fix: enhance execution phase handling in ProjectStore

- Added logic to correct stale execution phases when a task reaches a terminal status (human_review, done, pr_created) but the persisted executionPhase is outdated. - Improved the determination of execution progress to ensure accurate representation of task status, particularly for tasks that have transitioned from running to completed states. These changes improve the reliability of task status updates and enhance user experience by ensuring accurate visual feedback on task execution phases.
adjust for claude mem
2026-02-05 22:18:43 +01:00 · 2026-02-05 22:18:25 +01:00 · 2026-02-05 20:34:56 +01:00 · 2026-02-05 20:10:54 +01:00 · 2026-02-05 19:47:34 +01:00 · 2026-02-05 15:10:39 +01:00
345 changed files with 27344 additions and 9331 deletions
@@ -405,11 +405,11 @@ jobs:
      - name: Setup Node.js and install dependencies
        uses: ./.github/actions/setup-node-frontend

-      - name: Setup Flatpak
+      - name: Setup Flatpak and verification tools
        run: |
          set -e
          sudo apt-get update
-          sudo apt-get install -y flatpak flatpak-builder
+          sudo apt-get install -y flatpak flatpak-builder libarchive-tools
          flatpak remote-add --user --if-not-exists flathub https://flathub.org/repo/flathub.flatpakrepo
          flatpak install -y --user flathub org.freedesktop.Platform//25.08 org.freedesktop.Sdk//25.08
          flatpak install -y --user flathub org.electronjs.Electron2.BaseApp//25.08
@@ -447,6 +447,9 @@ jobs:
          SENTRY_TRACES_SAMPLE_RATE: ${{ secrets.SENTRY_TRACES_SAMPLE_RATE }}
          SENTRY_PROFILES_SAMPLE_RATE: ${{ secrets.SENTRY_PROFILES_SAMPLE_RATE }}

+      - name: Verify Linux packages
+        run: cd apps/frontend && npm run verify:linux
+
      - name: Upload artifacts
        uses: actions/upload-artifact@v4
        with:
@@ -344,10 +344,10 @@ jobs:
      - name: Setup Node.js and install dependencies
        uses: ./.github/actions/setup-node-frontend

-      - name: Setup Flatpak
+      - name: Setup Flatpak and verification tools
        run: |
          sudo apt-get update
-          sudo apt-get install -y flatpak flatpak-builder
+          sudo apt-get install -y flatpak flatpak-builder libarchive-tools
          flatpak remote-add --user --if-not-exists flathub https://flathub.org/repo/flathub.flatpakrepo
          flatpak install -y --user flathub org.freedesktop.Platform//25.08 org.freedesktop.Sdk//25.08
          flatpak install -y --user flathub org.electronjs.Electron2.BaseApp//25.08
@@ -383,6 +383,9 @@ jobs:
          SENTRY_TRACES_SAMPLE_RATE: ${{ secrets.SENTRY_TRACES_SAMPLE_RATE }}
          SENTRY_PROFILES_SAMPLE_RATE: ${{ secrets.SENTRY_PROFILES_SAMPLE_RATE }}

+      - name: Verify Linux packages
+        run: cd apps/frontend && npm run verify:linux
+
      - name: Upload artifacts
        uses: actions/upload-artifact@v4
        with:
@@ -40,6 +40,17 @@ Auto Claude is a desktop application (+ CLI) where users describe a goal and AI

 **PR target** — Always target the `develop` branch for PRs to AndyMik90/Auto-Claude, NOT `main`.

+## Memory (claude-mem)
+
+This project uses claude-mem for persistent memory across sessions. MCP search tools are available — use them proactively:
+
+- **Before modifying code** — Search for past work on the same files/features: `search(query="<file or feature>")`. Check if there are known bugs, decisions, or patterns to follow.
+- **When debugging** — Search for prior encounters with the same error or symptom: `search(query="<error message>", type="bugfix")`.
+- **When making architectural decisions** — Check for past decisions: `search(query="<topic>", type="decision")`.
+- **When resuming work** — Use `timeline(anchor=<recent_id>)` to understand where things left off.
+
+Follow the 3-layer workflow: `search` (cheap index) → `timeline` (context) → `get_observations` (full details only for relevant IDs). Never fetch full details without filtering first.
+
 ## Project Structure

 ```
@@ -35,18 +35,18 @@
 > ⚠️ Beta releases may contain bugs and breaking changes. [View all releases](https://github.com/AndyMik90/Auto-Claude/releases)

 <!-- BETA_VERSION_BADGE -->
-[![Beta](https://img.shields.io/badge/beta-2.7.2--beta.10-orange?style=flat-square)](https://github.com/AndyMik90/Auto-Claude/releases/tag/v2.7.2-beta.10)
+[![Beta](https://img.shields.io/badge/beta-2.7.6--beta.2-orange?style=flat-square)](https://github.com/AndyMik90/Auto-Claude/releases/tag/v2.7.6-beta.2)
 <!-- BETA_VERSION_BADGE_END -->

 <!-- BETA_DOWNLOADS -->
 | Platform | Download |
 |----------|----------|
-| **Windows** | [Auto-Claude-2.7.2-beta.10-win32-x64.exe](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.2-beta.10/Auto-Claude-2.7.2-beta.10-win32-x64.exe) |
-| **macOS (Apple Silicon)** | [Auto-Claude-2.7.2-beta.10-darwin-arm64.dmg](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.2-beta.10/Auto-Claude-2.7.2-beta.10-darwin-arm64.dmg) |
-| **macOS (Intel)** | [Auto-Claude-2.7.2-beta.10-darwin-x64.dmg](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.2-beta.10/Auto-Claude-2.7.2-beta.10-darwin-x64.dmg) |
-| **Linux** | [Auto-Claude-2.7.2-beta.10-linux-x86_64.AppImage](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.2-beta.10/Auto-Claude-2.7.2-beta.10-linux-x86_64.AppImage) |
-| **Linux (Debian)** | [Auto-Claude-2.7.2-beta.10-linux-amd64.deb](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.2-beta.10/Auto-Claude-2.7.2-beta.10-linux-amd64.deb) |
-| **Linux (Flatpak)** | [Auto-Claude-2.7.2-beta.10-linux-x86_64.flatpak](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.2-beta.10/Auto-Claude-2.7.2-beta.10-linux-x86_64.flatpak) |
+| **Windows** | [Auto-Claude-2.7.6-beta.2-win32-x64.exe](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.6-beta.2/Auto-Claude-2.7.6-beta.2-win32-x64.exe) |
+| **macOS (Apple Silicon)** | [Auto-Claude-2.7.6-beta.2-darwin-arm64.dmg](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.6-beta.2/Auto-Claude-2.7.6-beta.2-darwin-arm64.dmg) |
+| **macOS (Intel)** | [Auto-Claude-2.7.6-beta.2-darwin-x64.dmg](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.6-beta.2/Auto-Claude-2.7.6-beta.2-darwin-x64.dmg) |
+| **Linux** | [Auto-Claude-2.7.6-beta.2-linux-x86_64.AppImage](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.6-beta.2/Auto-Claude-2.7.6-beta.2-linux-x86_64.AppImage) |
+| **Linux (Debian)** | [Auto-Claude-2.7.6-beta.2-linux-amd64.deb](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.6-beta.2/Auto-Claude-2.7.6-beta.2-linux-amd64.deb) |
+| **Linux (Flatpak)** | [Auto-Claude-2.7.6-beta.2-linux-x86_64.flatpak](https://github.com/AndyMik90/Auto-Claude/releases/download/v2.7.6-beta.2/Auto-Claude-2.7.6-beta.2-linux-x86_64.flatpak) |
 <!-- BETA_DOWNLOADS_END -->

 > All releases include SHA256 checksums and VirusTotal scan results for security verification.
@@ -1,421 +0,0 @@
-# Token Encryption Investigation
-
-## Issue Summary
-
-Auto-Claude users are experiencing API 401 errors ("Invalid bearer token") because the Python backend is passing encrypted tokens (with `enc:` prefix) directly to the Claude Agent SDK without decryption. Standalone Claude Code terminals work correctly because they decrypt these tokens before use.
-
-**Key insight from user thehaffk:** "python cant unencrypt claude token and it launches session with CLAUDE_CODE_OAUTH_TOKEN=enc:djEwtxMGISt3tQ..."
-
-## Token Storage Format
-
-### Encrypted Token Format
-
-Claude Code CLI stores OAuth tokens in an encrypted format with the prefix `enc:`:
-
-```text
-enc:djEwtxMGISt3tQ...
-```
-
-This format is used when tokens are stored in:
- **macOS**: Keychain (service: "Claude Code-credentials")
- **Linux**: Secret Service API (DBus, via secretstorage library)
- **Windows**: Credential Manager / .credentials.json files
-
-### Decrypted Token Format
-
-Valid Claude OAuth tokens have the format:
-```text
-sk-ant-oat01-XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
-```
-
-## Current Token Flow (BROKEN)
-
-1. **Token Storage**: Claude Code CLI stores encrypted token with `enc:` prefix in system keychain
-2. **Token Retrieval**: `apps/backend/core/auth.py::get_auth_token()` retrieves token from:
-   - Environment variable `CLAUDE_CODE_OAUTH_TOKEN`
-   - OR system keychain via `get_token_from_keychain()`
-3. **❌ NO DECRYPTION**: Token is returned as-is with `enc:` prefix intact
-4. **SDK Initialization**: Encrypted token passed to Claude Agent SDK
-5. **API Call Fails**: SDK sends encrypted token to API → 401 error
-
-### Proof of Broken Flow
-
-Test in `apps/backend`:
-```python
-import os
-os.environ['CLAUDE_CODE_OAUTH_TOKEN'] = 'enc:test123'
-
-from core.auth import get_auth_token
-token = get_auth_token()
-print(f"Token: {token}")  # Output: "enc:test123"
-print(f"Encrypted: {token.startswith('enc:')}")  # Output: True
-```
-
-## How Standalone Claude Code CLI Handles Tokens
-
-### Current Understanding
-
-1. **Token Detection**: CLI checks if token starts with `enc:` prefix
-2. **Decryption**: If encrypted, CLI decrypts using platform-specific keyring access
-3. **Authentication**: Decrypted `sk-ant-oat01-` token is used for API calls
-
-### Missing Documentation
-
-Web search for "Claude Code CLI encrypted token enc: prefix decryption" found:
- Token storage formats (JSON with accessToken, refreshToken, expiresAt)
- Security issues (tokens exposed in debug logs before v2.1.0)
- Keychain access patterns for macOS/Linux/Windows
-
-**❌ NOT FOUND**: Specific documentation on how Claude Code CLI decrypts `enc:` tokens
-
-Sources:
- [Claude Code CLI over SSH on macOS: Fixing Keychain Access](https://phoenixtrap.com/2025/10/26/claude-code-cli-over-ssh-on-macos-fixing-keychain-access/)
- [Identity and Access Management - Claude Code Docs](https://code.claude.com/docs/en/iam)
- [Claude Code sessions should be encrypted | yoav.blog](https://yoav.blog/2026/01/09/claude-code-sessions-should-be-encrypted/)
-
-## Decryption Approach Options
-
-### Option 1: Claude Agent SDK Built-in Decryption
-
-**Status**: NEEDS VERIFICATION
-
-The Claude Agent SDK (`claude-agent-sdk>=0.1.19`) may handle decryption internally if:
- Token is passed to SDK still encrypted
- SDK detects `enc:` prefix
- SDK has access to system keyring for decryption
-
-**Action Required**: Check if SDK has decryption capabilities by examining:
- SDK source code or documentation
- Whether SDK expects encrypted vs decrypted tokens
- If SDK requires specific environment variables for decryption
-
-### Option 2: Python Backend Decryption (Recommended)
-
-**Approach**: Implement decryption in `apps/backend/core/auth.py` before passing to SDK
-
-**Implementation Pattern**:
-```python
-def get_auth_token() -> str | None:
-    """Get authentication token (decrypted if necessary)."""
-    token = _retrieve_token_from_sources()  # From env or keychain
-
-    if token and token.startswith("enc:"):
-        # Decrypt the token
-        token = decrypt_token(token)
-
-    return token
-
-def decrypt_token(encrypted_token: str) -> str:
-    """
-    Decrypt Claude Code encrypted token.
-
-    Args:
-        encrypted_token: Token with 'enc:' prefix
-
-    Returns:
-        Decrypted token in format 'sk-ant-oat01-...'
-    """
-    # Remove 'enc:' prefix
-    encrypted_data = encrypted_token[4:]
-
-    # TODO: Implement decryption logic
-    # Questions to answer:
-    # 1. What encryption algorithm does Claude Code use?
-    # 2. Where is the decryption key stored?
-    # 3. Is the decryption key platform-specific (per-user)?
-    # 4. Can we reuse Claude Code's decryption mechanism?
-
-    raise NotImplementedError("Token decryption not yet implemented")
-```
-
-### Option 3: Call Claude Code CLI for Decryption
-
-**Approach**: Use the Claude Code CLI binary to decrypt tokens
-
-```python
-def decrypt_token(encrypted_token: str) -> str:
-    """Decrypt token by invoking Claude Code CLI."""
-    # Find claude binary
-    claude_path = shutil.which("claude") or "~/.local/bin/claude"
-
-    # Use CLI command to get decrypted token
-    # (if such a command exists - needs research)
-    result = subprocess.run(
-        [claude_path, "auth", "decrypt", encrypted_token],
-        capture_output=True,
-        text=True
-    )
-
-    return result.stdout.strip()
-```
-
-**Issues**:
- Requires Claude Code CLI to be installed
- No documented CLI command for token decryption
- Adds external dependency
-
-## Required Investigation Steps
-
-### 1. Verify SDK Decryption Capabilities
-
-**Task**: Check if `claude-agent-sdk` handles `enc:` tokens automatically
-
-**Method**:
-```bash
-# In environment with SDK installed
-python3 << 'EOF'
-import os
-os.environ['CLAUDE_CODE_OAUTH_TOKEN'] = 'enc:...'  # Real encrypted token
-
-from claude_agent_sdk import Client
-# Try creating client - does it decrypt internally?
-client = Client()
-# Check if authentication works
-EOF
-```
-
-### 2. Reverse Engineer Claude Code CLI Decryption
-
-**Task**: Understand how Claude CLI decrypts tokens
-
-**Method**:
- Examine Claude CLI binary (if possible)
- Trace system calls when CLI runs (strace on Linux, dtruss on macOS)
- Check if CLI accesses specific keychain entries for decryption keys
- Look for encryption/decryption libraries used by CLI
-
-### 3. Find Decryption Key Storage
-
-**Task**: Locate where decryption keys are stored
-
-**Hypothesis**: Decryption key stored in:
- macOS: Keychain (separate entry from encrypted token)
- Linux: Secret Service API
- Windows: Credential Manager
-
-**Verification**:
-```bash
-# macOS: List all keychain entries
-security find-generic-password -a "$(whoami)" | grep -i claude
-
-# Linux: Use secretstorage to list all items
-python3 -c "import secretstorage; ..."
-```
-
-## Recommended Decryption Approach for Python Backend
-
-Based on investigation so far, the recommended approach is:
-
-1. **Detect encrypted tokens**: Check for `enc:` prefix in `get_auth_token()`
-2. **Decrypt before use**: Implement `decrypt_token()` function
-3. **Platform-specific decryption**: Use appropriate keyring library:
-   - macOS: Use `subprocess` with `/usr/bin/security` to access decryption key
-   - Linux: Use `secretstorage` library to access Secret Service API
-   - Windows: Access Credential Manager or credentials.json
-4. **Backward compatibility**: Support both encrypted and plaintext tokens
-5. **Error handling**: Provide clear error messages if decryption fails
-
-## Complete Token Flow Trace (Frontend → Backend)
-
-### 1. Token Retrieval (Frontend)
-
-**File**: `apps/frontend/src/main/services/profile-service.ts`
-
-The frontend retrieves the OAuth token from the system keychain but **does not decrypt it**. When no API profile is active (OAuth mode), the frontend returns an empty environment object, which means it relies on:
- The token already being in the environment as `CLAUDE_CODE_OAUTH_TOKEN`
- OR the Python backend retrieving it from the keychain
-
-**Key Code**:
-```typescript
-// Line 223: Returns empty object in OAuth mode, allowing
-// CLAUDE_CODE_OAUTH_TOKEN to be used from system keychain
-```
-
-### 2. Environment Variable Passing (Frontend → PTY)
-
-**File**: `apps/frontend/src/main/terminal/pty-manager.ts`
-
-The PTY manager spawns the terminal shell with environment variables, including `CLAUDE_CODE_OAUTH_TOKEN`:
-
-**Key Code** (Lines 149-152):
-```typescript
-// Remove ANTHROPIC_API_KEY to ensure Claude Code uses OAuth tokens
-// (CLAUDE_CODE_OAUTH_TOKEN from profileEnv) instead of API keys
-const { DEBUG: _DEBUG, ANTHROPIC_API_KEY: _ANTHROPIC_API_KEY, ...cleanEnv } = process.env;
-```
-
-**Important**: The frontend passes through whatever token value exists in the environment - it does NOT check for `enc:` prefix or decrypt it.
-
-### 3. Token Retrieval (Backend)
-
-**File**: `apps/backend/core/auth.py`
-
-#### 3.1. get_auth_token()
-
-This function retrieves the token from multiple sources:
-
-```python
-def get_auth_token() -> str | None:
-    # First check environment variables
-    for var in AUTH_TOKEN_ENV_VARS:  # CLAUDE_CODE_OAUTH_TOKEN, ANTHROPIC_AUTH_TOKEN
-        token = os.environ.get(var)
-        if token:
-            return token  # ❌ Returns immediately without checking for enc: prefix
-
-    # Fallback to system credential store
-    return get_token_from_keychain()  # ❌ Also returns without decryption
-```
-
-**Issue**: Returns token as-is with `enc:` prefix intact.
-
-#### 3.2. require_auth_token()
-
-This function calls `get_auth_token()` and raises an error if no token is found:
-
-```python
-def require_auth_token() -> str:
-    token = get_auth_token()  # ❌ Gets encrypted token
-    if not token:
-        raise ValueError("No OAuth token found...")
-    return token  # ❌ Returns encrypted token
-```
-
-**Issue**: No decryption step between retrieval and return.
-
-#### 3.3. ensure_claude_code_oauth_token()
-
-This function ensures the environment variable is set:
-
-```python
-def ensure_claude_code_oauth_token() -> None:
-    if os.environ.get("CLAUDE_CODE_OAUTH_TOKEN"):
-        return
-
-    token = get_auth_token()  # ❌ Gets encrypted token
-    if token:
-        os.environ["CLAUDE_CODE_OAUTH_TOKEN"] = token  # ❌ Sets encrypted token
-```
-
-**Issue**: Propagates encrypted token to environment variable.
-
-### 4. Token Usage in SDK Client Creation (Backend)
-
-#### 4.1. Full Client Creation
-
-**File**: `apps/backend/core/client.py` (see `create_client()` function)
-
-```python
-def create_client(...):
-    oauth_token = require_auth_token()  # ❌ Gets encrypted token
-    # Ensure SDK can access it via its expected env var
-    os.environ["CLAUDE_CODE_OAUTH_TOKEN"] = oauth_token  # ❌ Sets encrypted token
-```
-
-**Issue**: Encrypted token is passed to the Claude Agent SDK, which expects a decrypted `sk-ant-oat01-` token.
-
-#### 4.2. Simple Client Creation
-
-**File**: `apps/backend/core/simple_client.py` (see `create_simple_client()` function)
-
-```python
-def create_simple_client(...):
-    # Get authentication
-    oauth_token = require_auth_token()  # ❌ Gets encrypted token
-    import os
-    os.environ["CLAUDE_CODE_OAUTH_TOKEN"] = oauth_token  # ❌ Sets encrypted token
-```
-
-**Issue**: Same problem - encrypted token passed to SDK.
-
-#### 4.3. Other Usages
-
-**Files**:
- `apps/backend/core/workspace.py` (Line 1966) - AI merge operations
- `apps/backend/runners/insights_runner.py` - Insights analysis
- `apps/backend/runners/github/batch_issues.py` - GitHub batch operations
- `apps/backend/integrations/linear/updater.py` - Linear integration
- `apps/backend/commit_message.py` - Commit message generation
- `apps/backend/analysis/insight_extractor.py` - Code insights
- `apps/backend/merge/ai_resolver/claude_client.py` - Merge resolution
-
-**All follow the same pattern**: Call `ensure_claude_code_oauth_token()` → encrypted token in environment → SDK receives encrypted token → API 401 error.
-
-### 5. Where Decryption Should Be Inserted
-
-Based on the flow analysis, decryption should be added at **the earliest point of token retrieval** to avoid duplicating decryption logic:
-
-**RECOMMENDED INSERTION POINT**: `apps/backend/core/auth.py::get_auth_token()`
-
-```python
-def get_auth_token() -> str | None:
-    # First check environment variables
-    for var in AUTH_TOKEN_ENV_VARS:
-        token = os.environ.get(var)
-        if token:
-            # ✅ INSERT DECRYPTION HERE
-            if token.startswith("enc:"):
-                token = decrypt_token(token)
-            return token
-
-    # Fallback to system credential store
-    token = get_token_from_keychain()
-    # ✅ ALSO DECRYPT KEYCHAIN TOKENS
-    if token and token.startswith("enc:"):
-        token = decrypt_token(token)
-    return token
-```
-
-**Benefits of this approach**:
-1. Single location for decryption logic
-2. All downstream functions automatically get decrypted tokens
-3. Backward compatible (plaintext tokens pass through unchanged)
-4. Consistent behavior across all token sources (env vars and keychain)
-
-**Alternative insertion points** (NOT recommended):
- `require_auth_token()` - Would need similar logic in `get_auth_token()` for non-required usage
- `create_client()` - Would need duplication in `create_simple_client()` and all other clients
- `ensure_claude_code_oauth_token()` - Would miss direct `get_auth_token()` calls
-
-## Next Steps
-
-1. ✅ Document current token flow and identify issue (THIS FILE - COMPLETED)
-2. ✅ Trace token flow from frontend to backend (THIS FILE - COMPLETED)
-3. ✅ Identify where decryption should be inserted (THIS FILE - COMPLETED)
-4. ⏳ Verify if Claude Agent SDK handles decryption internally
-5. ⏳ Reverse engineer or document Claude Code CLI decryption mechanism
-6. ⏳ Implement `decrypt_token()` function in `apps/backend/core/auth.py`
-7. ⏳ Add encryption detection and auto-decryption to `get_auth_token()`
-8. ⏳ Test with real encrypted tokens on macOS and Linux
-9. ⏳ Add comprehensive error handling for decryption failures
-
-## Open Questions
-
-1. **What encryption algorithm does Claude Code use for `enc:` tokens?**
-   - Possible: AES-256, ChaCha20, or similar
-   - Key derivation method?
-
-2. **Where is the decryption key stored?**
-   - Same keychain entry as encrypted token?
-   - Separate keychain entry?
-   - Derived from system/user credentials?
-
-3. **Does Claude Agent SDK expect encrypted or decrypted tokens?**
-   - If it expects decrypted: we must decrypt before passing
-   - If it handles encryption: we may be missing SDK configuration
-
-4. **Is there a Claude Code CLI command to decrypt tokens?**
-   - `claude auth decrypt <token>`?
-   - `claude auth get-token`?
-   - No documented command found in research
-
-5. **Can we reuse Claude Code's decryption mechanism?**
-   - Import decryption functions from CLI?
-   - Call CLI as subprocess?
-   - Implement decryption ourselves?
-
-## References
-
- Issue: [GitHub #1223: API Error 401](https://github.com/AndyMik90/Auto-Claude/issues/1223)
- Current auth implementation: `apps/backend/core/auth.py`
- SDK client initialization: `apps/backend/core/client.py`
- Requirements: `apps/backend/requirements.txt` (includes `secretstorage>=3.3.3` for Linux)
@@ -19,5 +19,5 @@ Quick Start:
 See README.md for full documentation.
 """

-__version__ = "2.7.5"
+__version__ = "2.7.6-beta.2"
 __author__ = "Auto Claude Team"
@@ -6,6 +6,7 @@ Shared imports, types, and constants used across agent modules.
 """

 import logging
+import re

 # Configure logging
 logger = logging.getLogger(__name__)
@@ -13,3 +14,83 @@ logger = logging.getLogger(__name__)
 # Configuration constants
 AUTO_CONTINUE_DELAY_SECONDS = 3
 HUMAN_INTERVENTION_FILE = "PAUSE"
+
+# Retry configuration for 400 tool concurrency errors
+MAX_CONCURRENCY_RETRIES = 5  # Maximum number of retries for tool concurrency errors
+INITIAL_RETRY_DELAY_SECONDS = (
+    2  # Initial retry delay (doubles each retry: 2s, 4s, 8s, 16s, 32s)
+)
+MAX_RETRY_DELAY_SECONDS = 32  # Cap retry delay at 32 seconds
+
+# Pause file constants for intelligent error recovery
+# These files signal pause/resume between frontend and backend
+RATE_LIMIT_PAUSE_FILE = "RATE_LIMIT_PAUSE"  # Created when rate limited
+AUTH_FAILURE_PAUSE_FILE = "AUTH_PAUSE"  # Created when auth fails
+RESUME_FILE = "RESUME"  # Created by frontend to signal resume
+
+# Maximum time to wait for rate limit reset (2 hours)
+# If reset time is beyond this, task should fail rather than wait indefinitely
+MAX_RATE_LIMIT_WAIT_SECONDS = 7200
+
+# Wait intervals for pause/resume checking
+RATE_LIMIT_CHECK_INTERVAL_SECONDS = (
+    30  # Check for RESUME file every 30 seconds during rate limit wait
+)
+AUTH_RESUME_CHECK_INTERVAL_SECONDS = 10  # Check for re-authentication every 10 seconds
+AUTH_RESUME_MAX_WAIT_SECONDS = 86400  # Maximum wait for re-authentication (24 hours)
+
+
+def sanitize_error_message(error_message: str, max_length: int = 500) -> str:
+    """
+    Sanitize error messages to remove potentially sensitive information.
+
+    Redacts:
+    - API keys (sk-..., key-...)
+    - Bearer tokens
+    - Token/secret values
+
+    Args:
+        error_message: The raw error message to sanitize
+        max_length: Maximum length to truncate to (default 500)
+
+    Returns:
+        Sanitized and truncated error message
+    """
+    if not error_message:
+        return ""
+
+    # Redact patterns that look like API keys or tokens
+    # Pattern: sk-... (OpenAI/Anthropic keys like sk-ant-api03-...)
+    sanitized = re.sub(
+        r"\bsk-[a-zA-Z0-9._\-]{20,}\b", "[REDACTED_API_KEY]", error_message
+    )
+
+    # Pattern: key-... (generic API keys)
+    sanitized = re.sub(r"\bkey-[a-zA-Z0-9._\-]{20,}\b", "[REDACTED_API_KEY]", sanitized)
+
+    # Pattern: Bearer ... (bearer tokens)
+    sanitized = re.sub(
+        r"\bBearer\s+[a-zA-Z0-9._\-]{20,}\b", "Bearer [REDACTED_TOKEN]", sanitized
+    )
+
+    # Pattern: token= or token: followed by long strings
+    sanitized = re.sub(
+        r"(token[=:]\s*)[a-zA-Z0-9._\-]{20,}\b",
+        r"\1[REDACTED_TOKEN]",
+        sanitized,
+        flags=re.IGNORECASE,
+    )
+
+    # Pattern: secret= or secret: followed by strings
+    sanitized = re.sub(
+        r"(secret[=:]\s*)[a-zA-Z0-9._\-]{20,}\b",
+        r"\1[REDACTED_SECRET]",
+        sanitized,
+        flags=re.IGNORECASE,
+    )
+
+    # Truncate to max length
+    if len(sanitized) > max_length:
+        sanitized = sanitized[:max_length] + "..."
+
+    return sanitized
@@ -1,183 +1,183 @@
-"""
-Planner Agent Module
-====================
-
-Handles follow-up planner sessions for adding new subtasks to completed specs.
-"""
-
-import logging
-from pathlib import Path
-
-from core.client import create_client
-from phase_config import get_phase_model, get_phase_thinking_budget
-from phase_event import ExecutionPhase, emit_phase
-from task_logger import (
-    LogPhase,
-    get_task_logger,
-)
-from ui import (
-    BuildState,
-    Icons,
-    StatusManager,
-    bold,
-    box,
-    highlight,
-    icon,
-    muted,
-    print_status,
-)
-
-from .session import run_agent_session
-
-logger = logging.getLogger(__name__)
-
-
-async def run_followup_planner(
-    project_dir: Path,
-    spec_dir: Path,
-    model: str,
-    verbose: bool = False,
-) -> bool:
-    """
-    Run the follow-up planner to add new subtasks to a completed spec.
-
-    This is a simplified version of run_autonomous_agent that:
-    1. Creates a client
-    2. Loads the followup planner prompt
-    3. Runs a single planning session
-    4. Returns after the plan is updated (doesn't enter coding loop)
-
-    The planner agent will:
-    - Read FOLLOWUP_REQUEST.md for the new task
-    - Read the existing implementation_plan.json
-    - Add new phase(s) with pending subtasks
-    - Update the plan status back to in_progress
-
-    Args:
-        project_dir: Root directory for the project
-        spec_dir: Directory containing the completed spec
-        model: Claude model to use
-        verbose: Whether to show detailed output
-
-    Returns:
-        bool: True if planning completed successfully
-    """
-    from implementation_plan import ImplementationPlan
-    from prompts import get_followup_planner_prompt
-
-    # Initialize status manager for ccstatusline
-    status_manager = StatusManager(project_dir)
-    status_manager.set_active(spec_dir.name, BuildState.PLANNING)
-    emit_phase(ExecutionPhase.PLANNING, "Follow-up planning")
-
-    # Initialize task logger for persistent logging
-    task_logger = get_task_logger(spec_dir)
-
-    # Show header
-    content = [
-        bold(f"{icon(Icons.GEAR)} FOLLOW-UP PLANNER SESSION"),
-        "",
-        f"Spec: {highlight(spec_dir.name)}",
-        muted("Adding follow-up work to completed spec."),
-        "",
-        muted("The agent will read your FOLLOWUP_REQUEST.md and add new subtasks."),
-    ]
-    print()
-    print(box(content, width=70, style="heavy"))
-    print()
-
-    # Start planning phase in task logger
-    if task_logger:
-        task_logger.start_phase(LogPhase.PLANNING, "Starting follow-up planning...")
-        task_logger.set_session(1)
-
-    # Create client with phase-specific model and thinking budget
-    # Respects task_metadata.json configuration when no CLI override
-    planning_model = get_phase_model(spec_dir, "planning", model)
-    planning_thinking_budget = get_phase_thinking_budget(spec_dir, "planning")
-    client = create_client(
-        project_dir,
-        spec_dir,
-        planning_model,
-        max_thinking_tokens=planning_thinking_budget,
-    )
-
-    # Generate follow-up planner prompt
-    prompt = get_followup_planner_prompt(spec_dir)
-
-    print_status("Running follow-up planner...", "progress")
-    print()
-
-    try:
-        # Run single planning session
-        async with client:
-            status, response = await run_agent_session(
-                client, prompt, spec_dir, verbose, phase=LogPhase.PLANNING
-            )
-
-        # End planning phase in task logger
-        if task_logger:
-            task_logger.end_phase(
-                LogPhase.PLANNING,
-                success=(status != "error"),
-                message="Follow-up planning session completed",
-            )
-
-        if status == "error":
-            print()
-            print_status("Follow-up planning failed", "error")
-            status_manager.update(state=BuildState.ERROR)
-            return False
-
-        # Verify the plan was updated (should have pending subtasks now)
-        plan_file = spec_dir / "implementation_plan.json"
-        if plan_file.exists():
-            plan = ImplementationPlan.load(plan_file)
-
-            # Check if there are any pending subtasks
-            all_subtasks = [c for p in plan.phases for c in p.subtasks]
-            pending_subtasks = [c for c in all_subtasks if c.status.value == "pending"]
-
-            if pending_subtasks:
-                # Reset the plan status to in_progress (in case planner didn't)
-                plan.reset_for_followup()
-                await plan.async_save(plan_file)
-
-                print()
-                content = [
-                    bold(f"{icon(Icons.SUCCESS)} FOLLOW-UP PLANNING COMPLETE"),
-                    "",
-                    f"New pending subtasks: {highlight(str(len(pending_subtasks)))}",
-                    f"Total subtasks: {len(all_subtasks)}",
-                    "",
-                    muted("Next steps:"),
-                    f"  Run: {highlight(f'python auto-claude/run.py --spec {spec_dir.name}')}",
-                ]
-                print(box(content, width=70, style="heavy"))
-                print()
-                status_manager.update(state=BuildState.PAUSED)
-                return True
-            else:
-                print()
-                print_status(
-                    "Warning: No pending subtasks found after planning", "warning"
-                )
-                print(muted("The planner may not have added new subtasks."))
-                print(muted("Check implementation_plan.json manually."))
-                status_manager.update(state=BuildState.PAUSED)
-                return False
-        else:
-            print()
-            print_status(
-                "Error: implementation_plan.json not found after planning", "error"
-            )
-            status_manager.update(state=BuildState.ERROR)
-            return False
-
-    except Exception as e:
-        print()
-        print_status(f"Follow-up planning error: {e}", "error")
-        if task_logger:
-            task_logger.log_error(f"Follow-up planning error: {e}", LogPhase.PLANNING)
-        status_manager.update(state=BuildState.ERROR)
-        return False
+"""
+Planner Agent Module
+====================
+
+Handles follow-up planner sessions for adding new subtasks to completed specs.
+"""
+
+import logging
+from pathlib import Path
+
+from core.client import create_client
+from phase_config import get_phase_model, get_phase_thinking_budget
+from phase_event import ExecutionPhase, emit_phase
+from task_logger import (
+    LogPhase,
+    get_task_logger,
+)
+from ui import (
+    BuildState,
+    Icons,
+    StatusManager,
+    bold,
+    box,
+    highlight,
+    icon,
+    muted,
+    print_status,
+)
+
+from .session import run_agent_session
+
+logger = logging.getLogger(__name__)
+
+
+async def run_followup_planner(
+    project_dir: Path,
+    spec_dir: Path,
+    model: str,
+    verbose: bool = False,
+) -> bool:
+    """
+    Run the follow-up planner to add new subtasks to a completed spec.
+
+    This is a simplified version of run_autonomous_agent that:
+    1. Creates a client
+    2. Loads the followup planner prompt
+    3. Runs a single planning session
+    4. Returns after the plan is updated (doesn't enter coding loop)
+
+    The planner agent will:
+    - Read FOLLOWUP_REQUEST.md for the new task
+    - Read the existing implementation_plan.json
+    - Add new phase(s) with pending subtasks
+    - Update the plan status back to in_progress
+
+    Args:
+        project_dir: Root directory for the project
+        spec_dir: Directory containing the completed spec
+        model: Claude model to use
+        verbose: Whether to show detailed output
+
+    Returns:
+        bool: True if planning completed successfully
+    """
+    from implementation_plan import ImplementationPlan
+    from prompts import get_followup_planner_prompt
+
+    # Initialize status manager for ccstatusline
+    status_manager = StatusManager(project_dir)
+    status_manager.set_active(spec_dir.name, BuildState.PLANNING)
+    emit_phase(ExecutionPhase.PLANNING, "Follow-up planning")
+
+    # Initialize task logger for persistent logging
+    task_logger = get_task_logger(spec_dir)
+
+    # Show header
+    content = [
+        bold(f"{icon(Icons.GEAR)} FOLLOW-UP PLANNER SESSION"),
+        "",
+        f"Spec: {highlight(spec_dir.name)}",
+        muted("Adding follow-up work to completed spec."),
+        "",
+        muted("The agent will read your FOLLOWUP_REQUEST.md and add new subtasks."),
+    ]
+    print()
+    print(box(content, width=70, style="heavy"))
+    print()
+
+    # Start planning phase in task logger
+    if task_logger:
+        task_logger.start_phase(LogPhase.PLANNING, "Starting follow-up planning...")
+        task_logger.set_session(1)
+
+    # Create client with phase-specific model and thinking budget
+    # Respects task_metadata.json configuration when no CLI override
+    planning_model = get_phase_model(spec_dir, "planning", model)
+    planning_thinking_budget = get_phase_thinking_budget(spec_dir, "planning")
+    client = create_client(
+        project_dir,
+        spec_dir,
+        planning_model,
+        max_thinking_tokens=planning_thinking_budget,
+    )
+
+    # Generate follow-up planner prompt
+    prompt = get_followup_planner_prompt(spec_dir)
+
+    print_status("Running follow-up planner...", "progress")
+    print()
+
+    try:
+        # Run single planning session
+        async with client:
+            status, response, error_info = await run_agent_session(
+                client, prompt, spec_dir, verbose, phase=LogPhase.PLANNING
+            )
+
+        # End planning phase in task logger
+        if task_logger:
+            task_logger.end_phase(
+                LogPhase.PLANNING,
+                success=(status != "error"),
+                message="Follow-up planning session completed",
+            )
+
+        if status == "error":
+            print()
+            print_status("Follow-up planning failed", "error")
+            status_manager.update(state=BuildState.ERROR)
+            return False
+
+        # Verify the plan was updated (should have pending subtasks now)
+        plan_file = spec_dir / "implementation_plan.json"
+        if plan_file.exists():
+            plan = ImplementationPlan.load(plan_file)
+
+            # Check if there are any pending subtasks
+            all_subtasks = [c for p in plan.phases for c in p.subtasks]
+            pending_subtasks = [c for c in all_subtasks if c.status.value == "pending"]
+
+            if pending_subtasks:
+                # Reset the plan status to in_progress (in case planner didn't)
+                plan.reset_for_followup()
+                await plan.async_save(plan_file)
+
+                print()
+                content = [
+                    bold(f"{icon(Icons.SUCCESS)} FOLLOW-UP PLANNING COMPLETE"),
+                    "",
+                    f"New pending subtasks: {highlight(str(len(pending_subtasks)))}",
+                    f"Total subtasks: {len(all_subtasks)}",
+                    "",
+                    muted("Next steps:"),
+                    f"  Run: {highlight(f'python auto-claude/run.py --spec {spec_dir.name}')}",
+                ]
+                print(box(content, width=70, style="heavy"))
+                print()
+                status_manager.update(state=BuildState.PAUSED)
+                return True
+            else:
+                print()
+                print_status(
+                    "Warning: No pending subtasks found after planning", "warning"
+                )
+                print(muted("The planner may not have added new subtasks."))
+                print(muted("Check implementation_plan.json manually."))
+                status_manager.update(state=BuildState.PAUSED)
+                return False
+        else:
+            print()
+            print_status(
+                "Error: implementation_plan.json not found after planning", "error"
+            )
+            status_manager.update(state=BuildState.ERROR)
+            return False
+
+    except Exception as e:
+        print()
+        print_status(f"Follow-up planning error: {e}", "error")
+        if task_logger:
+            task_logger.log_error(f"Follow-up planning error: {e}", LogPhase.PLANNING)
+        status_manager.update(state=BuildState.ERROR)
+        return False
@@ -0,0 +1,347 @@
+"""
+PR Template Filler Agent Module
+================================
+
+Detects GitHub PR templates in a project and uses Claude to intelligently
+fill them based on code changes, spec context, commit history, and branch info.
+"""
+
+import logging
+from pathlib import Path
+
+from core.client import create_client
+from task_logger import LogPhase, get_task_logger
+
+from .session import run_agent_session
+
+logger = logging.getLogger(__name__)
+
+# Maximum diff size (in characters) before truncating to file-level summaries
+MAX_DIFF_CHARS = 30_000
+
+
+def detect_pr_template(project_dir: Path | str) -> str | None:
+    """
+    Detect a GitHub PR template in the project.
+
+    Searches for:
+    1. .github/PULL_REQUEST_TEMPLATE.md (single template)
+    2. .github/PULL_REQUEST_TEMPLATE/ directory (picks the first .md file)
+
+    Args:
+        project_dir: Root directory of the project
+
+    Returns:
+        The template content as a string, or None if no template is found.
+    """
+    project_dir = Path(project_dir)
+    # Check for single template file
+    single_template = project_dir / ".github" / "PULL_REQUEST_TEMPLATE.md"
+    if single_template.is_file():
+        try:
+            content = single_template.read_text(encoding="utf-8")
+            if content.strip():
+                logger.info(f"Found PR template: {single_template}")
+                return content
+        except Exception as e:
+            logger.warning(f"Failed to read PR template {single_template}: {e}")
+
+    # Check for template directory (pick first .md file alphabetically)
+    template_dir = project_dir / ".github" / "PULL_REQUEST_TEMPLATE"
+    if template_dir.is_dir():
+        try:
+            md_files = sorted(template_dir.glob("*.md"))
+            if md_files:
+                content = md_files[0].read_text(encoding="utf-8")
+                if content.strip():
+                    logger.info(f"Found PR template: {md_files[0]}")
+                    return content
+        except Exception as e:
+            logger.warning(f"Failed to read PR template from {template_dir}: {e}")
+
+    logger.info("No GitHub PR template found in project")
+    return None
+
+
+def _truncate_diff(diff_summary: str) -> str:
+    """
+    Truncate a large diff to file-level summaries to stay within token limits.
+
+    If the diff is within MAX_DIFF_CHARS, return it unchanged.
+    Otherwise, extract only file-level change summaries (e.g. file names
+    with insertions/deletions counts) and discard line-level detail.
+
+    Args:
+        diff_summary: The full diff summary text
+
+    Returns:
+        The original or truncated diff summary.
+    """
+    if len(diff_summary) <= MAX_DIFF_CHARS:
+        return diff_summary
+
+    lines = diff_summary.splitlines()
+    summary_lines: list[str] = []
+    summary_lines.append("(Diff truncated to file-level summaries due to size)")
+    summary_lines.append("")
+
+    for line in lines:
+        # Keep file-level summary lines (stat lines, file headers, etc.)
+        stripped = line.strip()
+        if (
+            stripped.startswith("diff --git")
+            or stripped.startswith("---")
+            or stripped.startswith("+++")
+            or "file changed" in stripped.lower()
+            or "files changed" in stripped.lower()
+            or "insertion" in stripped.lower()
+            or "deletion" in stripped.lower()
+            or stripped.startswith("rename")
+            or stripped.startswith("new file")
+            or stripped.startswith("deleted file")
+            or stripped.startswith("Binary files")
+        ):
+            summary_lines.append(line)
+
+    # If we couldn't extract meaningful summaries, take the first chunk
+    if len(summary_lines) <= 2:
+        truncated = diff_summary[:MAX_DIFF_CHARS]
+        return truncated + "\n\n(... diff truncated due to size)"
+
+    return "\n".join(summary_lines)
+
+
+def _strip_markdown_fences(content: str) -> str:
+    """
+    Strip markdown code fences from the response if present.
+
+    The AI sometimes wraps the output in ```markdown ... ``` even when instructed
+    not to. This ensures the PR body renders correctly on GitHub.
+
+    Args:
+        content: The response content to clean
+
+    Returns:
+        The content with markdown fences stripped.
+    """
+    result = content
+
+    # Strip opening fence (```markdown or just ```)
+    if result.startswith("```markdown"):
+        result = result[len("```markdown") :].lstrip("\n")
+    elif result.startswith("```md"):
+        result = result[len("```md") :].lstrip("\n")
+    elif result.startswith("```"):
+        result = result[3:].lstrip("\n")
+
+    # Strip closing fence
+    if result.endswith("```"):
+        result = result[:-3].rstrip("\n")
+
+    return result.strip()
+
+
+def _build_prompt(
+    template_content: str,
+    diff_summary: str,
+    spec_overview: str,
+    commit_log: str,
+    branch_name: str,
+    target_branch: str,
+) -> str:
+    """
+    Build the prompt for the PR template filler agent.
+
+    Combines the system prompt context variables into a single message
+    that includes the template and all change context.
+
+    Args:
+        template_content: The PR template markdown
+        diff_summary: Git diff summary (possibly truncated)
+        spec_overview: Spec.md content or summary
+        commit_log: Git log of commits in the PR
+        branch_name: Source branch name
+        target_branch: Target branch name
+
+    Returns:
+        The assembled prompt string.
+    """
+    return f"""Fill out the following GitHub PR template using the provided context.
+Return ONLY the filled template markdown — no preamble, no explanation, no code fences.
+
+## Checkbox Guidelines
+
+IMPORTANT: Be accurate and honest about what has and hasn't been verified.
+
+**Check these based on context (you can infer from the diff/spec):**
+- Base Branch targeting — check based on target_branch value
+- Type of Change (bug fix, feature, docs, refactor, test) — infer from diff and spec
+- Area (Frontend, Backend, Fullstack) — infer from changed file paths
+- Feature Toggle "N/A" — if the feature appears complete and not behind a flag
+- Breaking Changes "No" — if changes appear backward compatible
+
+**Leave UNCHECKED (these require human verification you cannot perform):**
+- "I've tested my changes locally" — you have not tested anything
+- "All CI checks pass" — CI has not run yet
+- "Windows/macOS/Linux tested" — requires manual testing on each platform
+- "All existing tests pass" — CI has not run yet
+- "New features include test coverage" — unless test files are clearly visible in the diff
+- "Bug fixes include regression tests" — unless test files are clearly visible in the diff
+
+**For platform/code quality checkboxes:**
+- "Used centralized platform/ module" — leave unchecked unless you can verify from the diff
+- "No hardcoded paths" — leave unchecked unless you can verify from the diff
+- "PR is small and focused (< 400 lines)" — check only if diff stats show < 400 lines changed
+
+**For the "I've synced with develop branch" checkbox:**
+- Leave unchecked — you cannot verify the sync status
+
+## PR Template
+
+{template_content}
+
+## Change Context
+
+### Branch Information
+- **Source branch:** {branch_name}
+- **Target branch:** {target_branch}
+
+### Git Diff Summary
+```
+{diff_summary}
+```
+
+### Spec Overview
+{spec_overview}
+
+### Commit History
+```
+{commit_log}
+```
+
+Fill every section of the PR template. Follow the checkbox guidelines above carefully.
+Output ONLY the completed template — no code fences, no preamble."""
+
+
+def _load_spec_overview(spec_dir: Path) -> str:
+    """
+    Load the spec.md content for context. Falls back to a brief note if unavailable.
+
+    Args:
+        spec_dir: Directory containing the spec files
+
+    Returns:
+        The spec content or a fallback message.
+    """
+    spec_file = spec_dir / "spec.md"
+    if spec_file.is_file():
+        try:
+            content = spec_file.read_text(encoding="utf-8")
+            # Truncate very long specs to keep prompt manageable
+            if len(content) > 8000:
+                return content[:8000] + "\n\n(... spec truncated for brevity)"
+            return content
+        except Exception as e:
+            logger.warning(f"Failed to read spec.md: {e}")
+    return "(No spec overview available)"
+
+
+async def run_pr_template_filler(
+    project_dir: Path,
+    spec_dir: Path,
+    model: str,
+    thinking_budget: int | None = None,
+    branch_name: str = "",
+    target_branch: str = "develop",
+    diff_summary: str = "",
+    commit_log: str = "",
+    verbose: bool = False,
+) -> str | None:
+    """
+    Run the PR template filler agent to generate a filled PR body.
+
+    Detects the project's PR template, gathers change context, and invokes
+    Claude to intelligently fill out the template sections.
+
+    Args:
+        project_dir: Root directory of the project
+        spec_dir: Directory containing the spec files
+        model: Claude model to use
+        thinking_budget: Max thinking tokens (None to disable extended thinking)
+        branch_name: Source branch name for the PR
+        target_branch: Target branch name for the PR
+        diff_summary: Git diff summary of changes
+        commit_log: Git log of commits included in the PR
+        verbose: Whether to show detailed output
+
+    Returns:
+        The filled template markdown string, or None if template detection fails
+        or the agent encounters an error.
+    """
+    # Detect PR template
+    template_content = detect_pr_template(project_dir)
+    if template_content is None:
+        logger.info("No PR template detected — skipping template filler")
+        return None
+
+    # Load spec overview
+    spec_overview = _load_spec_overview(spec_dir)
+
+    # Truncate diff if too large
+    truncated_diff = _truncate_diff(diff_summary)
+
+    # Build the prompt
+    prompt = _build_prompt(
+        template_content=template_content,
+        diff_summary=truncated_diff,
+        spec_overview=spec_overview,
+        commit_log=commit_log,
+        branch_name=branch_name,
+        target_branch=target_branch,
+    )
+
+    # Initialize task logger
+    task_logger = get_task_logger(spec_dir)
+    if task_logger:
+        task_logger.start_phase(LogPhase.CODING, "PR template filling")
+
+    # Create client following the pattern from planner.py
+    client = create_client(
+        project_dir,
+        spec_dir,
+        model,
+        agent_type="pr_template_filler",
+        max_thinking_tokens=thinking_budget,
+    )
+
+    try:
+        async with client:
+            status, response, _ = await run_agent_session(
+                client, prompt, spec_dir, verbose, phase=LogPhase.CODING
+            )
+
+        if task_logger:
+            task_logger.end_phase(
+                LogPhase.CODING,
+                success=(status != "error"),
+                message="PR template filling completed",
+            )
+
+        if status == "error":
+            logger.error("PR template filler agent returned an error")
+            return None
+
+        # The agent should return only the filled template markdown
+        if response and response.strip():
+            result = _strip_markdown_fences(response.strip())
+            logger.info("PR template filled successfully")
+            return result
+
+        logger.warning("PR template filler returned empty response")
+        return None
+
+    except Exception as e:
+        logger.error(f"PR template filler error: {e}")
+        if task_logger:
+            task_logger.log_error(f"PR template filler error: {e}", LogPhase.CODING)
+        return None
@@ -263,6 +263,12 @@ AGENT_CONFIGS = {
        "auto_claude_tools": [],
        "thinking_default": "low",
    },
+    "pr_template_filler": {
+        "tools": BASE_READ_TOOLS,  # Read-only — reads diff, template, spec
+        "mcp_servers": [],  # No MCP needed, context passed via prompt
+        "auto_claude_tools": [],
+        "thinking_default": "low",  # Fast utility task for structured fill-in
+    },
    "pr_reviewer": {
        "tools": BASE_READ_TOOLS + WEB_TOOLS,  # Read-only
        "mcp_servers": ["context7"],
@@ -270,18 +276,30 @@ AGENT_CONFIGS = {
        "thinking_default": "high",
    },
    "pr_orchestrator_parallel": {
-        "tools": BASE_READ_TOOLS + WEB_TOOLS,  # Read-only for parallel PR orchestrator
+        # Read-only for parallel PR orchestrator
+        # NOTE: Do NOT add "Task" here - the SDK auto-allows Task when agents are defined
+        # via the --agents flag. Explicitly adding it interferes with agent registration.
+        "tools": BASE_READ_TOOLS + WEB_TOOLS,
        "mcp_servers": ["context7"],
        "auto_claude_tools": [],
        "thinking_default": "high",
    },
    "pr_followup_parallel": {
-        "tools": BASE_READ_TOOLS
-        + WEB_TOOLS,  # Read-only for parallel followup reviewer
+        # Read-only for parallel followup reviewer
+        # NOTE: Do NOT add "Task" here - same reason as pr_orchestrator_parallel
+        "tools": BASE_READ_TOOLS + WEB_TOOLS,
        "mcp_servers": ["context7"],
        "auto_claude_tools": [],
        "thinking_default": "high",
    },
+    "pr_finding_validator": {
+        # Standalone validator for re-checking findings against actual code
+        # Called separately from orchestrator to validate findings with fresh context
+        "tools": BASE_READ_TOOLS,
+        "mcp_servers": [],
+        "auto_claude_tools": [],
+        "thinking_default": "medium",
+    },
    # ═══════════════════════════════════════════════════════════════════════
    # ANALYSIS PHASES
    # ═══════════════════════════════════════════════════════════════════════
@@ -56,14 +56,10 @@ def _apply_qa_update(
        "ready_for_qa_revalidation": status == "fixes_applied",
    }

-    # Update plan status to match QA result
-    # This ensures the UI shows the correct column after QA
-    if status == "approved":
-        plan["status"] = "human_review"
-        plan["planStatus"] = "review"
-    elif status == "rejected":
-        plan["status"] = "human_review"
-        plan["planStatus"] = "review"
+    # NOTE: Do NOT write plan["status"] or plan["planStatus"] here.
+    # The frontend XState task state machine owns status transitions.
+    # Writing status here races with XState's persistPlanStatusAndReasonSync()
+    # and can clobber the reviewReason field, causing tasks to appear "incomplete".

    plan["last_updated"] = datetime.now(timezone.utc).isoformat()

@@ -87,7 +87,10 @@ def handle_build_command(
        debug_success,
    )
    from phase_config import get_phase_model
-    from prompts_pkg.prompts import get_base_branch_from_metadata
+    from prompts_pkg.prompts import (
+        get_base_branch_from_metadata,
+        get_use_local_branch_from_metadata,
+    )
    from qa_loop import run_qa_validation_loop, should_run_qa

    from .utils import print_banner, validate_environment
@@ -203,6 +206,9 @@ def handle_build_command(
            base_branch = metadata_branch
            debug("run.py", f"Using base branch from task metadata: {base_branch}")

+    # Check if user requested local branch (preserves gitignored files like .env)
+    use_local_branch = get_use_local_branch_from_metadata(spec_dir)
+
    if workspace_mode == WorkspaceMode.ISOLATED:
        # Keep reference to original spec directory for syncing progress back
        source_spec_dir = spec_dir
@@ -213,6 +219,7 @@ def handle_build_command(
            workspace_mode,
            source_spec_dir=spec_dir,
            base_branch=base_branch,
+            use_local_branch=use_local_branch,
        )
        # Use the localized spec directory (inside worktree) for AI access
        if localized_spec_dir:
@@ -536,6 +536,175 @@ def handle_cleanup_worktrees_command(project_dir: Path) -> None:
    cleanup_all_worktrees(project_dir, confirm=True)


+def _detect_conflict_scenario(
+    project_dir: Path,
+    conflicting_files: list[str],
+    spec_branch: str,
+    base_branch: str,
+) -> dict:
+    """
+    Analyze conflicting files to determine the conflict scenario.
+
+    This helps distinguish between:
+    - 'already_merged': Task changes already identical in target branch
+    - 'superseded': Target has newer version of same feature
+    - 'diverged': Standard diverged branches (AI can resolve)
+    - 'normal_conflict': Actual conflicting changes
+
+    Returns dict with:
+    - scenario: 'already_merged' | 'superseded' | 'diverged' | 'normal_conflict'
+    - already_merged_files: files identical in task and target
+    - details: additional context
+    """
+    if not conflicting_files:
+        return {
+            "scenario": "normal_conflict",
+            "already_merged_files": [],
+            "details": "No conflicting files to analyze",
+        }
+
+    already_merged_files = []
+    superseded_files = []
+    diverged_files = []
+
+    try:
+        # Get the merge-base commit
+        merge_base_result = subprocess.run(
+            ["git", "merge-base", base_branch, spec_branch],
+            cwd=project_dir,
+            capture_output=True,
+            text=True,
+        )
+        if merge_base_result.returncode != 0:
+            debug_warning(
+                MODULE, "Could not find merge base for conflict scenario detection"
+            )
+            return {
+                "scenario": "normal_conflict",
+                "already_merged_files": [],
+                "details": "Could not determine merge base",
+            }
+
+        merge_base = merge_base_result.stdout.strip()
+
+        for file_path in conflicting_files:
+            try:
+                # Get content from spec branch (task's changes)
+                spec_content_result = subprocess.run(
+                    ["git", "show", f"{spec_branch}:{file_path}"],
+                    cwd=project_dir,
+                    capture_output=True,
+                    text=True,
+                )
+                # Get content from base branch (target)
+                base_content_result = subprocess.run(
+                    ["git", "show", f"{base_branch}:{file_path}"],
+                    cwd=project_dir,
+                    capture_output=True,
+                    text=True,
+                )
+                # Get content from merge-base (original state)
+                merge_base_content_result = subprocess.run(
+                    ["git", "show", f"{merge_base}:{file_path}"],
+                    cwd=project_dir,
+                    capture_output=True,
+                    text=True,
+                )
+
+                # Check file existence in each ref
+                spec_exists = spec_content_result.returncode == 0
+                base_exists = base_content_result.returncode == 0
+                merge_base_exists = merge_base_content_result.returncode == 0
+
+                if spec_exists and base_exists:
+                    spec_content = spec_content_result.stdout
+                    base_content = base_content_result.stdout
+
+                    # If contents are identical, the changes are already merged
+                    if spec_content == base_content:
+                        already_merged_files.append(file_path)
+                        debug(
+                            MODULE,
+                            f"File {file_path}: already merged (identical content)",
+                        )
+                    elif merge_base_exists:
+                        merge_base_content = merge_base_content_result.stdout
+                        # If base has changed from merge_base but spec matches merge_base,
+                        # the task's changes are superseded by newer changes
+                        if spec_content == merge_base_content:
+                            superseded_files.append(file_path)
+                            debug(
+                                MODULE,
+                                f"File {file_path}: superseded (base has newer changes)",
+                            )
+                        else:
+                            diverged_files.append(file_path)
+                            debug(
+                                MODULE,
+                                f"File {file_path}: diverged (both branches modified)",
+                            )
+                    else:
+                        diverged_files.append(file_path)
+                else:
+                    diverged_files.append(file_path)
+
+            except Exception as e:
+                debug_warning(
+                    MODULE, f"Error analyzing file {file_path} for scenario: {e}"
+                )
+                diverged_files.append(file_path)
+
+        # Determine overall scenario based on dominant pattern
+        total_files = len(conflicting_files)
+
+        if len(already_merged_files) == total_files:
+            scenario = "already_merged"
+            details = "All conflicting files have identical content in both branches"
+        elif len(already_merged_files) > total_files / 2:
+            scenario = "already_merged"
+            details = f"{len(already_merged_files)} of {total_files} files already have the same content"
+        elif len(superseded_files) == total_files:
+            scenario = "superseded"
+            details = "All task changes have been superseded by newer changes in the target branch"
+        elif len(superseded_files) > total_files / 2:
+            scenario = "superseded"
+            details = (
+                f"{len(superseded_files)} of {total_files} files have been superseded"
+            )
+        elif diverged_files:
+            scenario = "diverged"
+            details = f"{len(diverged_files)} files have diverged and need AI merge"
+        else:
+            scenario = "normal_conflict"
+            details = "Standard merge conflicts detected"
+
+        debug(
+            MODULE,
+            f"Conflict scenario: {scenario}",
+            already_merged=len(already_merged_files),
+            superseded=len(superseded_files),
+            diverged=len(diverged_files),
+        )
+
+        return {
+            "scenario": scenario,
+            "already_merged_files": already_merged_files,
+            "superseded_files": superseded_files,
+            "diverged_files": diverged_files,
+            "details": details,
+        }
+
+    except Exception as e:
+        debug_error(MODULE, f"Error detecting conflict scenario: {e}")
+        return {
+            "scenario": "normal_conflict",
+            "already_merged_files": [],
+            "superseded_files": [],
+            "diverged_files": [],
+            "details": f"Error during analysis: {e}",
+        }
+
+
 def _check_git_merge_conflicts(
    project_dir: Path, spec_name: str, base_branch: str | None = None
 ) -> dict:
@@ -879,6 +1048,24 @@ def handle_merge_preview_command(
            f for f in git_conflicts.get("conflicting_files", []) if not is_lock_file(f)
        ]

+        # Detect conflict scenario (already_merged, superseded, diverged, normal_conflict)
+        # This helps the UI show appropriate messaging and actions
+        conflict_scenario = None
+        if non_lock_conflicting_files:
+            conflict_scenario = _detect_conflict_scenario(
+                project_dir,
+                non_lock_conflicting_files,
+                git_conflicts["spec_branch"],
+                git_conflicts["base_branch"],
+            )
+            debug(
+                MODULE,
+                f"Conflict scenario detected: {conflict_scenario.get('scenario')}",
+                already_merged_files=len(
+                    conflict_scenario.get("already_merged_files", [])
+                ),
+            )
+
        # Use git diff file count as the authoritative totalFiles count
        # The semantic tracker may not track all files (e.g., test files, config files)
        # but we want to show the user all files that will be merged
@@ -952,6 +1139,16 @@ def handle_merge_preview_command(
                # Path-mapped files that need AI merge due to renames
                "pathMappedAIMerges": path_mapped_ai_merges,
                "totalRenames": len(path_mappings),
+                # Conflict scenario detection for better UX messaging
+                "scenario": conflict_scenario.get("scenario")
+                if conflict_scenario
+                else None,
+                "alreadyMergedFiles": conflict_scenario.get("already_merged_files", [])
+                if conflict_scenario
+                else [],
+                "scenarioMessage": conflict_scenario.get("details")
+                if conflict_scenario
+                else None,
            },
            "summary": {
                # Use git diff count, not semantic tracker count
@@ -6,6 +6,7 @@ for multiple environment variables, and SDK environment variable passthrough
 for custom API endpoints.
 """

+import hashlib
 import json
 import logging
 import os
@@ -65,6 +66,48 @@ SDK_ENV_VARS = [
 ]


+def _calculate_config_dir_hash(config_dir: str) -> str:
+    """
+    Calculate hash of config directory path for Keychain service name.
+
+    This MUST match the frontend's calculateConfigDirHash() in credential-utils.ts.
+    The frontend uses SHA256 hash of the config dir path, taking first 8 hex chars.
+
+    Args:
+        config_dir: Path to the config directory (should be absolute/expanded)
+
+    Returns:
+        8-character hex hash string (e.g., "d74c9506")
+    """
+    return hashlib.sha256(config_dir.encode()).hexdigest()[:8]
+
+
+def _get_keychain_service_name(config_dir: str | None = None) -> str:
+    """
+    Get the Keychain service name for credential storage.
+
+    This MUST match the frontend's getKeychainServiceName() in credential-utils.ts.
+    All profiles use hash-based keychain entries for isolation:
+    - Profile with configDir: "Claude Code-credentials-{hash}"
+    - No configDir (legacy/default): "Claude Code-credentials"
+
+    Args:
+        config_dir: Optional CLAUDE_CONFIG_DIR path. If provided, uses hash-based name.
+
+    Returns:
+        Keychain service name (e.g., "Claude Code-credentials-d74c9506")
+    """
+    if not config_dir:
+        return "Claude Code-credentials"
+
+    # Expand ~ to home directory (matching frontend normalization)
+    expanded_dir = os.path.expanduser(config_dir)
+
+    # Calculate hash and return hash-based service name
+    hash_suffix = _calculate_config_dir_hash(expanded_dir)
+    return f"Claude Code-credentials-{hash_suffix}"
+
+
 def is_encrypted_token(token: str | None) -> bool:
    """
    Check if a token is encrypted (has "enc:" prefix).
@@ -346,36 +389,50 @@ def _try_decrypt_token(token: str | None) -> str | None:
    return token


-def get_token_from_keychain() -> str | None:
+def get_token_from_keychain(config_dir: str | None = None) -> str | None:
    """
    Get authentication token from system credential store.

    Reads Claude Code credentials from:
-    - macOS: Keychain
+    - macOS: Keychain (uses hash-based service name if config_dir provided)
    - Windows: Credential Manager
    - Linux: Secret Service API (via dbus/secretstorage)

+    Args:
+        config_dir: Optional CLAUDE_CONFIG_DIR path for profile-specific credentials.
+                   When provided, reads from hash-based keychain entry matching
+                   the frontend's storage location.
+
    Returns:
        Token string if found, None otherwise
    """
    if is_macos():
-        return _get_token_from_macos_keychain()
+        return _get_token_from_macos_keychain(config_dir)
    elif is_windows():
-        return _get_token_from_windows_credential_files()
+        return _get_token_from_windows_credential_files(config_dir)
    else:
        # Linux: use secret-service API via DBus
-        return _get_token_from_linux_secret_service()
+        return _get_token_from_linux_secret_service(config_dir)


-def _get_token_from_macos_keychain() -> str | None:
-    """Get token from macOS Keychain."""
+def _get_token_from_macos_keychain(config_dir: str | None = None) -> str | None:
+    """Get token from macOS Keychain.
+
+    Args:
+        config_dir: Optional CLAUDE_CONFIG_DIR path. When provided, uses hash-based
+                   service name (e.g., "Claude Code-credentials-d74c9506") matching
+                   the frontend's credential storage location.
+    """
+    # Get the correct service name (hash-based if config_dir provided)
+    service_name = _get_keychain_service_name(config_dir)
+
    try:
        result = subprocess.run(
            [
                "/usr/bin/security",
                "find-generic-password",
                "-s",
-                "Claude Code-credentials",
+                service_name,
                "-w",
            ],
            capture_output=True,
@@ -384,6 +441,14 @@ def _get_token_from_macos_keychain() -> str | None:
        )

        if result.returncode != 0:
+            # If hash-based lookup fails and we have a config_dir, DON'T fall back
+            # to default service name - that would return the wrong profile's token.
+            # The config_dir was provided explicitly, so we should only use that.
+            if config_dir:
+                logger.debug(
+                    f"No keychain entry found for service '{service_name}' "
+                    f"(config_dir: {config_dir})"
+                )
            return None

        credentials_json = result.stdout.strip()
@@ -397,22 +462,51 @@ def _get_token_from_macos_keychain() -> str | None:
            return None

        # Validate token format (Claude OAuth tokens start with sk-ant-oat01-)
-        if not token.startswith("sk-ant-oat01-"):
+        # Also accept encrypted tokens (enc:) which will be decrypted later
+        if not (token.startswith("sk-ant-oat01-") or token.startswith("enc:")):
            return None

+        logger.debug(f"Found token in keychain service '{service_name}'")
        return token

    except (subprocess.TimeoutExpired, json.JSONDecodeError, KeyError, Exception):
        return None


-def _get_token_from_windows_credential_files() -> str | None:
+def _get_token_from_windows_credential_files(
+    config_dir: str | None = None,
+) -> str | None:
    """Get token from Windows credential files.

    Claude Code on Windows stores credentials in ~/.claude/.credentials.json
+    For custom profiles, uses the config_dir's .credentials.json file.
+
+    Args:
+        config_dir: Optional CLAUDE_CONFIG_DIR path for profile-specific credentials.
    """
    try:
-        # Claude Code stores credentials in ~/.claude/.credentials.json
+        # If config_dir is provided, read from that directory first
+        if config_dir:
+            expanded_dir = os.path.expanduser(config_dir)
+            profile_cred_paths = [
+                os.path.join(expanded_dir, ".credentials.json"),
+                os.path.join(expanded_dir, "credentials.json"),
+            ]
+            for cred_path in profile_cred_paths:
+                if os.path.exists(cred_path):
+                    with open(cred_path, encoding="utf-8") as f:
+                        data = json.load(f)
+                        token = data.get("claudeAiOauth", {}).get("accessToken")
+                        if token and (
+                            token.startswith("sk-ant-oat01-")
+                            or token.startswith("enc:")
+                        ):
+                            logger.debug(f"Found token in {cred_path}")
+                            return token
+            # If config_dir provided but no token found, don't fall back to default
+            return None
+
+        # Default Claude Code credential paths (no profile specified)
        cred_paths = [
            os.path.expandvars(r"%USERPROFILE%\.claude\.credentials.json"),
            os.path.expandvars(r"%USERPROFILE%\.claude\credentials.json"),
@@ -425,7 +519,9 @@ def _get_token_from_windows_credential_files() -> str | None:
                with open(cred_path, encoding="utf-8") as f:
                    data = json.load(f)
                    token = data.get("claudeAiOauth", {}).get("accessToken")
-                    if token and token.startswith("sk-ant-oat01-"):
+                    if token and (
+                        token.startswith("sk-ant-oat01-") or token.startswith("enc:")
+                    ):
                        return token

        return None
@@ -434,7 +530,7 @@ def _get_token_from_windows_credential_files() -> str | None:
        return None


-def _get_token_from_linux_secret_service() -> str | None:
+def _get_token_from_linux_secret_service(config_dir: str | None = None) -> str | None:
    """Get token from Linux Secret Service API via DBus.

    Claude Code on Linux stores credentials in the Secret Service API
@@ -442,9 +538,12 @@ def _get_token_from_linux_secret_service() -> str | None:
    uses the secretstorage library which communicates via DBus.

    The credential is stored with:
-    - Label: "Claude Code-credentials"
+    - Label: "Claude Code-credentials" or "Claude Code-credentials-{hash}" for profiles
    - Attributes: {application: "claude-code"}

+    Args:
+        config_dir: Optional CLAUDE_CONFIG_DIR path for profile-specific credentials.
+
    Returns:
        Token string if found, None otherwise
    """
@@ -452,6 +551,9 @@ def _get_token_from_linux_secret_service() -> str | None:
        # secretstorage not installed, fall back to env var
        return None

+    # Get the correct service name (hash-based if config_dir provided)
+    target_label = _get_keychain_service_name(config_dir)
+
    try:
        # Get the default collection (typically "login" keyring)
        # secretstorage handles DBus communication internally
@@ -476,10 +578,10 @@ def _get_token_from_linux_secret_service() -> str | None:
        items = collection.search_items({"application": "claude-code"})

        for item in items:
-            # Check if this is the Claude Code credentials item
+            # Check if this is the correct Claude Code credentials item
            label = item.get_label()
-            # Use exact match for "Claude Code-credentials" to avoid false positives
-            if label == "Claude Code-credentials":
+            # Use exact match for target label (profile-specific or default)
+            if label == target_label:
                # Get the secret (stored as JSON string)
                secret = item.get_secret()
                if not secret:
@@ -492,11 +594,23 @@ def _get_token_from_linux_secret_service() -> str | None:
                    data = json.loads(secret)
                    token = data.get("claudeAiOauth", {}).get("accessToken")

-                    if token and token.startswith("sk-ant-oat01-"):
+                    if token and (
+                        token.startswith("sk-ant-oat01-") or token.startswith("enc:")
+                    ):
+                        logger.debug(
+                            f"Found token in secret service with label '{target_label}'"
+                        )
                        return token
                except json.JSONDecodeError:
                    continue

+        # If config_dir was provided but no token found, don't fall back
+        if config_dir:
+            logger.debug(
+                f"No secret service entry found with label '{target_label}' "
+                f"(config_dir: {config_dir})"
+            )
+
        return None

    except (
@@ -589,13 +703,37 @@ def get_auth_token(config_dir: str | None = None) -> str | None:
    env_config_dir = os.environ.get("CLAUDE_CONFIG_DIR")
    effective_config_dir = config_dir or env_config_dir

-    # If a custom config directory is specified, read from there
+    # Debug: Log which config_dir is being used for credential resolution
+    debug = os.environ.get("DEBUG", "").lower() in ("true", "1")
+    if debug and effective_config_dir:
+        service_name = _get_keychain_service_name(effective_config_dir)
+        logger.info(
+            f"[Auth] Resolving credentials for profile config_dir: {effective_config_dir} "
+            f"(Keychain service: {service_name})"
+        )
+
+    # If a custom config directory is specified, read from there first
    if effective_config_dir:
+        # Try reading from .credentials.json file in the config directory
        token = _get_token_from_config_dir(effective_config_dir)
        if token:
            return _try_decrypt_token(token)

-    # Fallback to system credential store (default locations)
+        # Also try the system credential store with hash-based service name
+        # This is needed because macOS stores credentials in Keychain, not files
+        token = get_token_from_keychain(effective_config_dir)
+        if token:
+            return _try_decrypt_token(token)
+
+        # If config_dir was explicitly provided, DON'T fall back to default keychain
+        # - that would return the wrong profile's token
+        logger.debug(
+            f"No credentials found for config_dir '{effective_config_dir}' "
+            "in file or keychain"
+        )
+        return None
+
+    # No config_dir specified - use default system credential store
    return _try_decrypt_token(get_token_from_keychain())


@@ -616,10 +754,20 @@ def get_auth_token_source(config_dir: str | None = None) -> str | None:
    # Check if token came from custom config directory (profile's configDir)
    env_config_dir = os.environ.get("CLAUDE_CONFIG_DIR")
    effective_config_dir = config_dir or env_config_dir
-    if effective_config_dir and _get_token_from_config_dir(effective_config_dir):
-        return "CLAUDE_CONFIG_DIR"
+    if effective_config_dir:
+        # Check file-based storage
+        if _get_token_from_config_dir(effective_config_dir):
+            return "CLAUDE_CONFIG_DIR"
+        # Check hash-based keychain entry for this profile
+        if get_token_from_keychain(effective_config_dir):
+            if is_macos():
+                return "macOS Keychain (profile)"
+            elif is_windows():
+                return "Windows Credential Files (profile)"
+            else:
+                return "Linux Secret Service (profile)"

-    # Check if token came from system credential store
+    # Check if token came from default system credential store
    if get_token_from_keychain():
        if is_macos():
            return "macOS Keychain"
@@ -800,6 +948,57 @@ def get_sdk_env_vars() -> dict[str, str]:
    return env


+def configure_sdk_authentication(config_dir: str | None = None) -> None:
+    """
+    Configure SDK authentication based on environment variables.
+
+    Supports two authentication modes:
+    - API Profile mode (ANTHROPIC_BASE_URL set): uses ANTHROPIC_AUTH_TOKEN
+    - OAuth mode (default): uses CLAUDE_CODE_OAUTH_TOKEN
+
+    In API profile mode, explicitly removes CLAUDE_CODE_OAUTH_TOKEN from the
+    environment because the SDK gives OAuth priority over API keys when both
+    are present.
+
+    Args:
+        config_dir: Optional profile config directory for per-profile Keychain
+                    lookup. When set, enables multi-profile token storage.
+
+    Raises:
+        ValueError: If required tokens are missing for the active mode.
+                   - API profile mode: requires ANTHROPIC_AUTH_TOKEN
+                   - OAuth mode: requires CLAUDE_CODE_OAUTH_TOKEN (from Keychain or env)
+    """
+    api_profile_mode = bool(os.environ.get("ANTHROPIC_BASE_URL", "").strip())
+
+    if api_profile_mode:
+        # API profile mode: ensure ANTHROPIC_AUTH_TOKEN is present
+        if not os.environ.get("ANTHROPIC_AUTH_TOKEN"):
+            raise ValueError(
+                "API profile mode active (ANTHROPIC_BASE_URL is set) "
+                "but ANTHROPIC_AUTH_TOKEN is not set"
+            )
+        # Explicitly remove CLAUDE_CODE_OAUTH_TOKEN so SDK uses ANTHROPIC_AUTH_TOKEN
+        # SDK gives OAuth priority over API keys when both are present
+        os.environ.pop("CLAUDE_CODE_OAUTH_TOKEN", None)
+        logger.info("Using API profile authentication")
+    else:
+        # OAuth mode: require and validate OAuth token
+        # Get OAuth token - uses profile-specific Keychain lookup when config_dir is set
+        # This correctly reads from "Claude Code-credentials-{hash}" for non-default profiles
+        oauth_token = require_auth_token(config_dir)
+
+        # Validate token is not encrypted before passing to SDK
+        # Encrypted tokens (enc:...) should have been decrypted by require_auth_token()
+        # If we still have an encrypted token here, it means decryption failed or was skipped
+        validate_token_not_encrypted(oauth_token)
+
+        # Ensure SDK can access it via its expected env var
+        # This is required because the SDK doesn't know about per-profile Keychain naming
+        os.environ["CLAUDE_CODE_OAUTH_TOKEN"] = oauth_token
+        logger.info("Using OAuth authentication")
+
+
 def ensure_claude_code_oauth_token() -> None:
    """
    Ensure CLAUDE_CODE_OAUTH_TOKEN is set (for SDK compatibility).
@@ -139,9 +139,8 @@ from agents.tools_pkg import (
 from claude_agent_sdk import ClaudeAgentOptions, ClaudeSDKClient
 from claude_agent_sdk.types import HookMatcher
 from core.auth import (
+    configure_sdk_authentication,
    get_sdk_env_vars,
-    require_auth_token,
-    validate_token_not_encrypted,
 )
 from linear_updater import is_linear_enabled
 from prompts_pkg.project_context import detect_project_capabilities, load_project_index
@@ -490,20 +489,19 @@ def create_client(
       (see security.py for ALLOWED_COMMANDS)
    4. Tool filtering - Each agent type only sees relevant tools (prevents misuse)
    """
-    # Get OAuth token - Claude CLI handles token lifecycle internally
-    oauth_token = require_auth_token()
-
-    # Validate token is not encrypted before passing to SDK
-    # Encrypted tokens (enc:...) should have been decrypted by require_auth_token()
-    # If we still have an encrypted token here, it means decryption failed or was skipped
-    validate_token_not_encrypted(oauth_token)
-
-    # Ensure SDK can access it via its expected env var
-    os.environ["CLAUDE_CODE_OAUTH_TOKEN"] = oauth_token
-
-    # Collect env vars to pass to SDK (ANTHROPIC_BASE_URL, etc.)
+    # Collect env vars to pass to SDK (ANTHROPIC_BASE_URL, CLAUDE_CONFIG_DIR, etc.)
    sdk_env = get_sdk_env_vars()

+    # Get the config dir for profile-specific credential lookup
+    # CLAUDE_CONFIG_DIR enables per-profile Keychain entries with SHA256-hashed service names
+    config_dir = sdk_env.get("CLAUDE_CONFIG_DIR")
+
+    # Configure SDK authentication (OAuth or API profile mode)
+    configure_sdk_authentication(config_dir)
+
+    if config_dir:
+        logger.info(f"Using CLAUDE_CONFIG_DIR for profile: {config_dir}")
+
    # Debug: Log git-bash path detection on Windows
    if "CLAUDE_CODE_GIT_BASH_PATH" in sdk_env:
        logger.info(f"Git Bash path found: {sdk_env['CLAUDE_CODE_GIT_BASH_PATH']}")
@@ -0,0 +1,115 @@
+#!/usr/bin/env python3
+"""
+Git Provider Detection
+======================
+
+Utility to detect git hosting provider (GitHub, GitLab, or unknown) from git remote URLs.
+Supports both SSH and HTTPS remote formats, and self-hosted GitLab instances.
+"""
+
+import re
+from pathlib import Path
+
+from .git_executable import run_git
+
+
+def detect_git_provider(project_dir: str | Path, remote_name: str | None = None) -> str:
+    """Detect the git hosting provider from the git remote URL.
+
+    Args:
+        project_dir: Path to the git repository
+        remote_name: Name of the remote to check (defaults to "origin")
+
+    Returns:
+        'github' if GitHub remote detected
+        'gitlab' if GitLab remote detected (cloud or self-hosted)
+        'unknown' if no remote or unsupported provider
+
+    Examples:
+        >>> detect_git_provider('/path/to/repo')
+        'github'  # for git@github.com:user/repo.git
+        'gitlab'  # for git@gitlab.com:user/repo.git
+        'gitlab'  # for https://gitlab.company.com/user/repo.git
+        'unknown' # for no remote or other providers
+    """
+    try:
+        # Get the remote URL (use specified remote or default to origin)
+        remote = remote_name if remote_name else "origin"
+        result = run_git(
+            ["remote", "get-url", remote],
+            cwd=project_dir,
+            timeout=5,
+        )
+
+        # If command failed or no output, return unknown
+        if result.returncode != 0 or not result.stdout.strip():
+            return "unknown"
+
+        remote_url = result.stdout.strip()
+
+        # Parse ssh:// URL format: ssh://[user@]host[:port]/path
+        ssh_url_match = re.match(r"^ssh://(?:[^@]+@)?([^:/]+)(?::\d+)?/", remote_url)
+        if ssh_url_match:
+            hostname = ssh_url_match.group(1)
+            return _classify_hostname(hostname)
+
+        # Parse HTTPS/HTTP format: https://host/path or http://host/path
+        # Must check before scp-like format to avoid matching "https" as hostname
+        https_match = re.match(r"^https?://([^/]+)/", remote_url)
+        if https_match:
+            hostname = https_match.group(1)
+            return _classify_hostname(hostname)
+
+        # Parse scp-like format: [user@]host:path (any username, not just 'git')
+        # This handles git@github.com:user/repo.git and similar formats
+        scp_match = re.match(r"^(?:[^@]+@)?([^:]+):", remote_url)
+        if scp_match:
+            hostname = scp_match.group(1)
+            # Exclude paths that look like Windows drives (e.g., C:)
+            if len(hostname) > 1:
+                return _classify_hostname(hostname)
+
+        # Unrecognized URL format
+        return "unknown"
+
+    except Exception:
+        # Any error (subprocess issues, etc.) -> unknown
+        return "unknown"
+
+
+def _classify_hostname(hostname: str) -> str:
+    """Classify a hostname as github, gitlab, or unknown.
+
+    Args:
+        hostname: The git remote hostname (e.g., 'github.com', 'gitlab.example.com')
+
+    Returns:
+        'github', 'gitlab', or 'unknown'
+    """
+    hostname_lower = hostname.lower()
+
+    # Check for GitHub (cloud and self-hosted/enterprise)
+    # Match github.com, *.github.com, or domains where a segment is or starts with 'github'
+    hostname_parts = hostname_lower.split(".")
+    if (
+        hostname_lower == "github.com"
+        or hostname_lower.endswith(".github.com")
+        or any(
+            part == "github" or part.startswith("github-") for part in hostname_parts
+        )
+    ):
+        return "github"
+
+    # Check for GitLab (cloud and self-hosted)
+    # Match gitlab.com, *.gitlab.com, or domains where a segment is or starts with 'gitlab'
+    if (
+        hostname_lower == "gitlab.com"
+        or hostname_lower.endswith(".gitlab.com")
+        or any(
+            part == "gitlab" or part.startswith("gitlab-") for part in hostname_parts
+        )
+    ):
+        return "gitlab"
+
+    # Unknown provider
+    return "unknown"
@@ -0,0 +1,192 @@
+#!/usr/bin/env python3
+"""
+GitLab CLI Executable Finder
+============================
+
+Utility to find the glab (GitLab CLI) executable, with platform-specific fallbacks.
+"""
+
+import os
+import shutil
+import subprocess
+
+_cached_glab_path: str | None = None
+
+
+def invalidate_glab_cache() -> None:
+    """Invalidate the cached glab executable path.
+
+    Useful when glab may have been uninstalled, updated, or when
+    GITLAB_CLI_PATH environment variable has changed.
+    """
+    global _cached_glab_path
+    _cached_glab_path = None
+
+
+def _verify_glab_executable(path: str) -> bool:
+    """Verify that a path is a valid glab executable by checking version.
+
+    Args:
+        path: Path to the potential glab executable
+
+    Returns:
+        True if the path points to a valid glab executable, False otherwise
+    """
+    try:
+        result = subprocess.run(
+            [path, "--version"],
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            timeout=5,
+        )
+        return result.returncode == 0
+    except (subprocess.TimeoutExpired, OSError):
+        return False
+
+
+def _run_where_command() -> str | None:
+    """Run Windows 'where glab' command to find glab executable.
+
+    Returns:
+        First path found, or None if command failed
+    """
+    try:
+        result = subprocess.run(
+            "where glab",
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            timeout=5,
+            shell=True,  # Required: 'where' command must be executed through shell on Windows
+        )
+        if result.returncode == 0 and result.stdout.strip():
+            found_path = result.stdout.strip().split("\n")[0].strip()
+            if (
+                found_path
+                and os.path.isfile(found_path)
+                and _verify_glab_executable(found_path)
+            ):
+                return found_path
+    except (subprocess.TimeoutExpired, OSError):
+        # 'where' command failed or timed out - fall through to return None
+        pass
+    return None
+
+
+def get_glab_executable() -> str | None:
+    """Find the glab executable, with platform-specific fallbacks.
+
+    Returns the path to glab executable, or None if not found.
+
+    Priority order:
+    1. GITLAB_CLI_PATH env var (user-configured path from frontend)
+    2. shutil.which (if glab is in PATH)
+    3. Homebrew paths on macOS
+    4. Windows Program Files paths
+    5. Windows 'where' command
+
+    Caches the result after first successful find. Use invalidate_glab_cache()
+    to force re-detection (e.g., after glab installation/uninstallation).
+    """
+    global _cached_glab_path
+
+    # Return cached result if available AND still exists
+    if _cached_glab_path is not None and os.path.isfile(_cached_glab_path):
+        return _cached_glab_path
+
+    _cached_glab_path = _find_glab_executable()
+    return _cached_glab_path
+
+
+def _find_glab_executable() -> str | None:
+    """Internal function to find glab executable."""
+    # 1. Check GITLAB_CLI_PATH env var (set by Electron frontend)
+    env_path = os.environ.get("GITLAB_CLI_PATH")
+    if env_path and os.path.isfile(env_path) and _verify_glab_executable(env_path):
+        return env_path
+
+    # 2. Try shutil.which (works if glab is in PATH)
+    glab_path = shutil.which("glab")
+    if glab_path and _verify_glab_executable(glab_path):
+        return glab_path
+
+    # 3. macOS-specific: check Homebrew paths
+    if os.name != "nt":  # Unix-like systems (macOS, Linux)
+        homebrew_paths = [
+            "/opt/homebrew/bin/glab",  # Apple Silicon
+            "/usr/local/bin/glab",  # Intel Mac
+            "/home/linuxbrew/.linuxbrew/bin/glab",  # Linux Homebrew
+        ]
+        for path in homebrew_paths:
+            if os.path.isfile(path) and _verify_glab_executable(path):
+                return path
+
+    # 4. Windows-specific: check Program Files paths
+    # glab uses Inno Setup with DefaultDirName={autopf}\glab
+    if os.name == "nt":
+        windows_paths = [
+            os.path.expandvars(r"%PROGRAMFILES%\glab\glab.exe"),
+            os.path.expandvars(r"%PROGRAMFILES(X86)%\glab\glab.exe"),
+            os.path.expandvars(r"%LOCALAPPDATA%\Programs\glab\glab.exe"),
+        ]
+        for path in windows_paths:
+            if os.path.isfile(path) and _verify_glab_executable(path):
+                return path
+
+        # 5. Try 'where' command with shell=True (more reliable on Windows)
+        return _run_where_command()
+
+    return None
+
+
+def run_glab(
+    args: list[str],
+    cwd: str | None = None,
+    timeout: int = 60,
+    input_data: str | None = None,
+) -> subprocess.CompletedProcess:
+    """Run a glab command with proper executable finding.
+
+    Args:
+        args: glab command arguments (without 'glab' prefix)
+        cwd: Working directory for the command
+        timeout: Command timeout in seconds (default: 60)
+        input_data: Optional string data to pass to stdin
+
+    Returns:
+        CompletedProcess with command results.
+    """
+    glab = get_glab_executable()
+    if not glab:
+        return subprocess.CompletedProcess(
+            args=["glab"] + args,
+            returncode=-1,
+            stdout="",
+            stderr="GitLab CLI (glab) not found. Install from https://gitlab.com/gitlab-org/cli",
+        )
+    try:
+        return subprocess.run(
+            [glab] + args,
+            cwd=cwd,
+            input=input_data,
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            errors="replace",
+            timeout=timeout,
+        )
+    except subprocess.TimeoutExpired:
+        return subprocess.CompletedProcess(
+            args=[glab] + args,
+            returncode=-1,
+            stdout="",
+            stderr=f"Command timed out after {timeout} seconds",
+        )
+    except FileNotFoundError:
+        return subprocess.CompletedProcess(
+            args=[glab] + args,
+            returncode=-1,
+            stdout="",
+            stderr="GitLab CLI (glab) executable not found. Install from https://gitlab.com/gitlab-org/cli",
+        )
@@ -23,6 +23,9 @@ class ExecutionPhase(str, Enum):
    QA_FIXING = "qa_fixing"
    COMPLETE = "complete"
    FAILED = "failed"
+    # Pause states for intelligent error recovery
+    RATE_LIMIT_PAUSED = "rate_limit_paused"
+    AUTH_FAILURE_PAUSED = "auth_failure_paused"


 def emit_phase(
@@ -31,8 +34,19 @@ def emit_phase(
    *,
    progress: int | None = None,
    subtask: str | None = None,
+    reset_timestamp: int | None = None,
+    profile_id: str | None = None,
 ) -> None:
-    """Emit structured phase event to stdout for frontend parsing."""
+    """Emit structured phase event to stdout for frontend parsing.
+
+    Args:
+        phase: The execution phase (e.g., PLANNING, CODING, RATE_LIMIT_PAUSED)
+        message: Optional message describing the phase state
+        progress: Optional progress percentage (0-100)
+        subtask: Optional subtask identifier
+        reset_timestamp: Optional Unix timestamp for rate limit reset time
+        profile_id: Optional profile ID that triggered the pause
+    """
    phase_value = phase.value if isinstance(phase, ExecutionPhase) else phase

    payload: dict[str, Any] = {
@@ -48,6 +62,12 @@ def emit_phase(
    if subtask is not None:
        payload["subtask"] = subtask

+    if reset_timestamp is not None:
+        payload["reset_timestamp"] = reset_timestamp
+
+    if profile_id is not None:
+        payload["profile_id"] = profile_id
+
    try:
        print(f"{PHASE_MARKER_PREFIX}{json.dumps(payload, default=str)}", flush=True)
    except (OSError, UnicodeEncodeError) as e:
@@ -22,14 +22,14 @@ Example usage:
 """

 import logging
+import os
 from pathlib import Path

 from agents.tools_pkg import get_agent_config, get_default_thinking_level
 from claude_agent_sdk import ClaudeAgentOptions, ClaudeSDKClient
 from core.auth import (
+    configure_sdk_authentication,
    get_sdk_env_vars,
-    require_auth_token,
-    validate_token_not_encrypted,
 )
 from core.platform import validate_cli_path
 from phase_config import get_thinking_budget
@@ -72,21 +72,16 @@ def create_simple_client(
    Raises:
        ValueError: If agent_type is not found in AGENT_CONFIGS
    """
-    # Get authentication
-    oauth_token = require_auth_token()
-
-    # Validate token is not encrypted before passing to SDK
-    # Encrypted tokens (enc:...) should have been decrypted by require_auth_token()
-    # If we still have an encrypted token here, it means decryption failed or was skipped
-    validate_token_not_encrypted(oauth_token)
-
-    import os
-
-    os.environ["CLAUDE_CODE_OAUTH_TOKEN"] = oauth_token
-
-    # Get environment variables for SDK
+    # Get environment variables for SDK (including CLAUDE_CONFIG_DIR if set)
    sdk_env = get_sdk_env_vars()

+    # Get the config dir for profile-specific credential lookup
+    # CLAUDE_CONFIG_DIR enables per-profile Keychain entries with SHA256-hashed service names
+    config_dir = sdk_env.get("CLAUDE_CONFIG_DIR")
+
+    # Configure SDK authentication (OAuth or API profile mode)
+    configure_sdk_authentication(config_dir)
+
    # Get agent configuration (raises ValueError if unknown type)
    config = get_agent_config(agent_type)

@@ -0,0 +1,101 @@
+"""
+Task event protocol for frontend XState synchronization.
+
+Protocol: __TASK_EVENT__:{...}
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import sys
+from dataclasses import dataclass
+from datetime import datetime, timezone
+from pathlib import Path
+from uuid import uuid4
+
+TASK_EVENT_PREFIX = "__TASK_EVENT__:"
+_DEBUG = os.environ.get("DEBUG", "").lower() in ("1", "true", "yes")
+
+
+@dataclass
+class TaskEventContext:
+    task_id: str
+    spec_id: str
+    project_id: str
+    sequence_start: int = 0
+
+
+def _load_task_metadata(spec_dir: Path) -> dict:
+    metadata_path = spec_dir / "task_metadata.json"
+    if not metadata_path.exists():
+        return {}
+    try:
+        with open(metadata_path, encoding="utf-8") as f:
+            return json.load(f)
+    except (OSError, json.JSONDecodeError, UnicodeDecodeError):
+        return {}
+
+
+def _load_last_sequence(spec_dir: Path) -> int:
+    plan_path = spec_dir / "implementation_plan.json"
+    if not plan_path.exists():
+        return 0
+    try:
+        with open(plan_path, encoding="utf-8") as f:
+            plan = json.load(f)
+        last_event = plan.get("lastEvent") or {}
+        seq = last_event.get("sequence")
+        if isinstance(seq, int) and seq >= 0:
+            return seq + 1
+    except (OSError, json.JSONDecodeError, UnicodeDecodeError):
+        return 0
+    return 0
+
+
+def load_task_event_context(spec_dir: Path) -> TaskEventContext:
+    metadata = _load_task_metadata(spec_dir)
+    task_id = metadata.get("taskId") or metadata.get("task_id") or spec_dir.name
+    spec_id = metadata.get("specId") or metadata.get("spec_id") or spec_dir.name
+    project_id = metadata.get("projectId") or metadata.get("project_id") or ""
+    sequence_start = _load_last_sequence(spec_dir)
+    return TaskEventContext(
+        task_id=str(task_id),
+        spec_id=str(spec_id),
+        project_id=str(project_id),
+        sequence_start=sequence_start,
+    )
+
+
+class TaskEventEmitter:
+    def __init__(self, context: TaskEventContext) -> None:
+        self._context = context
+        self._sequence = context.sequence_start
+
+    @classmethod
+    def from_spec_dir(cls, spec_dir: Path) -> TaskEventEmitter:
+        return cls(load_task_event_context(spec_dir))
+
+    def emit(self, event_type: str, payload: dict | None = None) -> None:
+        event = {
+            "type": event_type,
+            "taskId": self._context.task_id,
+            "specId": self._context.spec_id,
+            "projectId": self._context.project_id,
+            "timestamp": datetime.now(timezone.utc).isoformat(),
+            "eventId": str(uuid4()),
+            "sequence": self._sequence,
+        }
+        if payload:
+            event.update(payload)
+
+        try:
+            print(f"{TASK_EVENT_PREFIX}{json.dumps(event, default=str)}", flush=True)
+            self._sequence += 1
+        except (OSError, UnicodeEncodeError) as e:
+            if _DEBUG:
+                try:
+                    sys.stderr.write(f"[task_event] emit failed: {e}\n")
+                    sys.stderr.flush()
+                except (OSError, UnicodeEncodeError):
+                    pass  # Silent on complete I/O failure
@@ -121,6 +121,7 @@ from merge import (
    FileTimelineTracker,
    MergeOrchestrator,
 )
+from merge.progress import MergeProgressCallback, MergeProgressStage, emit_progress

 MODULE = "workspace"

@@ -145,6 +146,26 @@ MODULE = "workspace"
 # - _heuristic_merge


+def _create_merge_progress_callback() -> MergeProgressCallback | None:
+    """
+    Create a progress callback for merge operations when running as a subprocess.
+
+    Returns emit_progress (writing JSON to stdout) only when stdout is piped
+    (i.e., running as a subprocess from the Electron frontend). Returns None
+    when running interactively in a terminal to avoid polluting CLI output.
+
+    This function must be called at runtime (not at import time) to ensure
+    sys.stdout state is accurate.
+    """
+    import sys
+
+    # Only emit progress JSON when stdout is piped (subprocess mode).
+    # In interactive CLI mode (TTY), progress JSON would clutter the output.
+    if not sys.stdout.isatty():
+        return emit_progress
+    return None
+
+
 def merge_existing_build(
    project_dir: Path,
    spec_name: str,
@@ -252,10 +273,11 @@ def merge_existing_build(
                had_conflicts = stats.get("conflicts_resolved", 0) > 0
                ai_assisted = stats.get("ai_assisted", 0) > 0
                direct_copy = stats.get("direct_copy", False)
+                git_merge_used = stats.get("git_merge", False)

-                if had_conflicts or ai_assisted or direct_copy:
-                    # AI resolved conflicts, assisted with merges, or direct copy was used
-                    # Changes are already written and staged - no need for git merge
+                if had_conflicts or ai_assisted or direct_copy or git_merge_used:
+                    # AI resolved conflicts, assisted with merges, git merge was used, or direct copy was used
+                    # Changes are already written and staged - no need for additional git merge
                    _print_merge_success(
                        no_commit, stats, spec_name=spec_name, keep_worktree=True
                    )
@@ -402,9 +424,20 @@ def _try_smart_merge_inner(
        no_commit=no_commit,
    )

+    # Create progress callback for subprocess mode (Electron frontend).
+    # Only emits JSON to stdout when piped, not in interactive CLI.
+    progress_callback = _create_merge_progress_callback()
+
    try:
        print(muted("  Analyzing changes with intent-aware merge..."))

+        if progress_callback is not None:
+            progress_callback(
+                MergeProgressStage.ANALYZING,
+                0,
+                "Starting merge analysis",
+            )
+
        # Capture worktree state in FileTimelineTracker before merge
        try:
            timeline_tracker = FileTimelineTracker(project_dir)
@@ -440,6 +473,13 @@ def _try_smart_merge_inner(
        )

        # Check for git-level conflicts first (branch divergence)
+        if progress_callback is not None:
+            progress_callback(
+                MergeProgressStage.DETECTING_CONFLICTS,
+                25,
+                "Checking for git-level conflicts",
+            )
+
        debug(MODULE, "Checking for git-level conflicts")
        git_conflicts = _check_git_conflicts(project_dir, spec_name)

@@ -492,12 +532,11 @@ def _try_smart_merge_inner(
                # If rebase succeeded and now there are no conflicts,
                # the diverged_but_no_conflicts path will handle the merge
            else:
-                # Rebase failed - continue with conflict resolution as before
-                # The AI resolver will handle the conflicts
-                print(
-                    warning(
-                        "  Rebase encountered issues, using AI conflict resolution..."
-                    )
+                # Rebase failed (likely due to worktree lock) - continue with merge
+                # Git merge or AI resolver will handle it depending on conflict state
+                debug(
+                    MODULE,
+                    "Rebase skipped or failed, continuing with merge flow",
                )

        if git_conflicts.get("has_conflicts"):
@@ -518,6 +557,18 @@ def _try_smart_merge_inner(
                num_conflicts=len(git_conflicts.get("conflicting_files", [])),
            )

+            if progress_callback is not None:
+                progress_callback(
+                    MergeProgressStage.RESOLVING,
+                    50,
+                    f"Resolving {len(git_conflicts.get('conflicting_files', []))} conflicting files with AI",
+                    {
+                        "conflicts_found": len(
+                            git_conflicts.get("conflicting_files", [])
+                        )
+                    },
+                )
+
            # Try to resolve git conflicts with AI
            resolution_result = _resolve_git_conflicts_with_ai(
                project_dir,
@@ -535,6 +586,22 @@ def _try_smart_merge_inner(
                    resolved_files=resolution_result.get("resolved_files", []),
                    stats=resolution_result.get("stats", {}),
                )
+
+                if progress_callback is not None:
+                    stats = resolution_result.get("stats", {})
+                    original_conflict_count = len(
+                        git_conflicts.get("conflicting_files", [])
+                    )
+                    progress_callback(
+                        MergeProgressStage.COMPLETE,
+                        100,
+                        "Merge complete",
+                        {
+                            "conflicts_found": original_conflict_count,
+                            "conflicts_resolved": stats.get("conflicts_resolved", 0),
+                        },
+                    )
+
                return resolution_result
            else:
                # AI couldn't resolve all conflicts
@@ -547,6 +614,26 @@ def _try_smart_merge_inner(
                    resolved_files=resolution_result.get("resolved_files", []),
                    error=resolution_result.get("error"),
                )
+
+                if progress_callback is not None:
+                    original_conflict_count = len(
+                        git_conflicts.get("conflicting_files", [])
+                    )
+                    remaining_count = len(
+                        resolution_result.get("remaining_conflicts", [])
+                    )
+                    progress_callback(
+                        MergeProgressStage.ERROR,
+                        0,
+                        "Some conflicts could not be resolved",
+                        {
+                            "conflicts_found": original_conflict_count,
+                            "conflicts_resolved": original_conflict_count
+                            - remaining_count,
+                            "conflicts_remaining": remaining_count,
+                        },
+                    )
+
                return {
                    "success": False,
                    "conflicts": resolution_result.get("remaining_conflicts", []),
@@ -555,148 +642,81 @@ def _try_smart_merge_inner(
                    "error": resolution_result.get("error"),
                }

-        # Check if branches diverged but no actual conflicts (can do direct copy)
+        # Check if branches diverged but no actual conflicts (use git merge)
        if git_conflicts.get("diverged_but_no_conflicts"):
-            debug(MODULE, "Branches diverged but no conflicts - doing direct file copy")
+            debug(MODULE, "Branches diverged but no conflicts - using git merge")
            print(muted("  Branches diverged but no conflicts detected"))
-            print(muted("  Copying changed files directly from worktree..."))
+            print(muted("  Using git merge to combine changes..."))

-            # Get changed files from spec branch
            spec_branch = f"auto-claude/{spec_name}"
-            base_branch = git_conflicts.get("base_branch", "main")

-            # Get merge-base for diff
-            merge_base_result = run_git(
-                ["merge-base", base_branch, spec_branch],
+            # Use git merge --no-commit to combine changes from both branches
+            # Since merge-tree confirmed no conflicts, this should succeed cleanly
+            merge_result = run_git(
+                ["merge", "--no-commit", "--no-ff", spec_branch],
                cwd=project_dir,
            )
-            merge_base = (
-                merge_base_result.stdout.strip()
-                if merge_base_result.returncode == 0
-                else None
-            )

-            if merge_base:
-                # Get list of changed files in spec branch
-                changed_files = _get_changed_files_from_branch(
-                    project_dir, merge_base, spec_branch
+            if merge_result.returncode == 0:
+                # Merge succeeded - get list of files that were merged
+                # Use git diff --cached to see what's staged
+                diff_result = run_git(
+                    ["diff", "--cached", "--name-only"],
+                    cwd=project_dir,
+                )
+                merged_files = [
+                    f.strip()
+                    for f in diff_result.stdout.splitlines()
+                    if f.strip() and not _is_auto_claude_file(f.strip())
+                ]
+
+                debug_success(
+                    MODULE,
+                    "Git merge succeeded",
+                    merged_files_count=len(merged_files),
                )

-                resolved_files = []
-                skipped_files = []  # Track files that failed to copy
-                files_to_stage = []
-                for file_path, status in changed_files:
-                    if _is_auto_claude_file(file_path):
-                        continue
+                for file_path in merged_files:
+                    print(success(f"    ✓ {file_path}"))

-                    try:
-                        target_path = project_dir / file_path
-
-                        if status == "D":
-                            # Deleted in worktree
-                            if target_path.exists():
-                                target_path.unlink()
-                            files_to_stage.append(file_path)
-                            resolved_files.append(file_path)
-                            print(success(f"    ✓ {file_path} (deleted)"))
-                        else:
-                            # New or modified - copy from spec branch
-                            target_path.parent.mkdir(parents=True, exist_ok=True)
-
-                            if _is_binary_file(file_path):
-                                binary_content = _get_binary_file_content_from_ref(
-                                    project_dir, spec_branch, file_path
-                                )
-                                if binary_content is not None:
-                                    target_path.write_bytes(binary_content)
-                                    files_to_stage.append(file_path)
-                                    resolved_files.append(file_path)
-                                    status_label = (
-                                        "new file" if status == "A" else "updated"
-                                    )
-                                    print(
-                                        success(f"    ✓ {file_path} ({status_label})")
-                                    )
-                                else:
-                                    skipped_files.append(file_path)
-                                    debug_warning(
-                                        MODULE,
-                                        f"Could not retrieve binary content for {file_path}",
-                                    )
-                            else:
-                                content = _get_file_content_from_ref(
-                                    project_dir, spec_branch, file_path
-                                )
-                                if content is not None:
-                                    target_path.write_text(content, encoding="utf-8")
-                                    files_to_stage.append(file_path)
-                                    resolved_files.append(file_path)
-                                    status_label = (
-                                        "new file" if status == "A" else "updated"
-                                    )
-                                    print(
-                                        success(f"    ✓ {file_path} ({status_label})")
-                                    )
-                                else:
-                                    skipped_files.append(file_path)
-                                    debug_warning(
-                                        MODULE,
-                                        f"Could not retrieve content for {file_path}",
-                                    )
-
-                    except Exception as e:
-                        skipped_files.append(file_path)
-                        debug_warning(MODULE, f"Could not copy {file_path}: {e}")
-
-                # Stage all files in a single git add call for efficiency
-                if files_to_stage:
-                    add_result = run_git(
-                        ["add"] + files_to_stage,
-                        cwd=project_dir,
+                if progress_callback is not None:
+                    progress_callback(
+                        MergeProgressStage.COMPLETE,
+                        100,
+                        f"Git merge complete ({len(merged_files)} files)",
                    )
-                    if add_result.returncode != 0:
-                        debug_warning(
-                            MODULE,
-                            f"Failed to stage files for direct copy: {add_result.stderr}",
-                        )
-                        # Return failure - files were written but not staged
-                        return {
-                            "success": False,
-                            "error": f"Failed to stage files: {add_result.stderr}",
-                            "resolved_files": [],
-                        }

-                # Build result - check for skipped files to detect partial merges
-                result = {
-                    "success": len(skipped_files) == 0,
-                    "resolved_files": resolved_files,
+                return {
+                    "success": True,
+                    "resolved_files": merged_files,
                    "stats": {
-                        "files_merged": len(resolved_files),
+                        "files_merged": len(merged_files),
                        "conflicts_resolved": 0,
                        "ai_assisted": 0,
-                        "auto_merged": len(resolved_files),
-                        "direct_copy": True,  # Flag indicating direct copy was used
-                        "skipped_count": len(skipped_files),
+                        "auto_merged": len(merged_files),
+                        "git_merge": True,  # Flag indicating git merge was used
                    },
                }
-                if skipped_files:
-                    result["skipped_files"] = skipped_files
-                    result["partial_success"] = len(resolved_files) > 0
-                    print()
-                    print(
-                        warning(
-                            f"  ⚠ {len(skipped_files)} file(s) could not be retrieved:"
-                        )
-                    )
-                    for skipped_file in skipped_files:
-                        print(muted(f"    - {skipped_file}"))
-                    print(muted("  These files may need manual review."))
-                return result
            else:
-                # merge-base failed - branches may not share history
+                # Merge failed unexpectedly - abort and fall back to semantic analysis
                debug_warning(
                    MODULE,
-                    "Could not find merge-base between branches - falling back to semantic analysis",
+                    "Git merge failed unexpectedly despite no conflicts detected",
+                    stderr=merge_result.stderr[:500] if merge_result.stderr else "",
+                )
+                # Abort the merge to restore clean state
+                abort_result = run_git(["merge", "--abort"], cwd=project_dir)
+                if abort_result.returncode != 0:
+                    debug_error(
+                        MODULE,
+                        "Failed to abort merge - repo may be in inconsistent state",
+                        stderr=abort_result.stderr,
+                    )
+                    return None  # Trigger fallback to avoid operating on inconsistent state
+                print(
+                    warning(
+                        "  Git merge failed unexpectedly, falling back to semantic analysis..."
+                    )
                )

        # No git conflicts - proceed with semantic analysis
@@ -725,6 +745,14 @@ def _try_smart_merge_inner(

        # All conflicts can be auto-merged or no conflicts
        print(muted("  All changes compatible, proceeding with merge..."))
+
+        if progress_callback is not None:
+            progress_callback(
+                MergeProgressStage.COMPLETE,
+                100,
+                f"Analysis complete ({files_to_merge} files compatible)",
+            )
+
        return {
            "success": True,
            "stats": {
@@ -737,6 +765,13 @@ def _try_smart_merge_inner(
        # If smart merge fails, fall back to git
        import traceback

+        if progress_callback is not None:
+            progress_callback(
+                MergeProgressStage.ERROR,
+                0,
+                f"Smart merge error: {e}",
+            )
+
        print(muted(f"  Smart merge error: {e}"))
        traceback.print_exc()
        return None
@@ -748,14 +783,11 @@ def _rebase_spec_branch(
    base_branch: str,
 ) -> bool:
    """
-    Rebase the spec branch onto the latest base branch.
+    Attempt to rebase the spec branch onto the latest base branch.

-    This performs an automatic rebase of the spec branch onto the current
-    base branch (main/develop) to bring it up to date before merging.
-    If conflicts occur during rebase, the function aborts and returns False
-    so that the caller can fall back to AI conflict resolution.
-
-    The function preserves the current HEAD by restoring it after completion.
+    NOTE: This will fail if the spec branch is checked out in a worktree,
+    which is the normal case. The caller should handle failure gracefully
+    by falling back to git merge or AI conflict resolution.

    Args:
        project_dir: The project directory
@@ -764,22 +796,36 @@ def _rebase_spec_branch(

    Returns:
        True if rebase succeeded cleanly or branch was already up-to-date,
-        False if rebase failed due to conflicts or other errors (aborted, no ref movement)
+        False if rebase failed (worktree lock, conflicts, or other errors)
    """
    spec_branch = f"auto-claude/{spec_name}"

    debug(
        MODULE,
-        "Rebasing spec branch",
+        "Attempting to rebase spec branch",
        spec_branch=spec_branch,
        base_branch=base_branch,
    )

-    # Save original branch to restore after rebase (HIGH: prevents leaving repo on spec branch)
+    # Check if spec branch is used by a worktree (common case)
+    # In this case, we can't checkout/rebase from the main repo
+    worktree_list_result = run_git(["worktree", "list", "--porcelain"], cwd=project_dir)
+    if worktree_list_result.returncode == 0:
+        # Check if spec_branch is in use by a worktree
+        output = worktree_list_result.stdout
+        if f"branch refs/heads/{spec_branch}" in output:
+            debug(
+                MODULE,
+                "Spec branch is checked out in a worktree - skipping rebase",
+                spec_branch=spec_branch,
+            )
+            # This is expected - return False to let caller use git merge instead
+            return False
+
+    # Save original branch to restore after rebase
    original_branch_result = run_git(
        ["rev-parse", "--abbrev-ref", "HEAD"], cwd=project_dir
    )
-    # Check returncode and validate stdout before using original_branch
    if original_branch_result.returncode != 0:
        debug_error(
            MODULE,
@@ -795,7 +841,6 @@ def _rebase_spec_branch(
        )
        return False

-    # Save current state for recovery
    # Get the current commit of spec_branch before rebase
    before_commit_result = run_git(["rev-parse", spec_branch], cwd=project_dir)
    if before_commit_result.returncode != 0:
@@ -804,8 +849,6 @@ def _rebase_spec_branch(
            "Could not get spec branch commit before rebase",
            stderr=before_commit_result.stderr,
        )
-        # Restore original branch before returning
-        run_git(["checkout", original_branch], cwd=project_dir)
        return False
    before_commit = before_commit_result.stdout.strip()

@@ -813,22 +856,18 @@ def _rebase_spec_branch(
    print(muted(f"  Rebasing {spec_branch} onto {base_branch}..."))

    try:
-        # Perform the rebase using safe/standard invocation:
-        # 1. Checkout the spec branch first
-        # 2. Run standard rebase (no strategy options - let conflicts stop the rebase)
-        # If conflicts occur, we'll abort and let AI handle them during merge
+        # Try to checkout the spec branch
        checkout_result = run_git(["checkout", spec_branch], cwd=project_dir)
        if checkout_result.returncode != 0:
-            debug_error(
+            # Checkout failed - likely due to worktree lock
+            debug(
                MODULE,
-                "Could not checkout spec branch for rebase",
-                stderr=checkout_result.stderr,
+                "Could not checkout spec branch for rebase (likely worktree lock)",
+                stderr=checkout_result.stderr[:200] if checkout_result.stderr else "",
            )
            return False

-        # Run standard rebase - will stop on conflicts so we can detect them
-        # Git syntax: git rebase [options] <upstream>
-        # where <upstream> is the branch to rebase onto
+        # Run standard rebase
        rebase_result = run_git(
            ["rebase", base_branch],
            cwd=project_dir,
@@ -838,9 +877,6 @@ def _rebase_spec_branch(
            # Rebase failed - check if it was due to conflicts
            status_result = run_git(["status", "--porcelain"], cwd=project_dir)

-            # MEDIUM: Properly parse git status output for conflict markers
-            # Git status --porcelain uses two-character status codes:
-            # UU = both modified, AA = both added, DD = both deleted, etc.
            has_unmerged = any(
                line[:2] in ("UU", "AA", "DD", "AU", "UA", "DU", "UD")
                for line in status_result.stdout.splitlines()
@@ -848,7 +884,6 @@ def _rebase_spec_branch(
            )

            # Abort the rebase to return to clean state
-            # NEW-002: If abort fails, immediately return False (repo in bad state)
            abort_result = run_git(["rebase", "--abort"], cwd=project_dir)
            if abort_result.returncode != 0:
                debug_error(
@@ -856,19 +891,16 @@ def _rebase_spec_branch(
                    "Failed to abort rebase - repo may be in inconsistent state",
                    stderr=abort_result.stderr,
                )
-                return False  # Abort failed - cannot safely continue
-
-            if has_unmerged:
-                # Rebase failed due to conflicts - we aborted, so no ref movement happened
-                debug_warning(
-                    MODULE,
-                    "Rebase encountered conflicts - aborted, will use AI conflict resolution",
-                    stderr=rebase_result.stderr[:200] if rebase_result.stderr else "",
-                )
-                # Return False since we aborted - no rebase occurred, caller should use AI
                return False

-            # Other error (not conflict-related)
+            if has_unmerged:
+                debug_warning(
+                    MODULE,
+                    "Rebase encountered conflicts - aborted, will use alternative merge",
+                    stderr=rebase_result.stderr[:200] if rebase_result.stderr else "",
+                )
+                return False
+
            debug_error(
                MODULE,
                "Rebase failed with unexpected error",
@@ -882,9 +914,7 @@ def _rebase_spec_branch(
        if after_commit_result.returncode == 0:
            after_commit_hash = after_commit_result.stdout.strip()

-            # Verify the branch actually moved (commit changed)
            if before_commit == after_commit_hash:
-                # MEDIUM: Branch already up-to-date is a success condition, not failure
                debug(
                    MODULE,
                    "Branch already up-to-date, no rebase needed",
@@ -904,8 +934,7 @@ def _rebase_spec_branch(
        debug_error(MODULE, "Could not verify spec branch commit after rebase")
        return False
    finally:
-        # HIGH: Always restore original branch, even on error/exception
-        # NEW-001: Log restoration failure (cannot modify return from finally block)
+        # Always restore original branch
        if original_branch:
            restore_result = run_git(["checkout", original_branch], cwd=project_dir)
            if restore_result.returncode != 0:
@@ -914,8 +943,6 @@ def _rebase_spec_branch(
                    f"Failed to restore original branch '{original_branch}'",
                    stderr=restore_result.stderr,
                )
-                # Note: Cannot modify return value from finally block,
-                # but restoration failure is rare and non-critical (user can manually switch back)


 def _check_git_conflicts(project_dir: Path, spec_name: str) -> dict:
@@ -325,6 +325,7 @@ def setup_workspace(
    mode: WorkspaceMode,
    source_spec_dir: Path | None = None,
    base_branch: str | None = None,
+    use_local_branch: bool = False,
 ) -> tuple[Path, WorktreeManager | None, Path | None]:
    """
    Set up the workspace based on user's choice.
@@ -337,6 +338,7 @@ def setup_workspace(
        mode: The workspace mode to use
        source_spec_dir: Optional source spec directory to copy to worktree
        base_branch: Base branch for worktree creation (default: current branch)
+        use_local_branch: If True, use local branch directly instead of preferring origin/branch

    Returns:
        Tuple of (working_directory, worktree_manager or None, localized_spec_dir or None)
@@ -357,7 +359,9 @@ def setup_workspace(
    # Ensure timeline tracking hook is installed (once per session)
    ensure_timeline_hook_installed(project_dir)

-    manager = WorktreeManager(project_dir, base_branch=base_branch)
+    manager = WorktreeManager(
+        project_dir, base_branch=base_branch, use_local_branch=use_local_branch
+    )
    manager.setup()

    # Get or create worktree for THIS SPECIFIC SPEC
@@ -15,6 +15,8 @@ This allows:
 """

 import asyncio
+import json
+import logging
 import os
 import re
 import shutil
@@ -28,8 +30,13 @@ from typing import TypedDict, TypeVar

 from core.gh_executable import get_gh_executable, invalidate_gh_cache
 from core.git_executable import get_git_executable, get_isolated_git_env, run_git
+from core.git_provider import detect_git_provider
+from core.glab_executable import get_glab_executable, invalidate_glab_cache
+from core.model_config import get_utility_model_config
 from debug import debug_warning

+logger = logging.getLogger(__name__)
+
 T = TypeVar("T")


@@ -136,6 +143,7 @@ class PushAndCreatePRResult(TypedDict, total=False):
    pushed: bool
    remote: str
    branch: str
+    provider: str  # 'github', 'gitlab', or 'unknown'
    pr_url: str | None  # None when PR was created but URL couldn't be extracted
    already_exists: bool
    error: str
@@ -175,12 +183,18 @@ class WorktreeManager:

    # Timeout constants for subprocess operations
    GIT_PUSH_TIMEOUT = 120  # 2 minutes for git push (network operations)
-    GH_CLI_TIMEOUT = 60  # 1 minute for gh CLI commands
-    GH_QUERY_TIMEOUT = 30  # 30 seconds for gh CLI queries
+    CLI_TIMEOUT = 60  # 1 minute for CLI commands (gh/glab)
+    CLI_QUERY_TIMEOUT = 30  # 30 seconds for CLI queries (gh/glab)

-    def __init__(self, project_dir: Path, base_branch: str | None = None):
+    def __init__(
+        self,
+        project_dir: Path,
+        base_branch: str | None = None,
+        use_local_branch: bool = False,
+    ):
        self.project_dir = project_dir
        self.base_branch = base_branch or self._detect_base_branch()
+        self.use_local_branch = use_local_branch
        self.worktrees_dir = project_dir / ".auto-claude" / "worktrees" / "tasks"
        self._merge_lock = asyncio.Lock()

@@ -687,18 +701,23 @@ class WorktreeManager:
        else:
            # Branch doesn't exist - create new branch from remote or local base
            # Determine the start point for the worktree
-            remote_ref = f"origin/{self.base_branch}"
            start_point = self.base_branch  # Default to local branch

-            # Check if remote ref exists and use it as the source of truth
-            check_remote = self._run_git(["rev-parse", "--verify", remote_ref])
-            if check_remote.returncode == 0:
-                start_point = remote_ref
-                print(f"Creating worktree from remote: {remote_ref}")
+            if self.use_local_branch:
+                # User explicitly requested local branch - skip auto-switch to remote
+                # This preserves gitignored files (.env, configs) that may not exist on remote
+                print(f"Creating worktree from local branch: {self.base_branch}")
            else:
-                print(
-                    f"Remote ref {remote_ref} not found, using local branch: {self.base_branch}"
-                )
+                # Check if remote ref exists and use it as the source of truth
+                remote_ref = f"origin/{self.base_branch}"
+                check_remote = self._run_git(["rev-parse", "--verify", remote_ref])
+                if check_remote.returncode == 0:
+                    start_point = remote_ref
+                    print(f"Creating worktree from remote: {remote_ref}")
+                else:
+                    print(
+                        f"Remote ref {remote_ref} not found, using local branch: {self.base_branch}"
+                    )

            # Create worktree with new branch from the start point
            result = self._run_git(
@@ -1192,8 +1211,22 @@ class WorktreeManager:
        target = target_branch or self.base_branch
        pr_title = title or f"auto-claude: {spec_name}"

-        # Get PR body from spec.md if available
-        pr_body = self._extract_spec_summary(spec_name)
+        # Try AI-powered PR body from project's PR template, fall back to spec summary
+        pr_body: str | None = None
+        try:
+            diff_summary, commit_log = self._gather_pr_context(spec_name, target)
+            pr_body = self._try_ai_pr_body(
+                spec_name=spec_name,
+                target_branch=target,
+                branch_name=info.branch,
+                diff_summary=diff_summary,
+                commit_log=commit_log,
+            )
+        except Exception as e:
+            logger.warning(f"AI PR body generation encountered an error: {e}")
+
+        if not pr_body:
+            pr_body = self._extract_spec_summary(spec_name)

        # Find gh executable before attempting PR creation
        gh_executable = get_gh_executable()
@@ -1236,7 +1269,7 @@ class WorktreeManager:
                    text=True,
                    encoding="utf-8",
                    errors="replace",
-                    timeout=self.GH_CLI_TIMEOUT,
+                    timeout=self.CLI_TIMEOUT,
                    env=get_isolated_git_env(),
                )

@@ -1312,9 +1345,328 @@ class WorktreeManager:
            invalidate_gh_cache()
            return PullRequestResult(
                success=False,
-                error="gh CLI not found. Install from https://cli.github.com/",
+                error="GitHub CLI (gh) not found. Install from https://cli.github.com/",
            )

+    def create_merge_request(
+        self,
+        spec_name: str,
+        target_branch: str | None = None,
+        title: str | None = None,
+        draft: bool = False,
+    ) -> PullRequestResult:
+        """
+        Create a GitLab merge request for a spec's branch using glab CLI with retry logic.
+
+        Args:
+            spec_name: The spec folder name
+            target_branch: Target branch for MR (defaults to base_branch)
+            title: MR title (defaults to spec name)
+            draft: Whether to create as draft MR
+
+        Returns:
+            PullRequestResult with keys:
+                - success: bool
+                - pr_url: str (if created)
+                - already_exists: bool (if MR already exists)
+                - error: str (if failed)
+        """
+        info = self.get_worktree_info(spec_name)
+        if not info:
+            return PullRequestResult(
+                success=False,
+                error=f"No worktree found for spec: {spec_name}",
+            )
+
+        target = target_branch or self.base_branch
+        mr_title = title or f"auto-claude: {spec_name}"
+
+        # Get MR body from spec.md if available
+        mr_body = self._extract_spec_summary(spec_name)
+
+        # Find glab executable before attempting MR creation
+        glab_executable = get_glab_executable()
+        if not glab_executable:
+            return PullRequestResult(
+                success=False,
+                error="GitLab CLI (glab) not found. Install from https://gitlab.com/gitlab-org/cli",
+            )
+
+        # Build glab mr create command
+        glab_args = [
+            glab_executable,
+            "mr",
+            "create",
+            "--target-branch",
+            target,
+            "--source-branch",
+            info.branch,
+            "--title",
+            mr_title,
+            "--description",
+            mr_body,
+        ]
+        if draft:
+            glab_args.append("--draft")
+
+        def is_mr_retryable(stderr: str) -> bool:
+            """Check if MR creation error is retryable (network or HTTP 5xx)."""
+            return _is_retryable_network_error(stderr) or _is_retryable_http_error(
+                stderr
+            )
+
+        def do_create_mr() -> tuple[bool, PullRequestResult | None, str]:
+            """Execute MR creation for retry wrapper."""
+            try:
+                result = subprocess.run(
+                    glab_args,
+                    cwd=info.path,
+                    capture_output=True,
+                    text=True,
+                    encoding="utf-8",
+                    errors="replace",
+                    timeout=self.CLI_TIMEOUT,
+                    env=get_isolated_git_env(),
+                )
+
+                # Check for "already exists" case (success, no retry needed)
+                if result.returncode != 0 and "already exists" in result.stderr.lower():
+                    existing_url = self._get_existing_mr_url(spec_name, target)
+                    result_dict = PullRequestResult(
+                        success=True,
+                        pr_url=existing_url,
+                        already_exists=True,
+                    )
+                    if existing_url is None:
+                        result_dict["message"] = (
+                            "MR already exists but URL could not be retrieved"
+                        )
+                    return (True, result_dict, "")
+
+                if result.returncode == 0:
+                    # Extract MR URL from output
+                    mr_url: str | None = result.stdout.strip()
+                    if not mr_url.startswith("http"):
+                        # Try to find URL in output
+                        # GitLab URL pattern: matches any HTTPS URL with /merge_requests/<number> or /-/merge_requests/<number> path
+                        match = re.search(
+                            r"https://[^\s]+(?:/merge_requests/|/-/merge_requests/)\d+",
+                            result.stdout,
+                        )
+                        if match:
+                            mr_url = match.group(0)
+                        else:
+                            # Invalid output - no valid URL found
+                            mr_url = None
+
+                    return (
+                        True,
+                        PullRequestResult(
+                            success=True,
+                            pr_url=mr_url,
+                            already_exists=False,
+                        ),
+                        "",
+                    )
+
+                return (False, None, result.stderr)
+
+            except FileNotFoundError:
+                # glab CLI not installed - not retryable, raise to exit retry loop
+                raise
+
+        max_retries = 3
+        try:
+            result, last_error = _with_retry(
+                operation=do_create_mr,
+                max_retries=max_retries,
+                is_retryable=is_mr_retryable,
+            )
+
+            if result:
+                return result
+
+            # Handle timeout error message
+            if last_error == "Operation timed out":
+                return PullRequestResult(
+                    success=False,
+                    error=f"MR creation timed out after {max_retries} attempts.",
+                )
+
+            return PullRequestResult(
+                success=False,
+                error=f"Failed to create MR: {last_error}",
+            )
+
+        except FileNotFoundError:
+            # Cached glab path became invalid - clear cache so next call re-discovers
+            invalidate_glab_cache()
+            return PullRequestResult(
+                success=False,
+                error="GitLab CLI (glab) not found. Install from https://gitlab.com/gitlab-org/cli",
+            )
+
+    def _gather_pr_context(self, spec_name: str, target_branch: str) -> tuple[str, str]:
+        """
+        Gather diff summary and commit log for PR template filling.
+
+        Args:
+            spec_name: The spec folder name
+            target_branch: The target branch for the PR
+
+        Returns:
+            Tuple of (diff_summary, commit_log)
+        """
+        worktree_path = self.get_worktree_path(spec_name)
+        info = self.get_worktree_info(spec_name)
+        branch = info.branch if info else self.get_branch_name(spec_name)
+
+        # Get diff summary (stat for overview)
+        diff_result = self._run_git(
+            ["diff", "--stat", f"{target_branch}...{branch}"],
+            cwd=worktree_path,
+            timeout=30,
+        )
+        diff_summary = diff_result.stdout.strip() if diff_result.returncode == 0 else ""
+
+        # Get shortstat for quick summary
+        shortstat_result = self._run_git(
+            ["diff", "--shortstat", f"{target_branch}...{branch}"],
+            cwd=worktree_path,
+            timeout=30,
+        )
+        if shortstat_result.returncode == 0 and shortstat_result.stdout.strip():
+            diff_summary += "\n\n" + shortstat_result.stdout.strip()
+
+        # Get actual code changes (patch format) for better AI context
+        # Truncate to 30k chars to avoid token limits while still providing meaningful context
+        patch_result = self._run_git(
+            ["diff", "-p", "--stat-width=999", f"{target_branch}...{branch}"],
+            cwd=worktree_path,
+            timeout=30,
+        )
+        if patch_result.returncode == 0 and patch_result.stdout.strip():
+            patch_content = patch_result.stdout.strip()
+            MAX_DIFF_CHARS = 30_000
+
+            if len(patch_content) > MAX_DIFF_CHARS:
+                # Truncate patch and add notice
+                truncated_patch = patch_content[:MAX_DIFF_CHARS]
+                diff_summary += (
+                    "\n\n" + truncated_patch + "\n\n(... diff truncated due to size)"
+                )
+            else:
+                diff_summary += "\n\n" + patch_content
+
+        # Get commit log
+        log_result = self._run_git(
+            [
+                "log",
+                "--oneline",
+                "--no-merges",
+                f"{target_branch}..{branch}",
+            ],
+            cwd=worktree_path,
+            timeout=30,
+        )
+        commit_log = log_result.stdout.strip() if log_result.returncode == 0 else ""
+
+        return diff_summary, commit_log
+
+    def _try_ai_pr_body(
+        self,
+        spec_name: str,
+        target_branch: str,
+        branch_name: str,
+        diff_summary: str,
+        commit_log: str,
+    ) -> str | None:
+        """
+        Attempt to generate a PR body using the AI template filler agent.
+
+        Runs the async agent synchronously with a 30-second timeout.
+        Returns None on any failure so the caller can fall back gracefully.
+
+        Args:
+            spec_name: The spec folder name
+            target_branch: The target branch for the PR
+            branch_name: The source branch name
+            diff_summary: Git diff summary of changes
+            commit_log: Git log of commits
+
+        Returns:
+            The AI-generated PR body string, or None if unavailable.
+        """
+        try:
+            from agents.pr_template_filler import (
+                detect_pr_template,
+                run_pr_template_filler,
+            )
+        except ImportError:
+            logger.warning(
+                "PR template filler module not available, skipping AI PR body"
+            )
+            return None
+
+        # Check if a PR template exists before doing any heavy lifting
+        template = detect_pr_template(self.project_dir)
+        if template is None:
+            return None
+
+        # Resolve spec directory
+        spec_dir = self.project_dir / ".auto-claude" / "specs" / spec_name
+        if not spec_dir.is_dir():
+            # Try worktree-local spec path
+            worktree_path = self.get_worktree_path(spec_name)
+            spec_dir = worktree_path / ".auto-claude" / "specs" / spec_name
+            if not spec_dir.is_dir():
+                logger.warning("Spec directory not found for AI PR body generation")
+                return None
+
+        # Get model configuration from environment (respects user settings)
+        model, thinking_budget = get_utility_model_config()
+
+        async def _run_with_timeout() -> str | None:
+            try:
+                return await asyncio.wait_for(
+                    run_pr_template_filler(
+                        project_dir=self.project_dir,
+                        spec_dir=spec_dir,
+                        model=model,
+                        thinking_budget=thinking_budget,
+                        branch_name=branch_name,
+                        target_branch=target_branch,
+                        diff_summary=diff_summary,
+                        commit_log=commit_log,
+                        verbose=False,
+                    ),
+                    timeout=30.0,
+                )
+            except asyncio.TimeoutError:
+                logger.warning("PR template filler timed out after 30s")
+                return None
+
+        try:
+            # Check if there's already a running event loop
+            try:
+                loop = asyncio.get_running_loop()
+            except RuntimeError:
+                loop = None
+
+            if loop and loop.is_running():
+                # We're already inside an async context — run in a new thread
+                import concurrent.futures
+
+                with concurrent.futures.ThreadPoolExecutor(max_workers=1) as pool:
+                    future = pool.submit(asyncio.run, _run_with_timeout())
+                    return future.result(timeout=35)
+            else:
+                return asyncio.run(_run_with_timeout())
+
+        except Exception as e:
+            logger.warning(f"AI PR body generation failed: {e}")
+            return None
+
    def _extract_spec_summary(self, spec_name: str) -> str:
        """Extract a summary from spec.md for PR body."""
        worktree_path = self.get_worktree_path(spec_name)
@@ -1389,7 +1741,7 @@ class WorktreeManager:
                text=True,
                encoding="utf-8",
                errors="replace",
-                timeout=self.GH_QUERY_TIMEOUT,
+                timeout=self.CLI_QUERY_TIMEOUT,
                env=get_isolated_git_env(),
            )
            if result.returncode == 0:
@@ -1408,6 +1760,57 @@ class WorktreeManager:

        return None

+    def _get_existing_mr_url(self, spec_name: str, target_branch: str) -> str | None:
+        """Get the URL of an existing MR for this branch."""
+        info = self.get_worktree_info(spec_name)
+        if not info:
+            return None
+
+        glab_executable = get_glab_executable()
+        if not glab_executable:
+            # glab CLI not found - return None and let caller handle it
+            return None
+
+        try:
+            result = subprocess.run(
+                [
+                    glab_executable,
+                    "mr",
+                    "view",
+                    info.branch,
+                    "--output",
+                    "json",
+                ],
+                cwd=info.path,
+                capture_output=True,
+                text=True,
+                encoding="utf-8",
+                errors="replace",
+                timeout=self.CLI_QUERY_TIMEOUT,
+                env=get_isolated_git_env(),
+            )
+            if result.returncode == 0 and result.stdout.strip():
+                # Parse JSON output to extract web_url (glab uses snake_case)
+                try:
+                    data = json.loads(result.stdout)
+                    return data.get("web_url")
+                except json.JSONDecodeError:
+                    # If JSON parsing fails, return None
+                    pass
+        except (
+            subprocess.TimeoutExpired,
+            FileNotFoundError,
+            subprocess.SubprocessError,
+        ) as e:
+            # Silently ignore errors when fetching existing MR URL - this is a best-effort
+            # lookup that may fail due to network issues, missing glab CLI, or auth problems.
+            # Returning None allows the caller to handle missing URLs gracefully.
+            if isinstance(e, FileNotFoundError):
+                invalidate_glab_cache()
+            debug_warning("worktree", f"Could not get existing MR URL: {e}")
+
+        return None
+
    def push_and_create_pr(
        self,
        spec_name: str,
@@ -1417,13 +1820,14 @@ class WorktreeManager:
        force_push: bool = False,
    ) -> PushAndCreatePRResult:
        """
-        Push branch and create a pull request in one operation.
+        Push branch and create a pull request/merge request in one operation.
+        Automatically detects git provider (GitHub or GitLab) and routes to the appropriate CLI.

        Args:
            spec_name: The spec folder name
-            target_branch: Target branch for PR (defaults to base_branch)
-            title: PR title (defaults to spec name)
-            draft: Whether to create as draft PR
+            target_branch: Target branch for PR/MR (defaults to base_branch)
+            title: PR/MR title (defaults to spec name)
+            draft: Whether to create as draft PR/MR
            force_push: Whether to force push the branch

        Returns:
@@ -1431,7 +1835,8 @@ class WorktreeManager:
                - success: bool
                - pr_url: str (if created)
                - pushed: bool (if push succeeded)
-                - already_exists: bool (if PR already exists)
+                - provider: str ('github', 'gitlab', or 'unknown')
+                - already_exists: bool (if PR/MR already exists)
                - error: str (if failed)
        """
        # Step 1: Push the branch
@@ -1445,20 +1850,44 @@ class WorktreeManager:
                error=push_result.get("error", "Push failed"),
            )

-        # Step 2: Create the PR
-        pr_result = self.create_pull_request(
-            spec_name=spec_name,
-            target_branch=target_branch,
-            title=title,
-            draft=draft,
+        # Step 2: Detect git provider (use the remote that was pushed to)
+        provider = detect_git_provider(
+            self.project_dir, remote_name=push_result.get("remote")
        )

+        # Step 3: Create the PR/MR based on provider
+        if provider == "github":
+            pr_result = self.create_pull_request(
+                spec_name=spec_name,
+                target_branch=target_branch,
+                title=title,
+                draft=draft,
+            )
+        elif provider == "gitlab":
+            pr_result = self.create_merge_request(
+                spec_name=spec_name,
+                target_branch=target_branch,
+                title=title,
+                draft=draft,
+            )
+        else:
+            # Unknown provider
+            return PushAndCreatePRResult(
+                success=False,
+                pushed=True,
+                remote=push_result.get("remote"),
+                branch=push_result.get("branch"),
+                provider=provider,
+                error="Unable to determine git hosting provider. Supported: GitHub, GitLab.",
+            )
+
        # Combine results
        return PushAndCreatePRResult(
            success=pr_result.get("success", False),
            pushed=True,
            remote=push_result.get("remote"),
            branch=push_result.get("branch"),
+            provider=provider,
            pr_url=pr_result.get("pr_url"),
            already_exists=pr_result.get("already_exists", False),
            error=pr_result.get("error"),
@@ -91,11 +91,14 @@ class IdeationGenerator:
            prompt += f"\n{additional_context}\n"

        # Create client with thinking budget
+        # Use agent_type="ideation" to avoid loading unnecessary MCP servers
+        # which can cause 60-second timeout delays
        client = create_client(
            self.project_dir,
            self.output_dir,
            resolve_model_id(self.model),
            max_thinking_tokens=self.thinking_budget,
+            agent_type="ideation",
        )

        try:
@@ -184,11 +187,13 @@ Common fixes:
 Write the fixed JSON to the file now.
 """

+        # Use agent_type="ideation" for recovery agent as well
        client = create_client(
            self.project_dir,
            self.output_dir,
            resolve_model_id(self.model),
            max_thinking_tokens=self.thinking_budget,
+            agent_type="ideation",
        )

        try:
@@ -28,6 +28,7 @@ from .types import IdeationPhaseResult

 # Configuration
 MAX_RETRIES = 3
+IDEATION_TIMEOUT_SECONDS = 5 * 60  # 5 minutes max for all ideation types


 class IdeationOrchestrator:
@@ -173,16 +174,45 @@ class IdeationOrchestrator:
            "progress",
        )

-        # Create tasks for all enabled types
-        ideation_tasks = [
-            self.output_streamer.stream_ideation_result(
-                ideation_type, self.phase_executor, MAX_RETRIES
+        # Create tasks explicitly so we can cancel them on timeout
+        ideation_task_objs = [
+            asyncio.create_task(
+                self.output_streamer.stream_ideation_result(
+                    ideation_type, self.phase_executor, MAX_RETRIES
+                )
            )
            for ideation_type in self.enabled_types
        ]

-        # Run all ideation types concurrently
-        ideation_results = await asyncio.gather(*ideation_tasks, return_exceptions=True)
+        # Run all ideation types concurrently with timeout protection
+        # 5 minute timeout prevents infinite hangs if one type stalls
+        try:
+            ideation_results = await asyncio.wait_for(
+                asyncio.gather(*ideation_task_objs, return_exceptions=True),
+                timeout=IDEATION_TIMEOUT_SECONDS,
+            )
+        except asyncio.TimeoutError:
+            print_status(
+                "Ideation generation timed out after 5 minutes",
+                "error",
+            )
+            # Cancel all pending tasks to prevent resource leaks
+            for task in ideation_task_objs:
+                if not task.done():
+                    task.cancel()
+            # Wait for cancellation to complete and preserve results from completed tasks
+            # Tasks that finished before timeout will return their results;
+            # cancelled tasks will return CancelledError
+            results_after_cancel = await asyncio.gather(
+                *ideation_task_objs, return_exceptions=True
+            )
+            # Convert CancelledError to timeout exception, preserve completed results
+            ideation_results = [
+                Exception("Ideation timed out")
+                if isinstance(res, asyncio.CancelledError)
+                else res
+                for res in results_after_cancel
+            ]

        # Process results
        for i, result in enumerate(ideation_results):
@@ -12,12 +12,8 @@ The refactored code is now organized as:
 - graphiti/search.py - Semantic search logic
 - graphiti/schema.py - Graph schema definitions

-This facade ensures existing imports continue to work:
-    from graphiti_memory import GraphitiMemory, is_graphiti_enabled
-
-New code should prefer importing from the graphiti package:
-    from graphiti import GraphitiMemory
-    from graphiti.schema import GroupIdMode
+Import from this module:
+    from integrations.graphiti.memory import GraphitiMemory, is_graphiti_enabled, GroupIdMode

 For detailed documentation on the memory system architecture and usage,
 see graphiti/graphiti.py.
@@ -62,7 +62,7 @@ async def get_graph_hints(
    try:
        from pathlib import Path

-        from graphiti_memory import GraphitiMemory, GroupIdMode
+        from integrations.graphiti.memory import GraphitiMemory, GroupIdMode

        # Determine project directory from project_id or use current dir
        project_dir = Path.cwd()
@@ -17,7 +17,7 @@ from core.sentry import capture_exception
 logger = logging.getLogger(__name__)

 if TYPE_CHECKING:
-    from graphiti_memory import GraphitiMemory
+    from integrations.graphiti.memory import GraphitiMemory


 def is_graphiti_memory_enabled() -> bool:
@@ -60,7 +60,7 @@ async def get_graphiti_memory(
        return None

    try:
-        from graphiti_memory import GraphitiMemory, GroupIdMode
+        from integrations.graphiti.memory import GraphitiMemory, GroupIdMode

        if project_dir is None:
            project_dir = spec_dir.parent.parent
@@ -17,6 +17,7 @@ import logging
 from .ai_resolver import AIResolver
 from .auto_merger import AutoMerger, MergeContext
 from .file_merger import apply_ai_merge, extract_location_content
+from .progress import MergeProgressCallback, MergeProgressStage
 from .types import (
    ConflictRegion,
    ConflictSeverity,
@@ -60,6 +61,7 @@ class ConflictResolver:
        baseline_content: str,
        task_snapshots: list[TaskSnapshot],
        conflicts: list[ConflictRegion],
+        progress_callback: MergeProgressCallback | None = None,
    ) -> MergeResult:
        """
        Resolve conflicts using AutoMerger and AIResolver.
@@ -69,6 +71,8 @@ class ConflictResolver:
            baseline_content: Original file content
            task_snapshots: Snapshots from all tasks modifying this file
            conflicts: List of detected conflicts
+            progress_callback: Optional callback for emitting per-conflict
+                resolution progress with details about current file and conflict count

        Returns:
            MergeResult with resolution details
@@ -78,8 +82,23 @@ class ConflictResolver:
        remaining: list[ConflictRegion] = []
        ai_calls = 0
        tokens_used = 0
+        total_conflicts = len(conflicts)

-        for conflict in conflicts:
+        for idx, conflict in enumerate(conflicts):
+            if progress_callback:
+                # Emit per-conflict progress within the resolving stage (50-75%)
+                # Calculate progress after processing (idx + 1) to reach 75% on last conflict
+                conflict_percent = 50 + int(((idx + 1) / max(total_conflicts, 1)) * 25)
+                progress_callback(
+                    stage=MergeProgressStage.RESOLVING,
+                    percent=conflict_percent,
+                    message=f"Resolving conflict {idx + 1}/{total_conflicts} in {file_path}",
+                    details={
+                        "current_file": file_path,
+                        "conflicts_found": total_conflicts,
+                        "conflicts_resolved": len(resolved),
+                    },
+                )
            # Try auto-merge first
            if conflict.can_auto_merge and conflict.merge_strategy:
                context = MergeContext(
@@ -18,6 +18,7 @@ import logging
 from .conflict_detector import ConflictDetector
 from .conflict_resolver import ConflictResolver
 from .file_merger import apply_single_task_changes, combine_non_conflicting_changes
+from .progress import MergeProgressCallback, MergeProgressStage
 from .types import (
    ChangeType,
    FileAnalysis,
@@ -57,6 +58,7 @@ class MergePipeline:
        file_path: str,
        baseline_content: str,
        task_snapshots: list[TaskSnapshot],
+        progress_callback: MergeProgressCallback | None = None,
    ) -> MergeResult:
        """
        Merge changes from multiple tasks for a single file.
@@ -65,6 +67,8 @@ class MergePipeline:
            file_path: Path to the file
            baseline_content: Original baseline content
            task_snapshots: Snapshots from tasks that modified this file
+            progress_callback: Optional callback for emitting per-file progress
+                within the 'resolving' stage (50-75% range)

        Returns:
            MergeResult with merged content or conflict info
@@ -72,6 +76,14 @@ class MergePipeline:
        task_ids = [s.task_id for s in task_snapshots]
        logger.info(f"Merging {file_path} with {len(task_snapshots)} task(s)")

+        if progress_callback:
+            progress_callback(
+                stage=MergeProgressStage.RESOLVING,
+                percent=50,
+                message=f"Merging file: {file_path}",
+                details={"current_file": file_path},
+            )
+
        # If only one task modified the file, no conflict possible
        if len(task_snapshots) == 1:
            snapshot = task_snapshots[0]
@@ -119,6 +131,7 @@ class MergePipeline:
            baseline_content=baseline_content,
            task_snapshots=task_snapshots,
            conflicts=conflicts,
+            progress_callback=progress_callback,
        )

    def _build_task_analyses(
@@ -33,6 +33,7 @@ from .merge_pipeline import MergePipeline

 # Re-export models for backwards compatibility
 from .models import MergeReport, MergeStats, TaskMergeRequest
+from .progress import MergeProgressCallback, MergeProgressStage
 from .semantic_analyzer import SemanticAnalyzer
 from .types import (
    ConflictRegion,
@@ -260,6 +261,7 @@ class MergeOrchestrator:
        task_id: str,
        worktree_path: Path | None = None,
        target_branch: str = "main",
+        progress_callback: MergeProgressCallback | None = None,
    ) -> MergeReport:
        """
        Merge a single task's changes into the target branch.
@@ -268,6 +270,8 @@ class MergeOrchestrator:
            task_id: The task identifier
            worktree_path: Path to the task's worktree (auto-detected if not provided)
            target_branch: Branch to merge into
+            progress_callback: Optional callback for progress updates.
+                Called with (stage, percent, message, details) at key pipeline stages.

        Returns:
            MergeReport with results
@@ -284,7 +288,20 @@ class MergeOrchestrator:
        report = MergeReport(started_at=datetime.now(), tasks_merged=[task_id])
        start_time = datetime.now()

+        def _emit(
+            stage: MergeProgressStage,
+            percent: int,
+            message: str,
+            details: dict[str, Any] | None = None,
+        ) -> None:
+            """Emit progress if a callback is provided."""
+            if progress_callback is not None:
+                progress_callback(stage, percent, message, details)
+
        try:
+            # --- ANALYZING stage (0-25%) ---
+            _emit(MergeProgressStage.ANALYZING, 0, "Starting merge analysis")
+
            # Find worktree if not provided
            if worktree_path is None:
                debug_detailed(MODULE, "Auto-detecting worktree path...")
@@ -293,16 +310,23 @@ class MergeOrchestrator:
                    debug_error(MODULE, f"Could not find worktree for task {task_id}")
                    report.success = False
                    report.error = f"Could not find worktree for task {task_id}"
+                    _emit(
+                        MergeProgressStage.ERROR,
+                        0,
+                        f"Could not find worktree for task {task_id}",
+                    )
                    return report
                debug_detailed(MODULE, f"Found worktree: {worktree_path}")

            # Ensure evolution data is up to date
+            _emit(MergeProgressStage.ANALYZING, 5, "Loading file evolution data")
            debug(MODULE, "Refreshing evolution data from git...")
            self.evolution_tracker.refresh_from_git(
                task_id, worktree_path, target_branch=target_branch
            )

            # Get files modified by this task
+            _emit(MergeProgressStage.ANALYZING, 15, "Running semantic analysis")
            modifications = self.evolution_tracker.get_task_modifications(task_id)
            debug(
                MODULE,
@@ -312,11 +336,39 @@ class MergeOrchestrator:
            if not modifications:
                debug_warning(MODULE, f"No modifications found for task {task_id}")
                logger.info(f"No modifications found for task {task_id}")
+                _emit(
+                    MergeProgressStage.COMPLETE,
+                    100,
+                    "No modifications found",
+                )
                report.completed_at = datetime.now()
                return report

-            # Process each modified file
-            for file_path, snapshot in modifications:
+            _emit(
+                MergeProgressStage.ANALYZING,
+                25,
+                f"Found {len(modifications)} modified files",
+            )
+
+            # --- DETECTING_CONFLICTS stage (25-50%) ---
+            _emit(
+                MergeProgressStage.DETECTING_CONFLICTS,
+                25,
+                "Detecting conflicts",
+            )
+
+            # --- RESOLVING stage (50-75%) ---
+            total_files = len(modifications)
+            for idx, (file_path, snapshot) in enumerate(modifications):
+                # Calculate progress after processing (idx + 1) to reach 75% on last file
+                file_percent = 50 + int(((idx + 1) / max(total_files, 1)) * 25)
+                _emit(
+                    MergeProgressStage.RESOLVING,
+                    file_percent,
+                    f"Merging file {idx + 1}/{total_files}",
+                    {"current_file": file_path},
+                )
+
                debug_detailed(
                    MODULE,
                    f"Processing file: {file_path}",
@@ -349,13 +401,31 @@ class MergeOrchestrator:
                    file=file_path,
                )

+            # --- VALIDATING stage (75-100%) ---
+            _emit(
+                MergeProgressStage.VALIDATING,
+                75,
+                "Validating merge results",
+                {
+                    "conflicts_found": report.stats.conflicts_detected,
+                    "conflicts_resolved": report.stats.conflicts_auto_resolved,
+                },
+            )
+
            report.success = report.stats.files_failed == 0

+            _emit(
+                MergeProgressStage.VALIDATING,
+                90,
+                "Validation complete",
+            )
+
        except Exception as e:
            debug_error(MODULE, f"Merge failed for task {task_id}", error=str(e))
            logger.exception(f"Merge failed for task {task_id}")
            report.success = False
            report.error = str(e)
+            _emit(MergeProgressStage.ERROR, 0, f"Merge failed: {e}")

        report.completed_at = datetime.now()
        report.stats.duration_seconds = (
@@ -366,6 +436,18 @@ class MergeOrchestrator:
        if not self.dry_run:
            self._save_report(report, task_id)

+        # --- COMPLETE stage (100%) ---
+        if report.success:
+            _emit(
+                MergeProgressStage.COMPLETE,
+                100,
+                f"Merge complete for {task_id}",
+                {
+                    "conflicts_found": report.stats.conflicts_detected,
+                    "conflicts_resolved": report.stats.conflicts_auto_resolved,
+                },
+            )
+
        debug_success(
            MODULE,
            f"Merge complete for {task_id}",
@@ -382,6 +464,7 @@ class MergeOrchestrator:
        self,
        requests: list[TaskMergeRequest],
        target_branch: str = "main",
+        progress_callback: MergeProgressCallback | None = None,
    ) -> MergeReport:
        """
        Merge multiple tasks' changes.
@@ -392,6 +475,8 @@ class MergeOrchestrator:
        Args:
            requests: List of merge requests (one per task)
            target_branch: Branch to merge into
+            progress_callback: Optional callback for progress updates.
+                Called with (stage, percent, message, details) at key pipeline stages.

        Returns:
            MergeReport with combined results
@@ -402,11 +487,33 @@ class MergeOrchestrator:
        )
        start_time = datetime.now()

+        def _emit(
+            stage: MergeProgressStage,
+            percent: int,
+            message: str,
+            details: dict[str, Any] | None = None,
+        ) -> None:
+            """Emit progress if a callback is provided."""
+            if progress_callback is not None:
+                progress_callback(stage, percent, message, details)
+
        try:
+            # --- ANALYZING stage (0-25%) ---
+            _emit(
+                MergeProgressStage.ANALYZING,
+                0,
+                f"Starting merge analysis for {len(requests)} tasks",
+            )
+
            # Sort by priority (higher first)
            requests = sorted(requests, key=lambda r: -r.priority)

            # Refresh evolution data for all tasks
+            _emit(
+                MergeProgressStage.ANALYZING,
+                5,
+                "Loading file evolution data",
+            )
            for request in requests:
                if request.worktree_path and request.worktree_path.exists():
                    self.evolution_tracker.refresh_from_git(
@@ -416,11 +523,38 @@ class MergeOrchestrator:
                    )

            # Find all files modified by any task
+            _emit(
+                MergeProgressStage.ANALYZING,
+                15,
+                "Running semantic analysis",
+            )
            task_ids = [r.task_id for r in requests]
            file_tasks = self.evolution_tracker.get_files_modified_by_tasks(task_ids)

-            # Process each file
-            for file_path, modifying_tasks in file_tasks.items():
+            _emit(
+                MergeProgressStage.ANALYZING,
+                25,
+                f"Found {len(file_tasks)} files to merge",
+            )
+
+            # --- DETECTING_CONFLICTS stage (25-50%) ---
+            _emit(
+                MergeProgressStage.DETECTING_CONFLICTS,
+                25,
+                "Detecting conflicts across tasks",
+            )
+
+            # --- RESOLVING stage (50-75%) ---
+            total_files = len(file_tasks)
+            for idx, (file_path, modifying_tasks) in enumerate(file_tasks.items()):
+                file_percent = 50 + int((idx / max(total_files, 1)) * 25)
+                _emit(
+                    MergeProgressStage.RESOLVING,
+                    file_percent,
+                    f"Merging file {idx + 1}/{total_files}",
+                    {"current_file": file_path},
+                )
+
                # Get snapshots from all tasks that modified this file
                evolution = self.evolution_tracker.get_file_evolution(file_path)
                if not evolution:
@@ -466,8 +600,25 @@ class MergeOrchestrator:
                report.file_results[file_path] = result
                self._update_stats(report.stats, result)

+            # --- VALIDATING stage (75-100%) ---
+            _emit(
+                MergeProgressStage.VALIDATING,
+                75,
+                "Validating merge results",
+                {
+                    "conflicts_found": report.stats.conflicts_detected,
+                    "conflicts_resolved": report.stats.conflicts_auto_resolved,
+                },
+            )
+
            report.success = report.stats.files_failed == 0

+            _emit(
+                MergeProgressStage.VALIDATING,
+                90,
+                "Validation complete",
+            )
+
        except Exception as e:
            debug_error(
                MODULE,
@@ -478,6 +629,7 @@ class MergeOrchestrator:
            logger.exception("Merge failed")
            report.success = False
            report.error = str(e)
+            _emit(MergeProgressStage.ERROR, 0, f"Merge failed: {e}")

        report.completed_at = datetime.now()
        report.stats.duration_seconds = (
@@ -489,6 +641,18 @@ class MergeOrchestrator:
            timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
            self._save_report(report, f"multi_{timestamp}")

+        # --- COMPLETE stage (100%) ---
+        if report.success:
+            _emit(
+                MergeProgressStage.COMPLETE,
+                100,
+                f"Merge complete for {len(requests)} tasks",
+                {
+                    "conflicts_found": report.stats.conflicts_detected,
+                    "conflicts_resolved": report.stats.conflicts_auto_resolved,
+                },
+            )
+
        return report

    def _merge_file(
@@ -0,0 +1,105 @@
+"""
+Merge Progress Emission
+=======================
+
+Structured progress event emission for the merge pipeline.
+
+This module provides the progress reporting infrastructure used by the
+merge orchestrator to communicate real-time status updates to the
+Electron frontend via stdout JSON lines.
+
+Progress events are emitted as JSON lines to stdout with type='progress',
+allowing the frontend to parse them separately from the final merge result.
+
+Components:
+- MergeProgressStage: Enum of pipeline stages
+- MergeProgressCallback: Protocol for type-safe callback threading
+- emit_progress: Function to emit structured progress events to stdout
+"""
+
+from __future__ import annotations
+
+import json
+from enum import Enum
+from typing import Any, Protocol
+
+
+class MergeProgressStage(Enum):
+    """
+    Stages of the merge pipeline.
+
+    Each stage corresponds to a phase of the merge process and maps
+    to a percentage range for progress reporting:
+    - ANALYZING: 0-25% — Loading file evolution, running semantic analysis
+    - DETECTING_CONFLICTS: 25-50% — Conflict detection and compatibility checks
+    - RESOLVING: 50-75% — Auto-merge and AI resolution of conflicts
+    - VALIDATING: 75-100% — Final validation of merged results
+    - COMPLETE: 100% — Merge finished successfully
+    - ERROR: N/A — Merge failed with an error
+    """
+
+    ANALYZING = "analyzing"
+    DETECTING_CONFLICTS = "detecting_conflicts"
+    RESOLVING = "resolving"
+    VALIDATING = "validating"
+    COMPLETE = "complete"
+    ERROR = "error"
+
+
+class MergeProgressCallback(Protocol):
+    """
+    Protocol for type-safe progress callback threading.
+
+    Implementations receive structured progress updates from the merge
+    pipeline stages and can forward them to any output channel.
+
+    Args:
+        stage: Current pipeline stage
+        percent: Progress percentage (0-100)
+        message: Human-readable status message
+        details: Optional additional context (conflicts_found, conflicts_resolved, current_file)
+    """
+
+    def __call__(
+        self,
+        stage: MergeProgressStage,
+        percent: int,
+        message: str,
+        details: dict[str, Any] | None = None,
+    ) -> None: ...
+
+
+def emit_progress(
+    stage: MergeProgressStage,
+    percent: int,
+    message: str,
+    details: dict[str, Any] | None = None,
+) -> None:
+    """
+    Emit a progress event as a JSON line to stdout.
+
+    The Electron main process parses these JSON lines from the merge
+    subprocess stdout and forwards them to the renderer via IPC.
+
+    Args:
+        stage: Current pipeline stage
+        percent: Progress percentage (0-100), clamped to valid range
+        message: Human-readable status message
+        details: Optional dict with additional context. Supported keys:
+            - conflicts_found (int): Number of conflicts detected
+            - conflicts_resolved (int): Number of conflicts resolved so far
+            - current_file (str): File currently being processed
+    """
+    percent = max(0, min(100, percent))
+
+    event: dict[str, Any] = {
+        "type": "progress",
+        "stage": stage.value,
+        "percent": percent,
+        "message": message,
+    }
+
+    if details:
+        event["details"] = details
+
+    print(json.dumps(event), flush=True)
@@ -13,15 +13,58 @@ environment at the start of each prompt in the "YOUR ENVIRONMENT" section. Pay c

 - **Working Directory**: This is your root - all paths are relative to here
 - **Spec Location**: Where your spec files live (usually `./auto-claude/specs/{spec-name}/`)
+- **Isolation Mode**: If present, you are in an isolated worktree (see below)

 **RULES:**
 1. ALWAYS use relative paths starting with `./`
-2. NEVER use absolute paths (like `/Users/...`)
+2. NEVER use absolute paths (like `/Users/...` or `/e/projects/...`)
 3. NEVER assume paths exist - check with `ls` first
 4. If a file doesn't exist where expected, check the spec location from YOUR ENVIRONMENT section

 ---

+## ⛔ WORKTREE ISOLATION (When Applicable)
+
+If your environment shows **"Isolation Mode: WORKTREE"**, you are working in an **isolated git worktree**.
+This is a complete copy of the project created for safe, isolated development.
+
+### Critical Rules for Worktree Mode:
+
+1. **NEVER navigate to the parent project path** shown in "FORBIDDEN PATH"
+   - If you see `cd /path/to/main/project` in your context, DO NOT run it
+   - The parent project is OFF LIMITS
+
+2. **All files exist locally via relative paths**
+   - `./prod/...` ✅ CORRECT
+   - `/path/to/main/project/prod/...` ❌ WRONG (escapes isolation)
+
+3. **Git commits in the wrong location = disaster**
+   - Commits made after escaping go to the WRONG branch
+   - This defeats the entire isolation system
+
+### Why You Might Be Tempted to Escape:
+
+You may see absolute paths like `/e/projects/myapp/prod/src/file.ts` in:
+- `spec.md` (file references)
+- `context.json` (discovered files)
+- Error messages
+
+**DO NOT** `cd` to these paths. Instead, convert them to relative paths:
+- `/e/projects/myapp/prod/src/file.ts` → `./prod/src/file.ts`
+
+### Quick Check:
+
+```bash
+# Verify you're still in the worktree
+pwd
+# Should show: .../.auto-claude/worktrees/tasks/{spec-name}/
+# Or (legacy): .../.worktrees/{spec-name}/
+# Or (PR review): .../.auto-claude/github/pr/worktrees/{pr-number}/
+# NOT: /path/to/main/project
+```
+
+---
+
 ## 🚨 CRITICAL: PATH CONFUSION PREVENTION 🚨

 **THE #1 BUG IN MONOREPOS: Doubled paths after `cd` commands**
@@ -84,65 +127,6 @@ git add [verified-path]

 ---

-## 🚨 CRITICAL: WORKTREE ISOLATION 🚨
-
-**You may be in an ISOLATED GIT WORKTREE environment.**
-
-Check the "YOUR ENVIRONMENT" section at the top of this prompt. If you see an
-**"ISOLATED WORKTREE - CRITICAL"** section, you are in a worktree.
-
-### What is a Worktree?
-
-A worktree is a **complete copy of the project** isolated from the main project.
-This allows safe development without affecting the main branch.
-
-### Worktree Rules (CRITICAL)
-
-**If you are in a worktree, the environment section will show:**
-
-* **YOUR LOCATION:** The path to your isolated worktree
-* **FORBIDDEN:** The parent project path you must NEVER `cd` to
-
-**CRITICAL RULES:**
-* **NEVER** `cd` to the forbidden parent path
-* **NEVER** use `cd ../..` to escape the worktree
-* **STAY** within your working directory at all times
-* **ALL** file operations use paths relative to your current location
-
-### Why This Matters
-
-Escaping the worktree causes:
-* ❌ Git commits going to the wrong branch
-* ❌ Files created/modified in the wrong location
-* ❌ Breaking worktree isolation guarantees
-* ❌ Losing the safety of isolated development
-
-### How to Stay Safe
-
-**Before ANY `cd` command:**
-
-```bash
-# 1. Check where you are
-pwd
-
-# 2. Verify the target is within your worktree
-# If pwd shows: /path/to/.auto-claude/worktrees/tasks/spec-name/
-# Then: cd ./apps/backend  ✅ SAFE
-# But:  cd /path/to/parent/project  ❌ FORBIDDEN - ESCAPES ISOLATION
-
-# 3. When in doubt, don't use cd at all
-# Use relative paths from your current directory instead
-git add ./apps/backend/file.py  # Works from anywhere in worktree
-```
-
-### The Golden Rule in Worktrees
-
-**If you're in a worktree, pretend the parent project doesn't exist.**
-
-Everything you need is in your worktree, accessible via relative paths.
-
---
-
 ## STEP 1: GET YOUR BEARINGS (MANDATORY)

 First, check your environment. The prompt should tell you your working directory and spec location.
@@ -0,0 +1,192 @@
+# PR Review System Quality Control Prompt
+
+You are a senior software architect tasked with quality-controlling an AI-powered PR review system. Your goal is to analyze the system holistically, identify gaps between intent and implementation, and provide actionable feedback.
+
+## System Overview
+
+This is a **parallel orchestrator PR review system** that:
+1. An orchestrator AI analyzes a PR and delegates to specialist agents
+2. Specialist agents (security, quality, logic, codebase-fit) perform deep reviews
+3. A finding-validator agent validates all findings against actual code
+4. The orchestrator synthesizes results into a final verdict
+
+**Key Design Principles (from vision document):**
+- Evidence-based validation (NOT confidence-based)
+- Pattern-triggered mandatory exploration (6 semantic triggers)
+- Understand intent BEFORE looking for issues
+- The diff is the question, not the answer
+
+---
+
+## FILES TO EXAMINE
+
+### Vision & Architecture
+- `docs/PR_REVIEW_99_TRUST.md` - The vision document defining 99% trust goal
+
+### Orchestrator Prompts
+- `apps/backend/prompts/github/pr_parallel_orchestrator.md` - Main orchestrator prompt
+- `apps/backend/prompts/github/pr_followup_orchestrator.md` - Follow-up review orchestrator
+
+### Specialist Agent Prompts
+- `apps/backend/prompts/github/pr_security_agent.md` - Security review agent
+- `apps/backend/prompts/github/pr_quality_agent.md` - Code quality agent
+- `apps/backend/prompts/github/pr_logic_agent.md` - Logic/correctness agent
+- `apps/backend/prompts/github/pr_codebase_fit_agent.md` - Codebase fit agent
+- `apps/backend/prompts/github/pr_finding_validator.md` - Finding validator agent
+
+### Implementation Code
+- `apps/backend/runners/github/services/parallel_orchestrator_reviewer.py` - Orchestrator implementation
+- `apps/backend/runners/github/services/parallel_followup_reviewer.py` - Follow-up implementation
+- `apps/backend/runners/github/services/pydantic_models.py` - Schema definitions (VerificationEvidence, etc.)
+- `apps/backend/runners/github/services/sdk_utils.py` - SDK utilities for running agents
+- `apps/backend/runners/github/services/review_tools.py` - Tools available to review agents
+- `apps/backend/runners/github/context_gatherer.py` - Gathers PR context (files, callers, dependents)
+
+### Models & Configuration
+- `apps/backend/runners/github/models.py` - Data models
+- `apps/backend/agents/tools_pkg/models.py` - Tool models
+
+---
+
+## ANALYSIS TASKS
+
+### 1. Vision Alignment Check
+Compare the implementation against `PR_REVIEW_99_TRUST.md`:
+
+- [ ] **Evidence-based validation**: Is the system truly evidence-based or does it still use confidence scores anywhere?
+- [ ] **6 Mandatory Triggers**: Are all 6 semantic triggers properly defined and enforced?
+  1. Output contract changed
+  2. Input contract changed
+  3. Behavioral contract changed
+  4. Side effect contract changed
+  5. Failure contract changed
+  6. Null/undefined contract changed
+- [ ] **Phase 0 (Understand Intent)**: Is it mandatory? Is it enforced before delegation?
+- [ ] **Phase 1 (Trigger Detection)**: Is it mandatory? Does it output explicit trigger analysis?
+- [ ] **Bounded Exploration**: Is exploration limited to depth 1 (direct callers only)?
+
+### 2. Prompt Quality Analysis
+For each agent prompt, check:
+
+- [ ] Does it explain WHAT to look for?
+- [ ] Does it explain HOW to verify findings?
+- [ ] Does it require evidence (code snippets, line numbers)?
+- [ ] Does it define when to STOP exploring?
+- [ ] Does it distinguish between "in scope" and "out of scope"?
+- [ ] Does it handle the "no issues found" case properly?
+
+### 3. Schema Enforcement
+Check `pydantic_models.py`:
+
+- [ ] Is `VerificationEvidence` required (not optional) on all finding types?
+- [ ] Does `VerificationEvidence` require:
+  - `code_examined` (actual code, not description)
+  - `line_range_examined` (specific lines)
+  - `verification_method` (how it was verified)
+- [ ] Are there any finding types that bypass evidence requirements?
+
+### 4. Information Flow
+Trace how information flows:
+
+- [ ] PR Context → Orchestrator: What context is provided?
+- [ ] Orchestrator → Specialists: Are triggers passed? Are known callers passed?
+- [ ] Specialists → Validator: Are all findings validated?
+- [ ] Validator → Final Output: Are false positives properly dismissed?
+
+### 5. False Positive Prevention
+Check mechanisms to prevent false positives:
+
+- [ ] Do specialists verify issues exist before reporting?
+- [ ] Does the validator re-read the actual code?
+- [ ] Are "missing X" claims (missing error handling, etc.) verified?
+- [ ] Are dismissed findings tracked for transparency?
+
+### 6. Log Analysis (ATTACH LOGS BELOW)
+When reviewing logs, check:
+
+- [ ] Did the orchestrator output PR UNDERSTANDING before delegating?
+- [ ] Did the orchestrator output TRIGGER DETECTION before delegating?
+- [ ] Were triggers passed to specialists in delegation prompts?
+- [ ] Did specialists actually explore when triggers were present?
+- [ ] Were findings validated with real code evidence?
+- [ ] Were any false positives caught by the validator?
+
+---
+
+## SPECIFIC QUESTIONS TO ANSWER
+
+1. **Trigger System Effectiveness**: Did the trigger detection system correctly identify semantic contract changes? Were there any missed triggers or false triggers?
+
+2. **Exploration Quality**: When exploration was mandated by a trigger, did specialists explore effectively? Did they stop at the right time?
+
+3. **Evidence Quality**: Are the `code_examined` fields in findings actual code snippets or just descriptions? Are line numbers accurate?
+
+4. **False Positive Rate**: How many findings were dismissed as false positives? What caused them?
+
+5. **Missing Issues**: Based on your understanding of the PR, were there any issues that SHOULD have been caught but weren't?
+
+6. **Prompt Gaps**: Are there any scenarios not covered by the current prompts?
+
+7. **Schema Gaps**: Are there any ways findings could bypass evidence requirements?
+
+---
+
+## OUTPUT FORMAT
+
+Provide your analysis in this structure:
+
+```markdown
+## Executive Summary
+[2-3 sentences on overall system health]
+
+## Vision Alignment Score: X/10
+[Brief explanation]
+
+## Critical Issues (Must Fix)
+1. [Issue]: [Description] → [Suggested Fix]
+2. ...
+
+## High Priority Improvements
+1. [Improvement]: [Why it matters] → [How to implement]
+2. ...
+
+## Medium Priority Improvements
+1. ...
+
+## Low Priority / Nice to Have
+1. ...
+
+## Log Analysis Findings
+### What Worked Well
+- ...
+
+### What Didn't Work
+- ...
+
+### Specific Recommendations from Log Analysis
+1. ...
+
+## Questions for the Team
+1. [Question that needs human input]
+2. ...
+```
+
+---
+
+## ATTACH LOGS BELOW
+
+Paste the PR review debug logs here for analysis:
+
+```
+[PASTE LOGS HERE]
+```
+
+---
+
+## IMPORTANT NOTES
+
+- Focus on **systemic issues**, not one-off bugs
+- Prioritize issues that cause **false positives** (annoying) over false negatives (missed issues)
+- Consider **language-agnostic** design - the system should work for any codebase
+- Think about **edge cases**: empty PRs, huge PRs, refactor-only PRs, CSS-only PRs
+- The goal is **99% trust** - developers should trust the review enough to act on it immediately
@@ -6,6 +6,81 @@ You are a focused codebase fit review agent. You have been spawned by the orches

 Ensure new code integrates well with the existing codebase. Check for consistency with project conventions, reuse of existing utilities, and architectural alignment. Focus ONLY on codebase fit - not security, logic correctness, or general quality.

+## Phase 1: Understand the PR Intent (BEFORE Looking for Issues)
+
+**MANDATORY** - Before searching for issues, understand what this PR is trying to accomplish.
+
+1. **Read the provided context**
+   - PR description: What does the author say this does?
+   - Changed files: What areas of code are affected?
+   - Commits: How did the PR evolve?
+
+2. **Identify the change type**
+   - Bug fix: Correcting broken behavior
+   - New feature: Adding new capability
+   - Refactor: Restructuring without behavior change
+   - Performance: Optimizing existing code
+   - Cleanup: Removing dead code or improving organization
+
+3. **State your understanding** (include in your analysis)
+   ```
+   PR INTENT: This PR [verb] [what] by [how].
+   RISK AREAS: [what could go wrong specific to this change type]
+   ```
+
+**Only AFTER completing Phase 1, proceed to looking for issues.**
+
+Why this matters: Understanding intent prevents flagging intentional design decisions as bugs.
+
+## TRIGGER-DRIVEN EXPLORATION (CHECK YOUR DELEGATION PROMPT)
+
+**FIRST**: Check if your delegation prompt contains a `TRIGGER:` instruction.
+
+- **If TRIGGER is present** → Exploration is **MANDATORY**, even if the diff looks correct
+- **If no TRIGGER** → Use your judgment to explore or not
+
+### How to Explore (Bounded)
+
+1. **Read the trigger** - What pattern did the orchestrator identify?
+2. **Form the specific question** - "Do similar functions elsewhere follow the same pattern?" (not "what's in the codebase?")
+3. **Use Grep** to find similar patterns, usages, or implementations
+4. **Use Read** to examine 3-5 relevant files
+5. **Answer the question** - Yes (report issue) or No (move on)
+6. **Stop** - Do not explore beyond the immediate question
+
+### Codebase-Fit-Specific Trigger Questions
+
+| Trigger | Codebase Fit Question to Answer |
+|---------|--------------------------------|
+| **Output contract changed** | Do other similar functions return the same type/structure? |
+| **Input contract changed** | Is this parameter change consistent with similar functions? |
+| **New pattern introduced** | Does this pattern already exist elsewhere that should be reused? |
+| **Naming changed** | Is the new naming consistent with project conventions? |
+| **Architecture changed** | Does this architectural change align with existing patterns? |
+
+### Example Exploration
+
+```
+TRIGGER: New pattern introduced (custom date formatter)
+QUESTION: Does a date formatting utility already exist?
+
+1. Grep for "formatDate\|dateFormat\|toDateString" → found utils/date.ts
+2. Read utils/date.ts → exports formatDate(date, format) with same functionality
+3. STOP - Found existing utility
+
+FINDINGS:
+- src/components/Report.tsx:45 - Implements custom date formatting
+  Existing utility: utils/date.ts exports formatDate() with same functionality
+  Suggestion: Use existing formatDate() instead of duplicating logic
+```
+
+### When NO Trigger is Given
+
+If the orchestrator doesn't specify a trigger, use your judgment:
+- Focus on pattern consistency in the changed code
+- Search for existing utilities that could be reused
+- Don't explore "just to be thorough"
+
 ## CRITICAL: PR Scope and Context

 ### What IS in scope (report these issues):
@@ -119,6 +194,92 @@ Before reporting ANY finding, you MUST:

 **Your evidence must prove the issue exists - not just that you suspect it.**

+## Evidence Requirements (MANDATORY)
+
+Every finding you report MUST include a `verification` object with ALL of these fields:
+
+### Required Fields
+
+**code_examined** (string, min 1 character)
+The **exact code snippet** you examined. Copy-paste directly from the file:
+```
+CORRECT: "cursor.execute(f'SELECT * FROM users WHERE id={user_id}')"
+WRONG:   "SQL query that uses string interpolation"
+```
+
+**line_range_examined** (array of 2 integers)
+The exact line numbers [start, end] where the issue exists:
+```
+CORRECT: [45, 47]
+WRONG:   [1, 100]  // Too broad - you didn't examine all 100 lines
+```
+
+**verification_method** (one of these exact values)
+How you verified the issue:
+- `"direct_code_inspection"` - Found the issue directly in the code at the location
+- `"cross_file_trace"` - Traced through imports/calls to confirm the issue
+- `"test_verification"` - Verified through examination of test code
+- `"dependency_analysis"` - Verified through analyzing dependencies
+
+### Conditional Fields
+
+**is_impact_finding** (boolean, default false)
+Set to `true` ONLY if this finding is about impact on OTHER files (not the changed file):
+```
+TRUE:  "This change in utils.ts breaks the caller in auth.ts"
+FALSE: "This code in utils.ts has a bug" (issue is in the changed file)
+```
+
+**checked_for_handling_elsewhere** (boolean, default false)
+For ANY claim about existing utilities or patterns:
+- Set `true` ONLY if you used Grep/Read tools to verify patterns exist/don't exist
+- Set `false` if you didn't search the codebase
+- **When true, include the search in your description:**
+  - "Searched `Grep('formatDate|dateFormat', 'src/utils/')` - found existing helper"
+  - "Searched `Grep('class.*Service', 'src/services/')` - confirmed naming pattern"
+
+```
+TRUE:  "Searched for date formatting helpers - found utils/date.ts:formatDate()"
+FALSE: "This should use an existing utility" (didn't verify one exists)
+```
+
+**If you cannot provide real evidence, you do not have a verified finding - do not report it.**
+
+**Search Before Claiming:** Never claim something "should use existing X" without first verifying X exists and fits the use case.
+
+## Valid Outputs
+
+Finding issues is NOT the goal. Accurate review is the goal.
+
+### Valid: No Significant Issues Found
+If the code is well-implemented, say so:
+```json
+{
+  "findings": [],
+  "summary": "Reviewed [files]. No codebase_fit issues found. The implementation correctly [positive observation about the code]."
+}
+```
+
+### Valid: Only Low-Severity Suggestions
+Minor improvements that don't block merge:
+```json
+{
+  "findings": [
+    {"severity": "low", "title": "Consider extracting magic number to constant", ...}
+  ],
+  "summary": "Code is sound. One minor suggestion for readability."
+}
+```
+
+### INVALID: Forced Issues
+Do NOT report issues just to have something to say:
+- Theoretical edge cases without evidence they're reachable
+- Style preferences not backed by project conventions
+- "Could be improved" without concrete problem
+- Pre-existing issues not introduced by this PR
+
+**Reporting nothing is better than reporting noise.** False positives erode trust faster than false negatives.
+
 ## Code Patterns to Flag

 ### Reinventing Existing Utilities
@@ -189,6 +350,13 @@ Provide findings in JSON format:
    "description": "This file implements custom date formatting, but the codebase already has `formatDate()` in `src/utils/date.ts` that does the same thing.",
    "category": "codebase_fit",
    "severity": "high",
+    "verification": {
+      "code_examined": "const formatted = `${date.getMonth()}/${date.getDate()}/${date.getFullYear()}`;",
+      "line_range_examined": [15, 15],
+      "verification_method": "cross_file_trace"
+    },
+    "is_impact_finding": false,
+    "checked_for_handling_elsewhere": false,
    "existing_code": "src/utils/date.ts:formatDate()",
    "suggested_fix": "Replace custom implementation with: import { formatDate } from '@/utils/date';",
    "confidence": 92
@@ -200,6 +368,13 @@ Provide findings in JSON format:
    "description": "This file uses 'customer' terminology but the rest of the codebase consistently uses 'user'. This creates confusion and makes search/navigation harder.",
    "category": "codebase_fit",
    "severity": "medium",
+    "verification": {
+      "code_examined": "export interface Customer { id: string; name: string; email: string; }",
+      "line_range_examined": [1, 5],
+      "verification_method": "direct_code_inspection"
+    },
+    "is_impact_finding": false,
+    "checked_for_handling_elsewhere": false,
    "codebase_pattern": "src/models/user.ts, src/api/users.ts, src/services/userService.ts",
    "suggested_fix": "Rename to use 'user' terminology to match codebase conventions",
    "confidence": 88
@@ -211,6 +386,13 @@ Provide findings in JSON format:
    "description": "This file is 847 lines and contains order validation, payment processing, inventory management, and notification sending. Each should be separate.",
    "category": "codebase_fit",
    "severity": "high",
+    "verification": {
+      "code_examined": "// File contains: validateOrder(), processPayment(), updateInventory(), sendNotification() - all in one file",
+      "line_range_examined": [1, 847],
+      "verification_method": "direct_code_inspection"
+    },
+    "is_impact_finding": false,
+    "checked_for_handling_elsewhere": false,
    "current_lines": 847,
    "suggested_fix": "Split into: orderValidator.ts, paymentProcessor.ts, inventoryManager.ts, notificationService.ts",
    "confidence": 95
@@ -33,6 +33,165 @@ For each finding you receive:
 4. **PROVIDE** concrete code evidence - the actual code that proves or disproves the issue
 5. **RETURN** validation status with evidence (binary decision based on what the code shows)

+## Batch Processing (Multiple Findings)
+
+You may receive multiple findings to validate at once. When processing batches:
+
+1. **Group by file** - Read each file once, validate all findings in that file together
+2. **Process systematically** - Validate each finding in order, don't skip any
+3. **Return all results** - Your response must include a validation result for EVERY finding received
+4. **Optimize reads** - If 3 findings are in the same file, read it once with enough context for all
+
+**Example batch input:**
+```
+Validate these findings:
+1. SEC-001: SQL injection at auth/login.ts:45
+2. QUAL-001: Missing error handling at auth/login.ts:78
+3. LOGIC-001: Off-by-one at utils/array.ts:23
+```
+
+**Expected output:** 3 separate validation results, one for each finding ID.
+
+## Hypothesis-Validation Structure (MANDATORY)
+
+For EACH finding you investigate, use this structured approach. This prevents rubber-stamping findings as valid without actually verifying them.
+
+### Step 1: State the Hypothesis
+
+Before reading any code, clearly state what you're testing:
+
+```
+HYPOTHESIS: The finding claims "{title}" at {file}:{line}
+
+This hypothesis is TRUE if:
+1. The code at {line} contains the specific pattern described
+2. No mitigation exists in surrounding context (+/- 20 lines)
+3. The issue is actually reachable/exploitable in this codebase
+
+This hypothesis is FALSE if:
+1. The code at {line} is different than described
+2. Mitigation exists (validation, sanitization, framework protection)
+3. The code is unreachable or purely theoretical
+```
+
+### Step 2: Gather Evidence
+
+Read the actual code. Copy-paste it into `code_evidence`.
+
+```
+FILE: {file}
+LINES: {line-20} to {line+20}
+ACTUAL CODE:
+[paste the code here - this is your proof]
+```
+
+### Step 3: Test Each Condition
+
+For each condition in your hypothesis:
+
+```
+CONDITION 1: Code contains {specific pattern from finding}
+EVIDENCE: [specific line from code_evidence that proves/disproves]
+RESULT: TRUE / FALSE / INCONCLUSIVE
+
+CONDITION 2: No mitigation in surrounding context
+EVIDENCE: [what you found or didn't find in ±20 lines]
+RESULT: TRUE / FALSE / INCONCLUSIVE
+
+CONDITION 3: Issue is reachable/exploitable
+EVIDENCE: [how input reaches this code, or why it doesn't]
+RESULT: TRUE / FALSE / INCONCLUSIVE
+```
+
+### Step 4: Conclude Based on Evidence
+
+Apply these rules strictly:
+
+| Conditions | Conclusion |
+|------------|------------|
+| ALL conditions TRUE | `confirmed_valid` |
+| ANY condition FALSE | `dismissed_false_positive` |
+| ANY condition INCONCLUSIVE, none FALSE | `needs_human_review` |
+
+**CRITICAL: Your conclusion MUST match your condition results.** If you found mitigation (Condition 2 = FALSE), you MUST conclude `dismissed_false_positive`, not `confirmed_valid`.
+
+### Worked Example
+
+```
+HYPOTHESIS: SQL injection at auth.py:45
+
+Conditions to test:
+1. User input directly in SQL string (not parameterized)
+2. No sanitization before this point
+3. Input reachable from HTTP request
+
+Evidence gathered:
+FILE: auth.py, lines 25-65
+ACTUAL CODE:
+```python
+def get_user(user_id: str) -> User:
+    # user_id comes from request.args["id"]
+    query = f"SELECT * FROM users WHERE id = {user_id}"  # Line 45
+    return db.execute(query).fetchone()
+```
+
+Testing conditions:
+CONDITION 1: User input in SQL string
+EVIDENCE: Line 45 uses f-string interpolation: f"SELECT * FROM users WHERE id = {user_id}"
+RESULT: TRUE
+
+CONDITION 2: No sanitization
+EVIDENCE: No validation between request.args["id"] (line 43) and query construction (line 45)
+RESULT: TRUE
+
+CONDITION 3: Input reachable
+EVIDENCE: Comment says "user_id comes from request.args", confirmed by caller on line 12
+RESULT: TRUE
+
+CONCLUSION: confirmed_valid (all conditions TRUE)
+CODE_EVIDENCE: "query = f\"SELECT * FROM users WHERE id = {user_id}\""
+LINE_RANGE: [45, 45]
+EXPLANATION: SQL injection confirmed - user input from request.args is interpolated directly into SQL query without parameterization or sanitization.
+```
+
+### Counter-Example: Dismissing a False Positive
+
+```
+HYPOTHESIS: XSS vulnerability at render.py:89
+
+Conditions to test:
+1. User input reaches output without encoding
+2. No sanitization in the call chain
+3. Output context allows script execution
+
+Evidence gathered:
+FILE: render.py, lines 70-110
+ACTUAL CODE:
+```python
+def render_comment(user_input: str) -> str:
+    sanitized = bleach.clean(user_input, tags=[], strip=True)  # Line 85
+    return f"<div class='comment'>{sanitized}</div>"  # Line 89
+```
+
+Testing conditions:
+CONDITION 1: User input reaches output
+EVIDENCE: Line 89 outputs user_input into HTML
+RESULT: TRUE
+
+CONDITION 2: No sanitization
+EVIDENCE: Line 85 uses bleach.clean() with tags=[] (strips ALL tags)
+RESULT: FALSE - sanitization exists
+
+CONDITION 3: Output allows scripts
+EVIDENCE: Even if injected, bleach.clean removes script tags
+RESULT: FALSE - mitigation prevents exploitation
+
+CONCLUSION: dismissed_false_positive (Condition 2 and 3 are FALSE)
+CODE_EVIDENCE: "sanitized = bleach.clean(user_input, tags=[], strip=True)"
+LINE_RANGE: [85, 89]
+EXPLANATION: The original finding missed the sanitization at line 85. bleach.clean() with tags=[] strips all HTML tags including script tags, making XSS impossible.
+```
+
 ## Investigation Process

 ### Step 1: Fetch the Code
@@ -47,6 +206,8 @@ Focus on lines around: {finding.line}

 ### Step 2: Analyze with Fresh Eyes - NEVER ASSUME

+**Follow the Hypothesis-Validation Structure above for each finding.** State your hypothesis, gather evidence, test each condition, then conclude based on the evidence. This structure prevents you from confirming findings just because they "sound plausible."
+
 **CRITICAL: Do NOT assume the original finding is correct.** The original reviewer may have:
 - Hallucinated line numbers that don't exist
 - Misread or misunderstood the code
@@ -194,6 +355,45 @@ These patterns often confirm the issue is real:
 4. **Missing error handling** in critical paths
 5. **Race conditions** with clear concurrent access

+## Cross-File Validation (For Specific Finding Types)
+
+Some findings require checking the CODEBASE, not just the flagged file:
+
+### Duplication Findings ("code is duplicated 3 times")
+
+**Before confirming a duplication finding, you MUST:**
+
+1. **Verify the duplicated code exists** - Read all locations mentioned
+2. **Check for existing helpers** - Use Grep to search for:
+   - Similar function names in `/utils/`, `/helpers/`, `/shared/`
+   - Common patterns that might already be abstracted
+   - Example: `Grep("formatDate|dateFormat|toDateString", "**/*.{ts,js}")`
+
+3. **Decide based on evidence:**
+   - If existing helper found → `dismissed_false_positive` (they should use it)
+   - Wait, no - if helper exists and they're NOT using it → `confirmed_valid` (finding is correct)
+   - If no helper exists → `confirmed_valid` (suggest creating one)
+
+**Example:**
+```
+Finding: "Duplicated YOLO mode check repeated 3 times"
+
+CROSS-FILE CHECK:
+1. Grep for "YOLO_MODE|yoloMode|bypassSecurity" in utils/ → No results
+2. Grep for existing env var pattern helpers → Found: utils/env.ts:getEnvFlag()
+3. CONCLUSION: confirmed_valid - getEnvFlag() exists but isn't being used
+   SUGGESTED_FIX: "Use existing getEnvFlag() helper from utils/env.ts"
+```
+
+### "Should Use Existing X" Findings
+
+**Before confirming, verify the existing X actually fits the use case:**
+
+1. Read the suggested existing code
+2. Check if it has the required interface/behavior
+3. If it doesn't match → `dismissed_false_positive` (can't use it)
+4. If it matches → `confirmed_valid` (should use it)
+
 ## Critical Rules

 1. **ALWAYS read the actual code** - Never rely on memory or the original finding description
@@ -204,6 +404,10 @@ These patterns often confirm the issue is real:
 6. **Look for mitigations** - Check surrounding code for sanitization/validation
 7. **Check the full context** - Read ±20 lines, not just the flagged line
 8. **Verify code exists** - Set `evidence_verified_in_file` to false if the code/line doesn't exist
+9. **SEARCH BEFORE CLAIMING ABSENCE** - If you claim something doesn't exist (no helper, no validation, no error handling), you MUST show the search you performed:
+   - Use Grep to search for the pattern
+   - Include the search command in your explanation
+   - Example: "Searched for `Grep('validateInput|sanitize', 'src/**/*.ts')` - no results found"

 ## Anti-Patterns to Avoid

@@ -45,13 +45,77 @@ Note: GitHub's API tells us IF there are conflicts but not WHICH files. The find

 ## Available Specialist Agents

-You have access to these specialist agents via the Task tool:
+You have access to these specialist agents via the Task tool.
+
+**You MUST use the Task tool with the exact `subagent_type` names listed below.** Do NOT use `general-purpose` or any other built-in agent - always use our custom specialists.
+
+### Exact Agent Names (use these in subagent_type)
+
+| Agent | subagent_type value |
+|-------|---------------------|
+| Resolution verifier | `resolution-verifier` |
+| New code reviewer | `new-code-reviewer` |
+| Comment analyzer | `comment-analyzer` |
+| Finding validator | `finding-validator` |
+
+### Task Tool Invocation Format
+
+When you invoke a specialist, use the Task tool like this:
+
+```
+Task(
+  subagent_type="resolution-verifier",
+  prompt="Verify resolution of these previous findings:\n\n1. [SEC-001] SQL injection in user.ts:45 - Check if parameterized queries now used\n2. [QUAL-002] Missing error handling in api.ts:89 - Check if try/catch was added",
+  description="Verify previous findings resolved"
+)
+```
+
+### Example: Complete Follow-up Review Workflow
+
+**Step 1: Verify previous findings are resolved**
+```
+Task(
+  subagent_type="resolution-verifier",
+  prompt="Previous findings to verify:\n\n1. [HIGH] is_impact_finding not propagated (parallel_orchestrator_reviewer.py:630)\n   - Original issue: Field not extracted from structured output\n   - Expected fix: Add is_impact_finding extraction and pass to PRReviewFinding\n\nCheck if the new commits resolve this issue. Examine the actual code.",
+  description="Verify previous findings"
+)
+```
+
+**Step 2: Validate unresolved findings (MANDATORY)**
+```
+Task(
+  subagent_type="finding-validator",
+  prompt="Validate these unresolved findings from resolution-verifier:\n\n1. [HIGH] is_impact_finding not propagated (parallel_orchestrator_reviewer.py:630)\n   - Status from resolution-verifier: unresolved\n   - Claimed issue: Field not extracted\n\nRead the ACTUAL code at line 630 and verify if this issue truly exists. Check for is_impact_finding extraction.",
+  description="Validate unresolved findings"
+)
+```
+
+**Step 3: Review new code (if substantial changes)**
+```
+Task(
+  subagent_type="new-code-reviewer",
+  prompt="Review new code in this diff for issues:\n- Security vulnerabilities\n- Logic errors\n- Edge cases not handled\n\nFocus on files: models.py, parallel_orchestrator_reviewer.py",
+  description="Review new code changes"
+)
+```
+
+### DO NOT USE
+
+- ❌ `general-purpose` - This is a generic built-in agent, NOT our specialist
+- ❌ `Explore` - This is for codebase exploration, NOT for PR review
+- ❌ `Plan` - This is for planning, NOT for PR review
+
+**Always use our specialist agents** (`resolution-verifier`, `new-code-reviewer`, `comment-analyzer`, `finding-validator`) for follow-up review tasks.
+
+---
+
+## Agent Descriptions

 ### 1. resolution-verifier
 **Use for**: Verifying whether previous findings have been addressed
 - Analyzes diffs to determine if issues are truly fixed
 - Checks for incomplete or incorrect fixes
- Provides confidence scores for each resolution
+- Provides evidence-based verification for each resolution
 - **Invoke when**: There are previous findings to verify

 ### 2. new-code-reviewer
@@ -93,34 +157,85 @@ Evaluate the follow-up context:
 - Are there previous findings to verify?
 - Are there new comments to process?

-### Phase 2: Delegate to Agents
-Based on your analysis, invoke the appropriate agents:
+### Phase 2: Delegate to Agents (USE TASK TOOL)

-**Always invoke** `resolution-verifier` if there are previous findings.
+**You MUST use the Task tool to invoke agents.** Simply saying "invoke resolution-verifier" does nothing - you must call the Task tool.

-**ALWAYS invoke** `finding-validator` for ALL unresolved findings from resolution-verifier.
-This is CRITICAL to prevent false positives from persisting.
+**If there are previous findings, invoke resolution-verifier FIRST:**

-**Invoke** `new-code-reviewer` if:
- Diff is substantial (>50 lines)
- Changes touch security-sensitive areas
- New files were added
- Complex logic was modified
+```
+Task(
+  subagent_type="resolution-verifier",
+  prompt="Verify resolution of these previous findings:\n\n[COPY THE PREVIOUS FINDINGS LIST HERE WITH IDs, FILES, LINES, AND DESCRIPTIONS]",
+  description="Verify previous findings resolved"
+)
+```

-**Invoke** `comment-analyzer` if:
- There are contributor comments since last review
- There are AI tool reviews to triage
- Questions remain unanswered
+**THEN invoke finding-validator for ALL unresolved findings:**
+
+```
+Task(
+  subagent_type="finding-validator",
+  prompt="Validate these unresolved findings:\n\n[COPY THE UNRESOLVED FINDINGS FROM RESOLUTION-VERIFIER]",
+  description="Validate unresolved findings"
+)
+```
+
+**Invoke new-code-reviewer if substantial changes:**
+
+```
+Task(
+  subagent_type="new-code-reviewer",
+  prompt="Review new code changes:\n\n[INCLUDE FILE LIST AND KEY CHANGES]",
+  description="Review new code"
+)
+```
+
+**Invoke comment-analyzer if there are comments:**
+
+```
+Task(
+  subagent_type="comment-analyzer",
+  prompt="Analyze these comments:\n\n[INCLUDE COMMENT LIST]",
+  description="Analyze comments"
+)
+```
+
+### Decision Matrix
+
+| Condition | Agent to Invoke |
+|-----------|-----------------|
+| Previous findings exist | `resolution-verifier` (ALWAYS) |
+| Unresolved findings exist | `finding-validator` (ALWAYS - MANDATORY) |
+| Diff > 50 lines | `new-code-reviewer` |
+| New comments exist | `comment-analyzer` |
+
+### Phase 3: Validate ALL Findings (MANDATORY)
+
+**⚠️ ABSOLUTE RULE: You MUST invoke finding-validator for EVERY finding, regardless of severity.**
+This includes unresolved findings from resolution-verifier AND any new findings from new-code-reviewer.
+- CRITICAL/HIGH/MEDIUM/LOW: ALL must be validated
+- There are NO exceptions — every finding the user sees must be independently verified
+
+After resolution-verifier and new-code-reviewer return their findings:
+1. **Batch findings for validation:**
+   - For ≤10 findings: Send all to finding-validator in one call
+   - For >10 findings: Group by file or category, invoke 2-4 validator calls in parallel
+   - This reduces overhead while maintaining thorough validation

-### Phase 3: Validate Unresolved Findings
-After resolution-verifier returns findings marked as unresolved:
-1. Pass ALL unresolved findings to finding-validator
 2. finding-validator will read the actual code at each location
 3. For each finding, it returns:
-   - `confirmed_valid`: Issue IS real → keep as unresolved
+   - `confirmed_valid`: Issue IS real → keep as finding
   - `dismissed_false_positive`: Original finding was WRONG → remove from findings
   - `needs_human_review`: Cannot determine → flag for human

+**Every finding in the final output MUST have:**
+- `validation_status`: One of "confirmed_valid" or "needs_human_review"
+- `validation_evidence`: The actual code snippet examined during validation
+- `validation_explanation`: Why the finding was confirmed or flagged
+
+**If any finding is missing validation_status in the final output, the review is INVALID.**
+
 ### Phase 4: Synthesize Results
 After all agents complete:
 1. Combine resolution verifications
@@ -177,9 +292,10 @@ After all agents complete:
 ## Cross-Validation

 When multiple agents report on the same area:
- **Agreement boosts confidence**: If resolution-verifier and new-code-reviewer both flag an issue, increase severity
+- **Agreement strengthens evidence**: If resolution-verifier and new-code-reviewer both flag an issue, this is strong signal
 - **Conflicts need resolution**: If agents disagree, investigate and document your reasoning
 - **Track consensus**: Note which findings have cross-agent validation
+- **Evidence-based, not confidence-based**: Multiple agents agreeing doesn't skip validation - all findings still verified

 ## Output Format

@@ -198,16 +314,14 @@ Provide your synthesis as a structured response matching the ParallelFollowupRes
      "validation_status": "confirmed_valid",
      "code_evidence": "const query = `SELECT * FROM users WHERE id = ${userId}`;",
      "line_range": [45, 45],
-      "explanation": "SQL injection is present - user input is concatenated...",
-      "confidence": 0.92
+      "explanation": "SQL injection is present - user input is concatenated directly into query"
    },
    {
      "finding_id": "QUAL-002",
      "validation_status": "dismissed_false_positive",
      "code_evidence": "const sanitized = DOMPurify.sanitize(data);",
      "line_range": [23, 26],
-      "explanation": "Original finding claimed XSS but code uses DOMPurify...",
-      "confidence": 0.88
+      "explanation": "Original finding claimed XSS but code uses DOMPurify for sanitization"
    }
  ],
  "new_findings": [...],
@@ -6,6 +6,83 @@ You are a focused logic and correctness review agent. You have been spawned by t

 Verify that the code logic is correct, handles all edge cases, and doesn't introduce subtle bugs. Focus ONLY on logic and correctness issues - not style, security, or general quality.

+## Phase 1: Understand the PR Intent (BEFORE Looking for Issues)
+
+**MANDATORY** - Before searching for issues, understand what this PR is trying to accomplish.
+
+1. **Read the provided context**
+   - PR description: What does the author say this does?
+   - Changed files: What areas of code are affected?
+   - Commits: How did the PR evolve?
+
+2. **Identify the change type**
+   - Bug fix: Correcting broken behavior
+   - New feature: Adding new capability
+   - Refactor: Restructuring without behavior change
+   - Performance: Optimizing existing code
+   - Cleanup: Removing dead code or improving organization
+
+3. **State your understanding** (include in your analysis)
+   ```
+   PR INTENT: This PR [verb] [what] by [how].
+   RISK AREAS: [what could go wrong specific to this change type]
+   ```
+
+**Only AFTER completing Phase 1, proceed to looking for issues.**
+
+Why this matters: Understanding intent prevents flagging intentional design decisions as bugs.
+
+## TRIGGER-DRIVEN EXPLORATION (CHECK YOUR DELEGATION PROMPT)
+
+**FIRST**: Check if your delegation prompt contains a `TRIGGER:` instruction.
+
+- **If TRIGGER is present** → Exploration is **MANDATORY**, even if the diff looks correct
+- **If no TRIGGER** → Use your judgment to explore or not
+
+### How to Explore (Bounded)
+
+1. **Read the trigger** - What pattern did the orchestrator identify?
+2. **Form the specific question** - "Do callers handle the new return type?" (not "what do callers do?")
+3. **Use Grep** to find call sites of the changed function/method
+4. **Use Read** to examine 3-5 callers
+5. **Answer the question** - Yes (report issue) or No (move on)
+6. **Stop** - Do not explore callers of callers (depth > 1)
+
+### Trigger-Specific Questions
+
+| Trigger | What to Check in Callers |
+|---------|-------------------------|
+| **Output contract changed** | Do callers assume the old return type/structure? |
+| **Input contract changed** | Do callers pass the old arguments/defaults? |
+| **Behavioral contract changed** | Does code after the call assume old ordering/timing? |
+| **Side effect removed** | Did callers depend on the removed effect? |
+| **Failure contract changed** | Can callers handle the new failure mode? |
+| **Null contract changed** | Do callers have explicit null checks or tri-state logic? |
+
+### Example Exploration
+
+```
+TRIGGER: Output contract changed (array → single object)
+QUESTION: Do callers use array methods?
+
+1. Grep for "getUserSettings(" → found 8 call sites
+2. Read dashboard.tsx:45 → uses .find() on result → ISSUE
+3. Read profile.tsx:23 → uses result.email directly → OK
+4. Read settings.tsx:67 → uses .map() on result → ISSUE
+5. STOP - Found 2 confirmed issues, pattern established
+
+FINDINGS:
+- dashboard.tsx:45 - uses .find() which doesn't exist on object
+- settings.tsx:67 - uses .map() which doesn't exist on object
+```
+
+### When NO Trigger is Given
+
+If the orchestrator doesn't specify a trigger, use your judgment:
+- Focus on the changed code first
+- Only explore callers if you suspect an issue from the diff
+- Don't explore "just to be thorough"
+
 ## CRITICAL: PR Scope and Context

 ### What IS in scope (report these issues):
@@ -140,6 +217,92 @@ Before reporting ANY finding, you MUST:

 **Your evidence must prove the issue exists - not just that you suspect it.**

+## Evidence Requirements (MANDATORY)
+
+Every finding you report MUST include a `verification` object with ALL of these fields:
+
+### Required Fields
+
+**code_examined** (string, min 1 character)
+The **exact code snippet** you examined. Copy-paste directly from the file:
+```
+CORRECT: "cursor.execute(f'SELECT * FROM users WHERE id={user_id}')"
+WRONG:   "SQL query that uses string interpolation"
+```
+
+**line_range_examined** (array of 2 integers)
+The exact line numbers [start, end] where the issue exists:
+```
+CORRECT: [45, 47]
+WRONG:   [1, 100]  // Too broad - you didn't examine all 100 lines
+```
+
+**verification_method** (one of these exact values)
+How you verified the issue:
+- `"direct_code_inspection"` - Found the issue directly in the code at the location
+- `"cross_file_trace"` - Traced through imports/calls to confirm the issue
+- `"test_verification"` - Verified through examination of test code
+- `"dependency_analysis"` - Verified through analyzing dependencies
+
+### Conditional Fields
+
+**is_impact_finding** (boolean, default false)
+Set to `true` ONLY if this finding is about impact on OTHER files (not the changed file):
+```
+TRUE:  "This change in utils.ts breaks the caller in auth.ts"
+FALSE: "This code in utils.ts has a bug" (issue is in the changed file)
+```
+
+**checked_for_handling_elsewhere** (boolean, default false)
+For ANY "missing X" claim (missing null check, missing bounds check, missing edge case handling):
+- Set `true` ONLY if you used Grep/Read tools to verify X is not handled elsewhere
+- Set `false` if you didn't search other files
+- **When true, include the search in your description:**
+  - "Searched `Grep('if.*null|!= null|\?\?', 'src/utils/')` - no null check found"
+  - "Checked callers via `Grep('processArray\(', '**/*.ts')` - none validate input"
+
+```
+TRUE:  "Searched for null checks in this file and callers - none found"
+FALSE: "This function should check for null" (didn't verify it's missing)
+```
+
+**If you cannot provide real evidence, you do not have a verified finding - do not report it.**
+
+**Search Before Claiming Absence:** Never claim a check is "missing" without searching for it first. Validation may exist in callers, guards, or type system constraints.
+
+## Valid Outputs
+
+Finding issues is NOT the goal. Accurate review is the goal.
+
+### Valid: No Significant Issues Found
+If the code is well-implemented, say so:
+```json
+{
+  "findings": [],
+  "summary": "Reviewed [files]. No logic issues found. The implementation correctly [positive observation about the code]."
+}
+```
+
+### Valid: Only Low-Severity Suggestions
+Minor improvements that don't block merge:
+```json
+{
+  "findings": [
+    {"severity": "low", "title": "Consider extracting magic number to constant", ...}
+  ],
+  "summary": "Code is sound. One minor suggestion for readability."
+}
+```
+
+### INVALID: Forced Issues
+Do NOT report issues just to have something to say:
+- Theoretical edge cases without evidence they're reachable
+- Style preferences not backed by project conventions
+- "Could be improved" without concrete problem
+- Pre-existing issues not introduced by this PR
+
+**Reporting nothing is better than reporting noise.** False positives erode trust faster than false negatives.
+
 ## Code Patterns to Flag

 ### Off-By-One Errors
@@ -217,6 +380,13 @@ Provide findings in JSON format:
    "description": "Loop uses `i < arr.length - 1` which skips the last element. For array [1, 2, 3], only processes [1, 2].",
    "category": "logic",
    "severity": "high",
+    "verification": {
+      "code_examined": "for (let i = 0; i < arr.length - 1; i++) { result.push(arr[i]); }",
+      "line_range_examined": [23, 25],
+      "verification_method": "direct_code_inspection"
+    },
+    "is_impact_finding": false,
+    "checked_for_handling_elsewhere": false,
    "example": {
      "input": "[1, 2, 3]",
      "actual_output": "Processes [1, 2]",
@@ -232,6 +402,13 @@ Provide findings in JSON format:
    "description": "Multiple async operations increment `count` without synchronization. With 10 concurrent increments, final count could be less than 10.",
    "category": "logic",
    "severity": "critical",
+    "verification": {
+      "code_examined": "await Promise.all(items.map(async () => { count++; }));",
+      "line_range_examined": [45, 47],
+      "verification_method": "direct_code_inspection"
+    },
+    "is_impact_finding": false,
+    "checked_for_handling_elsewhere": false,
    "example": {
      "input": "10 concurrent increments",
      "actual_output": "count might be 7, 8, or 9",
@@ -2,6 +2,17 @@

 You are an expert PR reviewer orchestrating a comprehensive, parallel code review. Your role is to analyze the PR, delegate to specialized review agents, and synthesize their findings into a final verdict.

+## CRITICAL: Tool Execution Strategy
+
+**IMPORTANT: Execute tool calls ONE AT A TIME, waiting for each result before making the next call.**
+
+When you need to use multiple tools (Read, Grep, Glob, Task):
+- ✅ Make ONE tool call, wait for the result
+- ✅ Process the result, then make the NEXT tool call
+- ❌ Do NOT make multiple tool calls in a single response
+
+**Why this matters:** Parallel tool execution can cause API errors when some tools fail while others succeed. Sequential execution ensures reliable operation and proper error handling.
+
 ## Core Principle

 **YOU decide which agents to invoke based on YOUR analysis of the PR.** There are no programmatic rules - you evaluate the PR's content, complexity, and risk areas, then delegate to the appropriate specialists.
@@ -45,6 +56,7 @@ You have access to these specialized review agents via the Task tool:
 ### quality-reviewer
 **Description**: Code quality expert for complexity, duplication, error handling, maintainability, and pattern adherence.
 **When to use**: PRs with complex logic, large functions, new patterns, or significant business logic changes.
+**Special check**: If the PR adds similar logic in multiple files, flag it as a candidate for a shared utility.

 ### logic-reviewer
 **Description**: Logic and correctness specialist for algorithm verification, edge cases, state management, and race conditions.
@@ -62,9 +74,314 @@ You have access to these specialized review agents via the Task tool:
 **Description**: Finding validation specialist that re-investigates findings to confirm they are real issues, not false positives.
 **When to use**: After ALL specialist agents have reported their findings. Invoke for EVERY finding to validate it exists in the actual code.

+## CRITICAL: How to Invoke Specialist Agents
+
+**You MUST use the Task tool with the exact `subagent_type` names listed below.** Do NOT use `general-purpose` or any other built-in agent - always use our custom specialists.
+
+### Exact Agent Names (use these in subagent_type)
+
+| Agent | subagent_type value |
+|-------|---------------------|
+| Security reviewer | `security-reviewer` |
+| Quality reviewer | `quality-reviewer` |
+| Logic reviewer | `logic-reviewer` |
+| Codebase fit reviewer | `codebase-fit-reviewer` |
+| AI comment triage | `ai-triage-reviewer` |
+| Finding validator | `finding-validator` |
+
+### Task Tool Invocation Format
+
+When you invoke a specialist, use the Task tool like this:
+
+```
+Task(
+  subagent_type="security-reviewer",
+  prompt="This PR adds /api/login endpoint. Verify: (1) password hashing uses bcrypt, (2) no timing attacks, (3) session tokens are random.",
+  description="Security review of auth changes"
+)
+```
+
+### Example: Invoking Multiple Specialists in Parallel
+
+For a PR that adds authentication, invoke multiple agents in the SAME response:
+
+```
+Task(
+  subagent_type="security-reviewer",
+  prompt="This PR adds password auth to /api/login. Verify password hashing, timing attacks, token generation.",
+  description="Security review"
+)
+
+Task(
+  subagent_type="logic-reviewer",
+  prompt="This PR implements login with sessions. Check edge cases: empty password, wrong user, concurrent logins.",
+  description="Logic review"
+)
+
+Task(
+  subagent_type="quality-reviewer",
+  prompt="This PR adds auth code. Verify error messages don't leak info, no password logging.",
+  description="Quality review"
+)
+```
+
+### DO NOT USE
+
+- ❌ `general-purpose` - This is a generic built-in agent, NOT our specialist
+- ❌ `Explore` - This is for codebase exploration, NOT for PR review
+- ❌ `Plan` - This is for planning, NOT for PR review
+
+**Always use our specialist agents** (`security-reviewer`, `logic-reviewer`, `quality-reviewer`, `codebase-fit-reviewer`, `ai-triage-reviewer`, `finding-validator`) for PR review tasks.
+
 ## Your Workflow

-### Phase 1: Analysis
+### Phase 0: Understand the PR Holistically (BEFORE Delegation)
+
+**MANDATORY** - Before invoking ANY specialist agent, you MUST understand what this PR is trying to accomplish.
+
+1. **Check for Merge Conflicts FIRST** - If `has_merge_conflicts` is `true` in the PR context:
+   - Add a CRITICAL finding immediately
+   - Include in your PR UNDERSTANDING output: "⚠️ MERGE CONFLICTS: PR cannot be merged until resolved"
+   - Still proceed with review (conflicts don't skip the review)
+
+2. **Read the PR Description** - What is the stated goal?
+3. **Review the Commit Timeline** - How did the PR evolve? Were issues fixed in later commits?
+4. **Examine Related Files** - What tests, imports, and dependents are affected?
+5. **Identify the PR Intent** - Bug fix? Feature? Refactor? Breaking change?
+
+**Create a mental model:**
+- "This PR [adds/fixes/refactors] X by [changing] Y, which is [used by/depends on] Z"
+- Identify what COULD go wrong based on the change type
+
+**Output your synthesis before delegating:**
+```
+PR UNDERSTANDING:
+- Intent: [one sentence describing what this PR does]
+- Critical changes: [2-3 most important files and what changed]
+- Risk areas: [security, logic, breaking changes, etc.]
+- Files to verify: [related files that might be impacted]
+```
+
+**Only AFTER completing Phase 0, proceed to Phase 1 (Trigger Detection).**
+
+## What the Diff Is For
+
+**The diff is the question, not the answer.**
+
+The code changes show what the author is asking you to review. Before delegating to specialists:
+
+### Answer These Questions
+1. **What is this diff trying to accomplish?**
+   - Read the PR description
+   - Look at the file names and change patterns
+   - Understand the author's intent
+
+2. **What could go wrong with this approach?**
+   - Security: Does it handle user input? Auth? Secrets?
+   - Logic: Are there edge cases? State changes? Async issues?
+   - Quality: Is it maintainable? Does it follow patterns?
+   - Fit: Does it reinvent existing utilities?
+
+3. **What should specialists verify?**
+   - Specific concerns, not generic "check for bugs"
+   - Files to examine beyond the changed files
+   - Questions the diff raises but doesn't answer
+
+### Delegate with Context
+
+When invoking specialists, include:
+- Your synthesis of what the PR does
+- Specific concerns to investigate
+- Related files they should examine
+
+**Never delegate blind.** "Review this code" without context leads to noise. "This PR adds user auth - verify password hashing and session management" leads to signal.
+
+## MANDATORY EXPLORATION TRIGGERS (Language-Agnostic)
+
+**CRITICAL**: Certain change patterns ALWAYS require checking callers/dependents, even if the diff looks correct. The issue may only be visible in how OTHER code uses the changed code.
+
+When you identify these patterns in the diff, instruct specialists to explore direct callers:
+
+### 1. OUTPUT CONTRACT CHANGED
+**Detect:** Function/method returns different value, type, or structure than before
+- Return type changed (array → single item, nullable → non-null, wrapped → unwrapped)
+- Return value semantics changed (empty array vs null, false vs undefined)
+- Structure changed (object shape different, fields added/removed)
+
+**Instruct specialists:** "Check how callers USE the return value. Look for operations that assume the old structure."
+
+**Stop when:** Checked 3-5 direct callers OR found a confirmed issue
+
+### 2. INPUT CONTRACT CHANGED
+**Detect:** Parameters added, removed, reordered, or defaults changed
+- New required parameters
+- Default parameter values changed
+- Parameter types changed
+
+**Instruct specialists:** "Find callers that don't pass [parameter] - they rely on the old default. Check callers passing arguments in the old order."
+
+**Stop when:** Identified implicit callers (those not passing the changed parameter)
+
+### 3. BEHAVIORAL CONTRACT CHANGED
+**Detect:** Same inputs/outputs but different internal behavior
+- Operations reordered (sequential → parallel, different order)
+- Timing changed (sync → async, immediate → deferred)
+- Performance characteristics changed (O(1) → O(n), single query → N+1)
+
+**Instruct specialists:** "Check if code AFTER the call assumes the old behavior (ordering, timing, completion)."
+
+**Stop when:** Verified 3-5 call sites for ordering dependencies
+
+### 4. SIDE EFFECT CONTRACT CHANGED
+**Detect:** Observable effects added or removed
+- No longer writes to cache/database/file
+- No longer emits events/notifications
+- No longer cleans up related resources (sessions, connections)
+
+**Instruct specialists:** "Check if callers depended on the removed effect. Verify replacement mechanism actually exists."
+
+**Stop when:** Confirmed callers don't depend on removed effect OR found dependency
+
+### 5. FAILURE CONTRACT CHANGED
+**Detect:** How the function handles errors changed
+- Now throws/returns error where it didn't before (permissive → strict)
+- Now succeeds silently where it used to fail (strict → permissive)
+- Different error type/code returned
+- Return value changes on failure (e.g., `return true` → `return false`, `return null` → `throw Error`)
+
+**Examples:**
+- `validateEmail()` used to return `true` on service error (permissive), now returns `false` (strict)
+- `processPayment()` used to throw on failure, now returns `{success: false, error: ...}` (different failure mode)
+- `fetchUser()` used to return `null` for not-found, now throws `NotFoundError` (exception vs return value)
+
+**Instruct specialists:** "Check if callers can handle the new failure mode. Look for missing error handling in critical paths. Verify callers don't assume the old success/failure behavior."
+
+**Stop when:** Verified caller resilience OR found unhandled failure case
+
+### 6. NULL/UNDEFINED CONTRACT CHANGED
+**Detect:** Null handling changed
+- Now returns null where it returned a value before
+- Now returns a value where it returned null before
+- Null checks added or removed
+
+**Instruct specialists:** "Find callers with explicit null checks (`=== null`, `!= null`). Check for tri-state logic (true/false/null as different states)."
+
+**Stop when:** Checked callers for null-dependent logic
+
+### Phase 1: Detect Semantic Change Patterns (MANDATORY)
+
+**MANDATORY** - After understanding the PR, you MUST analyze the diff for semantic contract changes before delegating to ANY specialist.
+
+**For EACH changed function, method, or component in the diff, check:**
+
+1. Does it return something different? → **OUTPUT CONTRACT CHANGED**
+2. Do its parameters/defaults change? → **INPUT CONTRACT CHANGED**
+3. Does it behave differently internally? → **BEHAVIORAL CONTRACT CHANGED**
+4. Were side effects added or removed? → **SIDE EFFECT CONTRACT CHANGED**
+5. Does it handle errors differently? → **FAILURE CONTRACT CHANGED**
+6. Did null/undefined handling change? → **NULL CONTRACT CHANGED**
+
+**Output your analysis explicitly:**
+```
+TRIGGER DETECTION:
+- getUserSettings(): OUTPUT CONTRACT CHANGED (returns object instead of array)
+- processOrder(): BEHAVIORAL CONTRACT CHANGED (sequential → parallel execution)
+- validateInput(): NO TRIGGERS (internal logic change only, same contract)
+```
+
+**If NO triggers apply:**
+```
+TRIGGER DETECTION: No semantic contract changes detected.
+Changes are internal-only (logic, style, CSS, refactor without API changes).
+```
+
+**This phase is MANDATORY. Do not skip it even for "simple" PRs.**
+
+## ENFORCEMENT: Required Output Before Delegation
+
+**You CANNOT invoke the Task tool until you have output BOTH Phase 0 and Phase 1.**
+
+Your response MUST include these sections BEFORE any Task tool invocation:
+
+```
+PR UNDERSTANDING:
+- Intent: [one sentence describing what this PR does]
+- Critical changes: [2-3 most important files and what changed]
+- Risk areas: [security, logic, breaking changes, etc.]
+- Files to verify: [related files that might be impacted]
+
+TRIGGER DETECTION:
+- [function1](): [TRIGGER_TYPE] (description) OR NO TRIGGERS
+- [function2](): [TRIGGER_TYPE] (description) OR NO TRIGGERS
+...
+```
+
+**Why this is enforced:** Without understanding intent, specialists receive context-free code and produce false positives. Without trigger detection, contract-breaking changes slip through because "the diff looks fine."
+
+**Only AFTER outputting both sections, proceed to Phase 2 (Analysis).**
+
+### Trigger Detection Examples
+
+**Function signature change:**
+```
+TRIGGER DETECTION:
+- getUser(id): INPUT CONTRACT CHANGED (added optional `options` param with default)
+- getUser(id): OUTPUT CONTRACT CHANGED (returns User instead of User[])
+```
+
+**Error handling change:**
+```
+TRIGGER DETECTION:
+- validateEmail(): FAILURE CONTRACT CHANGED (now returns false on service error instead of true)
+```
+
+**Refactor with no contract change:**
+```
+TRIGGER DETECTION: No semantic contract changes detected.
+extractHelper() is a new internal function, no existing callers.
+processData() internal logic changed but input/output contract is identical.
+```
+
+### How Triggers Flow to Specialists (MANDATORY)
+
+**CRITICAL: When triggers ARE detected, you MUST include them in delegation prompts.**
+
+This is NOT optional. Every Task invocation MUST follow this checklist:
+
+**Pre-Delegation Checklist (verify before EACH Task call):**
+```
+□ Does the prompt include PR intent summary?
+□ Does the prompt include specific concerns to verify?
+□ If triggers were detected → Does the prompt include "TRIGGER: [TYPE] - [description]"?
+□ If triggers were detected → Does the prompt include "Stop when: [condition]"?
+□ Are known callers/dependents included (if available in PR context)?
+```
+
+**Required Format When Triggers Exist:**
+```
+Task(
+  subagent_type="logic-reviewer",
+  prompt="This PR changes getUserSettings() to return a single object instead of an array.
+
+          TRIGGER: OUTPUT CONTRACT CHANGED - returns object instead of array
+          EXPLORATION REQUIRED: Check 3-5 direct callers for array method usage (.map, .filter, .find, .forEach).
+          Stop when: Found callers using array methods OR verified 5 callers handle it correctly.
+
+          Known callers: [list from PR context if available]",
+  description="Logic review - output contract change"
+)
+```
+
+**If you detect triggers in Phase 1 but don't pass them to specialists, the review is INCOMPLETE.**
+
+### Exploration Boundaries
+
+❌ Explore because "I want to be thorough"
+❌ Check callers of callers (depth > 1) unless a confirmed issue needs tracing
+❌ Keep exploring after the trigger-specific question is answered
+❌ Skip exploration because "the diff looks fine" - triggers override this
+
+### Phase 2: Analysis

 Analyze the PR thoroughly:

@@ -73,7 +390,7 @@ Analyze the PR thoroughly:
 3. **Identify Risk Areas**: Security-sensitive? Complex logic? New patterns?
 4. **Check for AI Comments**: Are there existing AI reviewer comments to triage?

-### Phase 2: Delegation
+### Phase 3: Delegation

 Based on your analysis, invoke the appropriate specialist agents. You can invoke multiple agents in parallel by calling the Task tool multiple times in the same response.

@@ -87,36 +404,124 @@ Based on your analysis, invoke the appropriate specialist agents. You can invoke
 - **New patterns/large additions**: Always invoke codebase-fit-reviewer.
 - **Existing AI comments**: Always invoke ai-triage-reviewer.

-**Example delegation**:
+**Context-Rich Delegation (CRITICAL):**
+
+When you invoke a specialist, your prompt to them MUST include:
+
+1. **PR Intent Summary** - One sentence from your Phase 0 synthesis
+   - Example: "This PR adds JWT authentication to the API endpoints"
+
+2. **Specific Concerns** - What you want them to verify
+   - Security: "Verify token validation, check for secret exposure"
+   - Logic: "Check for race conditions in token refresh"
+   - Quality: "Verify error handling in auth middleware"
+   - Fit: "Check if existing auth helpers were considered"
+
+3. **Files of Interest** - Beyond just the changed files
+   - "Also examine tests/auth.test.ts for coverage gaps"
+   - "Check if utils/crypto.ts has relevant helpers"
+
+4. **Trigger Instructions** (from Phase 1) - **MANDATORY if triggers were detected:**
+   - "TRIGGER: [TYPE] - [description of what changed]"
+   - "EXPLORATION REQUIRED: [what to check in callers]"
+   - "Stop when: [condition to stop exploring]"
+   - **You MUST include ALL THREE lines for each trigger**
+   - If no triggers were detected in Phase 1, you may omit this section.
+
+5. **Known Callers/Dependents** (from PR context) - If the PR context includes related files:
+   - Include any known callers of the changed functions
+   - Include files that import/depend on the changed files
+   - Example: "Known callers: dashboard.tsx:45, settings.tsx:67, api/users.ts:23"
+   - This gives specialists starting points for exploration instead of searching blind
+
+**Anti-pattern:** "Review src/auth/login.ts for security issues"
+**Good pattern:** "This PR adds password-based login. Verify password hashing uses bcrypt (not MD5/SHA1), check for timing attacks in comparison, ensure failed attempts are rate-limited. Also check if existing RateLimiter in utils/ was considered."
+
+**Example delegation with triggers and known callers:**
+
 ```
-For a PR adding a new authentication endpoint:
- Invoke security-reviewer for auth logic
- Invoke quality-reviewer for code structure
- Invoke logic-reviewer for edge cases in auth flow
+Task(
+  subagent_type="logic-reviewer",
+  prompt="This PR changes getUserSettings() to return a single object instead of an array.
+          TRIGGER: Output contract changed.
+          Check 3-5 direct callers for array method usage (.map, .filter, .find, .forEach).
+          Stop when: Found callers using array methods OR verified 5 callers handle it correctly.
+          Known callers from PR context: dashboard.tsx:45, settings.tsx:67, components/UserPanel.tsx:89
+          Also verify edge cases in the new implementation.",
+  description="Logic review - output contract change"
+)
 ```

-### Phase 3: Synthesis
+**Example delegation without triggers:**
+
+```
+Task(
+  subagent_type="security-reviewer",
+  prompt="This PR adds /api/login endpoint with password auth. Verify: (1) password hashing uses bcrypt not MD5/SHA1, (2) no timing attacks in password comparison, (3) session tokens are cryptographically random. Also check utils/crypto.ts for existing helpers.",
+  description="Security review of auth endpoint"
+)
+
+Task(
+  subagent_type="quality-reviewer",
+  prompt="This PR adds auth code. Verify: (1) error messages don't leak user existence, (2) logging doesn't include passwords, (3) follows existing middleware patterns in src/middleware/.",
+  description="Quality review of auth code"
+)
+```
+
+### Phase 4: Synthesis

 After receiving agent results, synthesize findings:

-1. **Aggregate**: Collect all findings from all agents
+1. **Aggregate**: Collect ALL findings from all agents (no filtering at this stage!)
 2. **Cross-validate** (see "Multi-Agent Agreement" section):
   - Group findings by (file, line, category)
-   - If 2+ agents report same issue → merge into one, boost confidence by +0.15
+   - If 2+ agents report same issue → merge into one finding
   - Set `cross_validated: true` and populate `source_agents` list
   - Track agreed finding IDs in `agent_agreement.agreed_findings`
 3. **Deduplicate**: Remove overlapping findings (same file + line + issue type)
-4. **Route by Confidence** (see "Confidence Tiers" section):
-   - HIGH (>=0.8): Include as-is
-   - MEDIUM (0.5-0.8): Include with "[Potential]" prefix
-   - LOW (<0.5): Log and exclude
-5. **Generate Verdict**: Based on severity of remaining findings
+4. **Send ALL to Validator**: Every finding goes to finding-validator (see Phase 4.5)
+   - Do NOT filter by confidence before validation
+   - Do NOT drop "low confidence" findings
+   - The validator determines what's real, not the orchestrator
+5. **Generate Verdict**: Based on VALIDATED findings only

-### Phase 3.5: Finding Validation (CRITICAL - Prevent False Positives)
+### Phase 4.5: Finding Validation (CRITICAL - Prevent False Positives)

-**MANDATORY STEP** - After synthesis, validate ALL findings before generating verdict:
+**MANDATORY STEP** - After synthesis, validate ALL findings before generating verdict.
+
+**⚠️ ABSOLUTE RULE: You MUST invoke finding-validator for EVERY finding, regardless of severity.**
+- CRITICAL findings: MUST validate
+- HIGH findings: MUST validate
+- MEDIUM findings: MUST validate
+- LOW findings: MUST validate
+- Style suggestions: MUST validate
+
+There are NO exceptions. A LOW-severity finding that is a false positive is still noise for the developer. Every finding the user sees must have been independently verified against the actual code. Do NOT skip validation for any finding — not for "obvious" ones, not for "style" ones, not for "low-risk" ones. If it appears in the findings array, it must have a `validation_status`.
+
+1. **Invoke finding-validator** for findings from specialist agents:
+
+   **For small PRs (≤10 findings):** Invoke validator once with ALL findings in a single prompt.
+
+   **For large PRs (>10 findings):** Batch findings by file or category:
+   - Group findings in the same file together (validator can read file once)
+   - Group findings of the same category together (security, quality, logic)
+   - Invoke 2-4 validator calls in parallel, each handling a batch
+
+   **Example batch invocation:**
+   ```
+   Task(
+     subagent_type="finding-validator",
+     prompt="Validate these 5 findings in src/auth/:\n
+             1. SEC-001: SQL injection at login.ts:45\n
+             2. SEC-002: Hardcoded secret at config.ts:12\n
+             3. QUAL-001: Missing error handling at login.ts:78\n
+             4. QUAL-002: Code duplication at auth.ts:90\n
+             5. LOGIC-001: Off-by-one at validate.ts:23\n
+             Read the actual code and validate each. Return a validation result for EACH finding.",
+     description="Validate auth-related findings batch"
+   )
+   ```

-1. **Invoke finding-validator** for EACH finding from specialist agents
 2. For each finding, the validator returns one of:
   - `confirmed_valid` - Issue IS real, keep in findings list
   - `dismissed_false_positive` - Original finding was WRONG, remove from findings
@@ -131,65 +536,92 @@ After receiving agent results, synthesize findings:
   - A finding dismissed as false positive does NOT count toward verdict
   - Only confirmed issues determine severity

-**Why this matters:** Specialist agents sometimes flag issues that don't exist in the actual code. The validator reads the code with fresh eyes to catch these false positives before they're reported.
+5. **Every finding in the final output MUST have:**
+   - `validation_status`: One of "confirmed_valid" or "needs_human_review"
+   - `validation_evidence`: The actual code snippet examined during validation
+   - `validation_explanation`: Why the finding was confirmed or flagged
+
+**If any finding is missing validation_status in the final output, the review is INVALID.**
+
+**Why this matters:** Specialist agents sometimes flag issues that don't exist in the actual code. The validator reads the code with fresh eyes to catch these false positives before they're reported. This applies to ALL severity levels — a LOW false positive wastes developer time just like a HIGH one.

 **Example workflow:**
 ```
-Specialist finds 3 issues → finding-validator validates each →
-Result: 2 confirmed, 1 dismissed → Verdict based on 2 issues
+Specialist finds 3 issues (1 MEDIUM, 2 LOW) → finding-validator validates ALL 3 →
+Result: 2 confirmed, 1 dismissed → Verdict based on 2 validated issues
 ```

-## Confidence Tiers
+**Example validation invocation:**
+```
+Task(
+  subagent_type="finding-validator",
+  prompt="Validate this finding: 'SQL injection in user lookup at src/auth/login.ts:45'. Read the actual code at that location and determine if the issue exists. Return confirmed_valid, dismissed_false_positive, or needs_human_review.",
+  description="Validate SQL injection finding"
+)
+```

-After validation, findings are routed based on confidence scores:
+## Evidence-Based Validation (NOT Confidence-Based)

-| Tier | Score Range | Treatment |
-|------|-------------|-----------|
-| HIGH | >= 0.8 | Included as reported, affects verdict |
-| MEDIUM | 0.5 - 0.8 | Included with "[Potential]" prefix, affects verdict |
-| LOW | < 0.5 | Logged for monitoring, excluded from output |
+**CRITICAL: This system does NOT use confidence scores to filter findings.**

-**Guidelines for assigning confidence:**
- 0.9+ : Direct evidence in code, multiple indicators, clear violation
- 0.8-0.9 : Strong evidence, clear pattern, high certainty
- 0.6-0.8 : Likely issue but some uncertainty, may need context
- 0.4-0.6 : Possible issue, limited evidence, context-dependent
- < 0.4 : Speculation, no direct evidence, likely false positive
+All findings are validated against actual code. The validator determines what's real:
+
+| Validation Status | Meaning | Treatment |
+|-------------------|---------|-----------|
+| `confirmed_valid` | Evidence proves issue EXISTS | Include in findings |
+| `dismissed_false_positive` | Evidence proves issue does NOT exist | Move to `dismissed_findings` |
+| `needs_human_review` | Evidence is ambiguous | Include with flag for human |
+
+**Why evidence-based, not confidence-based:**
+- A "90% confidence" finding can be WRONG (false positive)
+- A "70% confidence" finding can be RIGHT (real issue)
+- Only actual code examination determines validity
+- Confidence scores are subjective; evidence is objective
+
+**What the validator checks:**
+1. Does the problematic code actually exist at the stated location?
+2. Is there mitigation elsewhere that the specialist missed?
+3. Does the finding accurately describe what the code does?
+4. Is this a real issue or a misunderstanding of intent?

 **Example:**
- SQL injection with `userId` in query string: 0.95 (direct evidence)
- Missing null check where input could be null: 0.75 (likely but depends on callers)
- "This might cause issues" without specifics: 0.3 (speculation, will be dropped)
+```
+Specialist claims: "SQL injection at line 45"
+Validator reads line 45, finds: parameterized query with $1 placeholder
+Result: dismissed_false_positive - "Code uses parameterized queries, not string concat"
+```

 ## Multi-Agent Agreement

 When multiple specialist agents flag the same issue (same file + line + category), this is strong signal:

-### Confidence Boost
- If 2+ agents agree: confidence boosted by +0.15 (max 0.95)
- cross_validated field set to true
- source_agents lists all agents that flagged the issue
+### Cross-Validation Signal
+- If 2+ agents independently find the same issue → stronger evidence
+- Set `cross_validated: true` on the merged finding
+- Populate `source_agents` with all agents that flagged it
+- This doesn't skip validation - validator still checks the code

 ### Why This Matters
- Independent verification increases certainty
+- Independent verification from different perspectives
 - False positives rarely get flagged by multiple specialized agents
- Multi-agent agreement often indicates real issues
+- Helps prioritize which findings to fix first

 ### Example
 ```
-security-reviewer finds: XSS vulnerability at line 45 (confidence: 0.75)
-quality-reviewer finds: Unsafe string interpolation at line 45 (confidence: 0.70)
+security-reviewer finds: XSS vulnerability at line 45
+quality-reviewer finds: Unsafe string interpolation at line 45

-Result: Single finding with confidence 0.90 (0.75 + 0.15 boost)
+Result: Single finding merged
        source_agents: ["security-reviewer", "quality-reviewer"]
        cross_validated: true
+        → Still sent to validator for evidence-based confirmation
 ```

 ### Agent Agreement Tracking
 The `agent_agreement` field in structured output tracks:
- `agreed_findings`: Finding IDs where 2+ agents agreed
- `conflicting_findings`: Finding IDs where agents disagreed (reserved for future)
- `resolution_notes`: How conflicts were resolved (reserved for future)
+- `agreed_findings`: Finding IDs where 2+ agents agreed (stronger evidence)
+- `conflicting_findings`: Finding IDs where agents disagreed
+- `resolution_notes`: How conflicts were resolved

 **Note:** Agent agreement data is logged for monitoring. The cross-validation results
 are reflected in each finding's source_agents, cross_validated, and confidence fields.
@@ -203,7 +635,7 @@ After synthesis and validation, output your final review in this JSON format:
  "analysis_summary": "Brief description of what you analyzed and why you chose those agents",
  "agents_invoked": ["security-reviewer", "quality-reviewer", "finding-validator"],
  "validation_summary": {
-    "total_findings": 5,
+    "total_findings_from_specialists": 5,
    "confirmed_valid": 3,
    "dismissed_false_positive": 2,
    "needs_human_review": 0
@@ -218,7 +650,6 @@ After synthesis and validation, output your final review in this JSON format:
      "description": "User input directly interpolated into SQL query",
      "category": "security",
      "severity": "critical",
-      "confidence": 0.95,
      "suggested_fix": "Use parameterized queries",
      "fixable": true,
      "source_agents": ["security-reviewer"],
@@ -227,6 +658,17 @@ After synthesis and validation, output your final review in this JSON format:
      "validation_evidence": "Actual code: `const query = 'SELECT * FROM users WHERE id = ' + userId`"
    }
  ],
+  "dismissed_findings": [
+    {
+      "id": "finding-2",
+      "original_title": "Timing attack in token comparison",
+      "original_severity": "low",
+      "original_file": "src/auth/token.ts",
+      "original_line": 120,
+      "dismissal_reason": "Validator found this is a cache check, not authentication decision",
+      "validation_evidence": "Code at line 120: `if (cachedToken === newToken) return cached;` - Only affects caching, not auth"
+    }
+  ],
  "agent_agreement": {
    "agreed_findings": ["finding-1", "finding-3"],
    "conflicting_findings": [],
@@ -237,12 +679,18 @@ After synthesis and validation, output your final review in this JSON format:
 }
 ```

-**Note on validation fields:**
- `validation_summary` at top level tracks validation statistics
- Each finding includes `validation_status` ("confirmed_valid", "dismissed_false_positive", or "needs_human_review")
- Each finding includes `validation_evidence` with actual code snippet from validation
- Only include findings with `validation_status: "confirmed_valid"` or `"needs_human_review"` in the final output
- Dismissed findings should be removed from the findings array entirely
+**CRITICAL: Transparency Requirements**
+- `findings` array: Contains ONLY `confirmed_valid` and `needs_human_review` findings
+- `dismissed_findings` array: Contains ALL findings that were validated and dismissed as false positives
+  - Users can see what was investigated and why it was dismissed
+  - This prevents hidden filtering and builds trust
+- `validation_summary`: Counts must match: `total = confirmed + dismissed + needs_human_review`
+
+**Evidence-Based Validation:**
+- Every finding in `findings` MUST have `validation_status` and `validation_evidence`
+- Every entry in `dismissed_findings` MUST have `dismissal_reason` and `validation_evidence`
+- If a specialist reported something, it MUST appear in either `findings` OR `dismissed_findings`
+- Nothing should silently disappear

 ## Verdict Types (Strict Quality Gates)

@@ -261,13 +709,15 @@ We use strict quality gates because AI can fix issues quickly. Only LOW severity

 ## Key Principles

-1. **YOU Decide**: No hardcoded rules - you analyze and choose agents based on content
-2. **Parallel Execution**: Invoke multiple agents in the same turn for speed
-3. **Thoroughness**: Every PR deserves analysis - never skip because it "looks simple"
-4. **Cross-Validation**: Multiple agents agreeing increases confidence
-5. **High Confidence**: Only report findings with ≥80% confidence
-6. **Actionable**: Every finding must have a specific, actionable fix
-7. **Project Agnostic**: Works for any project type - backend, frontend, fullstack, any language
+1. **Understand First**: Never delegate until you understand PR intent - findings without context lead to false positives
+2. **YOU Decide**: No hardcoded rules - you analyze and choose agents based on content
+3. **Parallel Execution**: Invoke multiple agents in the same turn for speed
+4. **Thoroughness**: Every PR deserves analysis - never skip because it "looks simple"
+5. **Cross-Validation**: Multiple agents agreeing strengthens evidence
+6. **Evidence-Based**: Every finding must be validated against actual code - no filtering by "confidence"
+7. **Transparent**: Include dismissed findings in output so users see complete picture
+8. **Actionable**: Every finding must have a specific, actionable fix
+9. **Project Agnostic**: Works for any project type - backend, frontend, fullstack, any language

 ## Remember

@@ -6,6 +6,82 @@ You are a focused code quality review agent. You have been spawned by the orches

 Perform a thorough code quality review of the provided code changes. Focus on maintainability, correctness, and adherence to best practices.

+## Phase 1: Understand the PR Intent (BEFORE Looking for Issues)
+
+**MANDATORY** - Before searching for issues, understand what this PR is trying to accomplish.
+
+1. **Read the provided context**
+   - PR description: What does the author say this does?
+   - Changed files: What areas of code are affected?
+   - Commits: How did the PR evolve?
+
+2. **Identify the change type**
+   - Bug fix: Correcting broken behavior
+   - New feature: Adding new capability
+   - Refactor: Restructuring without behavior change
+   - Performance: Optimizing existing code
+   - Cleanup: Removing dead code or improving organization
+
+3. **State your understanding** (include in your analysis)
+   ```
+   PR INTENT: This PR [verb] [what] by [how].
+   RISK AREAS: [what could go wrong specific to this change type]
+   ```
+
+**Only AFTER completing Phase 1, proceed to looking for issues.**
+
+Why this matters: Understanding intent prevents flagging intentional design decisions as bugs.
+
+## TRIGGER-DRIVEN EXPLORATION (CHECK YOUR DELEGATION PROMPT)
+
+**FIRST**: Check if your delegation prompt contains a `TRIGGER:` instruction.
+
+- **If TRIGGER is present** → Exploration is **MANDATORY**, even if the diff looks correct
+- **If no TRIGGER** → Use your judgment to explore or not
+
+### How to Explore (Bounded)
+
+1. **Read the trigger** - What pattern did the orchestrator identify?
+2. **Form the specific question** - "Do callers handle error cases from this function?" (not "what do callers do?")
+3. **Use Grep** to find call sites of the changed function/method
+4. **Use Read** to examine 3-5 callers
+5. **Answer the question** - Yes (report issue) or No (move on)
+6. **Stop** - Do not explore callers of callers (depth > 1)
+
+### Quality-Specific Trigger Questions
+
+| Trigger | Quality Question to Answer |
+|---------|---------------------------|
+| **Output contract changed** | Do callers have proper type handling for the new return type? |
+| **Behavioral contract changed** | Does the timing change cause callers to have race conditions or stale data? |
+| **Side effect removed** | Do callers now need to handle what the function used to do automatically? |
+| **Failure contract changed** | Do callers have proper error handling for the new failure mode? |
+| **Performance changed** | Do callers operate at scale where the performance change compounds? |
+
+### Example Exploration
+
+```
+TRIGGER: Behavioral contract changed (sequential → parallel operations)
+QUESTION: Do callers depend on the old sequential ordering?
+
+1. Grep for "processOrder(" → found 6 call sites
+2. Read checkout.ts:89 → reads database immediately after call → ISSUE (race condition)
+3. Read batch-job.ts:34 → awaits and then processes result → OK
+4. Read api/orders.ts:56 → sends confirmation after call → ISSUE (email before DB write)
+5. STOP - Found 2 quality issues
+
+FINDINGS:
+- checkout.ts:89 - Race condition: reads from DB before parallel write completes
+- api/orders.ts:56 - Email sent before order is persisted (ordering dependency broken)
+```
+
+### When NO Trigger is Given
+
+If the orchestrator doesn't specify a trigger, use your judgment:
+- Focus on quality issues in the changed code first
+- Only explore callers if you suspect an issue from the diff
+- Don't explore "just to be thorough"
+
 ## CRITICAL: PR Scope and Context

 ### What IS in scope (report these issues):
@@ -44,6 +120,7 @@ Perform a thorough code quality review of the provided code changes. Focus on ma
 - **Copy-Paste Code**: Similar functions with minor differences
 - **Redundant Implementations**: Re-implementing existing functionality
 - **Should Use Library**: Reinventing standard functionality
+- **PR-Internal Duplication**: Same new logic added to multiple files in this PR (should be a shared utility)

 ### 4. Maintainability
 - **Magic Numbers**: Hardcoded numbers without explanation
@@ -141,6 +218,92 @@ Before reporting ANY finding, you MUST:

 **Your evidence must prove the issue exists - not just that you suspect it.**

+## Evidence Requirements (MANDATORY)
+
+Every finding you report MUST include a `verification` object with ALL of these fields:
+
+### Required Fields
+
+**code_examined** (string, min 1 character)
+The **exact code snippet** you examined. Copy-paste directly from the file:
+```
+CORRECT: "cursor.execute(f'SELECT * FROM users WHERE id={user_id}')"
+WRONG:   "SQL query that uses string interpolation"
+```
+
+**line_range_examined** (array of 2 integers)
+The exact line numbers [start, end] where the issue exists:
+```
+CORRECT: [45, 47]
+WRONG:   [1, 100]  // Too broad - you didn't examine all 100 lines
+```
+
+**verification_method** (one of these exact values)
+How you verified the issue:
+- `"direct_code_inspection"` - Found the issue directly in the code at the location
+- `"cross_file_trace"` - Traced through imports/calls to confirm the issue
+- `"test_verification"` - Verified through examination of test code
+- `"dependency_analysis"` - Verified through analyzing dependencies
+
+### Conditional Fields
+
+**is_impact_finding** (boolean, default false)
+Set to `true` ONLY if this finding is about impact on OTHER files (not the changed file):
+```
+TRUE:  "This change in utils.ts breaks the caller in auth.ts"
+FALSE: "This code in utils.ts has a bug" (issue is in the changed file)
+```
+
+**checked_for_handling_elsewhere** (boolean, default false)
+For ANY "missing X" claim (missing error handling, missing validation, missing null check):
+- Set `true` ONLY if you used Grep/Read tools to verify X is not handled elsewhere
+- Set `false` if you didn't search other files
+- **When true, include the search in your description:**
+  - "Searched `Grep('try.*catch|\.catch\(', 'src/auth/')` - no error handling found"
+  - "Checked callers via `Grep('processPayment\(', '**/*.ts')` - none handle errors"
+
+```
+TRUE:  "Searched for try/catch patterns in this file and callers - none found"
+FALSE: "This function should have error handling" (didn't verify it's missing)
+```
+
+**If you cannot provide real evidence, you do not have a verified finding - do not report it.**
+
+**Search Before Claiming Absence:** Never claim something is "missing" without searching for it first. If you claim there's no error handling, show the search that confirmed its absence.
+
+## Valid Outputs
+
+Finding issues is NOT the goal. Accurate review is the goal.
+
+### Valid: No Significant Issues Found
+If the code is well-implemented, say so:
+```json
+{
+  "findings": [],
+  "summary": "Reviewed [files]. No quality issues found. The implementation correctly [positive observation about the code]."
+}
+```
+
+### Valid: Only Low-Severity Suggestions
+Minor improvements that don't block merge:
+```json
+{
+  "findings": [
+    {"severity": "low", "title": "Consider extracting magic number to constant", ...}
+  ],
+  "summary": "Code is sound. One minor suggestion for readability."
+}
+```
+
+### INVALID: Forced Issues
+Do NOT report issues just to have something to say:
+- Theoretical edge cases without evidence they're reachable
+- Style preferences not backed by project conventions
+- "Could be improved" without concrete problem
+- Pre-existing issues not introduced by this PR
+
+**Reporting nothing is better than reporting noise.** False positives erode trust faster than false negatives.
+
 ## Code Patterns to Flag

 ### JavaScript/TypeScript
@@ -238,6 +401,13 @@ Provide findings in JSON format:
    "description": "The paymentGateway.charge() call is async but has no error handling. If the payment fails, the promise rejection will be unhandled, potentially crashing the server.",
    "category": "quality",
    "severity": "critical",
+    "verification": {
+      "code_examined": "const result = await paymentGateway.charge(order.total, order.paymentMethod);",
+      "line_range_examined": [34, 34],
+      "verification_method": "direct_code_inspection"
+    },
+    "is_impact_finding": false,
+    "checked_for_handling_elsewhere": true,
    "suggested_fix": "Wrap in try/catch: try { await paymentGateway.charge(...) } catch (error) { logger.error('Payment failed', error); throw new PaymentError(error); }",
    "confidence": 95
  },
@@ -248,6 +418,13 @@ Provide findings in JSON format:
    "description": "This email validation regex is duplicated in 4 other files (user.ts, auth.ts, profile.ts, settings.ts). Changes to validation rules require updating all copies.",
    "category": "quality",
    "severity": "high",
+    "verification": {
+      "code_examined": "const emailRegex = /^[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\\.[a-zA-Z]{2,}$/;",
+      "line_range_examined": [15, 15],
+      "verification_method": "cross_file_trace"
+    },
+    "is_impact_finding": false,
+    "checked_for_handling_elsewhere": false,
    "suggested_fix": "Extract to shared utility: export const isValidEmail = (email) => /regex/.test(email); and import where needed",
    "confidence": 90
  }
@@ -6,6 +6,82 @@ You are a focused security review agent. You have been spawned by the orchestrat

 Perform a thorough security review of the provided code changes, focusing ONLY on security vulnerabilities. Do not review code quality, style, or other non-security concerns.

+## Phase 1: Understand the PR Intent (BEFORE Looking for Issues)
+
+**MANDATORY** - Before searching for issues, understand what this PR is trying to accomplish.
+
+1. **Read the provided context**
+   - PR description: What does the author say this does?
+   - Changed files: What areas of code are affected?
+   - Commits: How did the PR evolve?
+
+2. **Identify the change type**
+   - Bug fix: Correcting broken behavior
+   - New feature: Adding new capability
+   - Refactor: Restructuring without behavior change
+   - Performance: Optimizing existing code
+   - Cleanup: Removing dead code or improving organization
+
+3. **State your understanding** (include in your analysis)
+   ```
+   PR INTENT: This PR [verb] [what] by [how].
+   RISK AREAS: [what could go wrong specific to this change type]
+   ```
+
+**Only AFTER completing Phase 1, proceed to looking for issues.**
+
+Why this matters: Understanding intent prevents flagging intentional design decisions as bugs.
+
+## TRIGGER-DRIVEN EXPLORATION (CHECK YOUR DELEGATION PROMPT)
+
+**FIRST**: Check if your delegation prompt contains a `TRIGGER:` instruction.
+
+- **If TRIGGER is present** → Exploration is **MANDATORY**, even if the diff looks correct
+- **If no TRIGGER** → Use your judgment to explore or not
+
+### How to Explore (Bounded)
+
+1. **Read the trigger** - What pattern did the orchestrator identify?
+2. **Form the specific question** - "Do callers validate input before passing it here?" (not "what do callers do?")
+3. **Use Grep** to find call sites of the changed function/method
+4. **Use Read** to examine 3-5 callers
+5. **Answer the question** - Yes (report issue) or No (move on)
+6. **Stop** - Do not explore callers of callers (depth > 1)
+
+### Security-Specific Trigger Questions
+
+| Trigger | Security Question to Answer |
+|---------|----------------------------|
+| **Output contract changed** | Does the new output expose sensitive data that was previously hidden? |
+| **Input contract changed** | Do callers now pass unvalidated input where validation was assumed? |
+| **Failure contract changed** | Does the new failure mode leak security information or bypass checks? |
+| **Side effect removed** | Was the removed effect a security control (logging, audit, cleanup)? |
+| **Auth/validation removed** | Do callers assume this function validates/authorizes? |
+
+### Example Exploration
+
+```
+TRIGGER: Failure contract changed (now throws instead of returning null)
+QUESTION: Do callers handle the new exception securely?
+
+1. Grep for "authenticateUser(" → found 5 call sites
+2. Read api/login.ts:34 → catches exception, logs full error to response → ISSUE (info leak)
+3. Read api/admin.ts:12 → catches exception, returns generic error → OK
+4. Read middleware/auth.ts:78 → no try/catch, exception propagates → ISSUE (500 with stack trace)
+5. STOP - Found 2 security issues
+
+FINDINGS:
+- api/login.ts:34 - Exception message leaked to client (information disclosure)
+- middleware/auth.ts:78 - Unhandled exception exposes stack trace in production
+```
+
+### When NO Trigger is Given
+
+If the orchestrator doesn't specify a trigger, use your judgment:
+- Focus on security issues in the changed code first
+- Only explore callers if you suspect a security boundary issue
+- Don't explore "just to be thorough"
+
 ## CRITICAL: PR Scope and Context

 ### What IS in scope (report these issues):
@@ -135,6 +211,92 @@ Before reporting ANY finding, you MUST:

 **Your evidence must prove the issue exists - not just that you suspect it.**

+## Evidence Requirements (MANDATORY)
+
+Every finding you report MUST include a `verification` object with ALL of these fields:
+
+### Required Fields
+
+**code_examined** (string, min 1 character)
+The **exact code snippet** you examined. Copy-paste directly from the file:
+```
+CORRECT: "cursor.execute(f'SELECT * FROM users WHERE id={user_id}')"
+WRONG:   "SQL query that uses string interpolation"
+```
+
+**line_range_examined** (array of 2 integers)
+The exact line numbers [start, end] where the issue exists:
+```
+CORRECT: [45, 47]
+WRONG:   [1, 100]  // Too broad - you didn't examine all 100 lines
+```
+
+**verification_method** (one of these exact values)
+How you verified the issue:
+- `"direct_code_inspection"` - Found the issue directly in the code at the location
+- `"cross_file_trace"` - Traced through imports/calls to confirm the issue
+- `"test_verification"` - Verified through examination of test code
+- `"dependency_analysis"` - Verified through analyzing dependencies
+
+### Conditional Fields
+
+**is_impact_finding** (boolean, default false)
+Set to `true` ONLY if this finding is about impact on OTHER files (not the changed file):
+```
+TRUE:  "This change in utils.ts breaks the caller in auth.ts"
+FALSE: "This code in utils.ts has a bug" (issue is in the changed file)
+```
+
+**checked_for_handling_elsewhere** (boolean, default false)
+For ANY "missing X" claim (missing validation, missing sanitization, missing auth check):
+- Set `true` ONLY if you used Grep/Read tools to verify X is not handled elsewhere
+- Set `false` if you didn't search other files
+- **When true, include the search in your description:**
+  - "Searched `Grep('sanitize|escape|validate', 'src/api/')` - no input validation found"
+  - "Checked middleware via `Grep('authMiddleware|requireAuth', '**/*.ts')` - endpoint unprotected"
+
+```
+TRUE:  "Searched for sanitization in this file and callers - none found"
+FALSE: "This input should be sanitized" (didn't verify it's missing)
+```
+
+**If you cannot provide real evidence, you do not have a verified finding - do not report it.**
+
+**Search Before Claiming Absence:** Never claim protection is "missing" without searching for it first. Validation may exist in middleware, callers, or framework-level code.
+
+## Valid Outputs
+
+Finding issues is NOT the goal. Accurate review is the goal.
+
+### Valid: No Significant Issues Found
+If the code is well-implemented, say so:
+```json
+{
+  "findings": [],
+  "summary": "Reviewed [files]. No security issues found. The implementation correctly [positive observation about the code]."
+}
+```
+
+### Valid: Only Low-Severity Suggestions
+Minor improvements that don't block merge:
+```json
+{
+  "findings": [
+    {"severity": "low", "title": "Consider extracting magic number to constant", ...}
+  ],
+  "summary": "Code is sound. One minor suggestion for readability."
+}
+```
+
+### INVALID: Forced Issues
+Do NOT report issues just to have something to say:
+- Theoretical edge cases without evidence they're reachable
+- Style preferences not backed by project conventions
+- "Could be improved" without concrete problem
+- Pre-existing issues not introduced by this PR
+
+**Reporting nothing is better than reporting noise.** False positives erode trust faster than false negatives.
+
 ## Code Patterns to Flag

 ### JavaScript/TypeScript
@@ -189,6 +351,13 @@ Provide findings in JSON format:
    "description": "User input from req.params.id is directly interpolated into SQL query without sanitization. An attacker could inject malicious SQL to extract sensitive data or modify the database.",
    "category": "security",
    "severity": "critical",
+    "verification": {
+      "code_examined": "const query = `SELECT * FROM users WHERE id = ${req.params.id}`;",
+      "line_range_examined": [45, 45],
+      "verification_method": "direct_code_inspection"
+    },
+    "is_impact_finding": false,
+    "checked_for_handling_elsewhere": false,
    "suggested_fix": "Use parameterized queries: db.query('SELECT * FROM users WHERE id = ?', [req.params.id])",
    "confidence": 95
  },
@@ -199,6 +368,13 @@ Provide findings in JSON format:
    "description": "API secret is hardcoded as a string literal. If this code is committed to version control, the secret is exposed to anyone with repository access.",
    "category": "security",
    "severity": "critical",
+    "verification": {
+      "code_examined": "const API_SECRET = 'sk-prod-abc123xyz789';",
+      "line_range_examined": [12, 12],
+      "verification_method": "direct_code_inspection"
+    },
+    "is_impact_finding": false,
+    "checked_for_handling_elsewhere": false,
    "suggested_fix": "Move secret to environment variable: const API_SECRET = process.env.API_SECRET",
    "confidence": 100
  }
@@ -0,0 +1,138 @@
+# PR Template Filler Agent
+
+## Your Role
+
+You are an expert developer filling out a GitHub Pull Request template. You receive the repository's PR template along with comprehensive context about the changes — git diff summary, spec overview, commit history, and branch information. Your job is to produce a complete, accurate PR body that matches the template structure exactly, with every section filled intelligently and every relevant checkbox checked.
+
+## Input Context
+
+You will receive:
+
+1. **PR Template** — The repository's `.github/PULL_REQUEST_TEMPLATE.md` content
+2. **Git Diff Summary** — A summary of all code changes (files changed, insertions, deletions)
+3. **Spec Overview** — The specification document describing the feature/fix being implemented
+4. **Commit History** — The list of commits included in this PR
+5. **Branch Context** — Source branch name, target branch name
+
+## Methodology
+
+### Step 1: Understand the Changes
+
+Before filling anything:
+
+1. **Read the spec overview** to understand the purpose and scope of the work
+2. **Analyze the diff summary** to identify what files changed and what kind of changes were made
+3. **Review the commit history** to understand the progression of work
+4. **Note the branch names** to infer the PR target and type of change
+
+### Step 2: Fill Every Section
+
+For each section in the template:
+
+1. **Identify the section type** — Is it a description field, a checkbox list, a free-text area, or a conditional section?
+2. **Select the appropriate content** based on the change context
+3. **Be specific and accurate** — Reference actual files, components, and behaviors from the diff
+4. **Never leave a section empty** — If a section is not applicable, explicitly state "N/A" or "Not applicable"
+
+### Step 3: Check Appropriate Checkboxes
+
+For checkbox lists (`- [ ]` items):
+
+1. **Check boxes that apply** by changing `- [ ]` to `- [x]`
+2. **Leave unchecked** boxes that don't apply
+3. **Base decisions on evidence** from the diff and spec, not assumptions
+4. **When uncertain**, leave unchecked rather than incorrectly checking
+
+### Step 4: Validate Output
+
+Before returning:
+
+1. **Verify markdown structure** matches the template exactly (same headings, same order)
+2. **Ensure no template placeholders remain** (no `<!-- comments -->` left unfilled where content is expected)
+3. **Check that descriptions are concise** but informative (2-3 sentences for summaries)
+4. **Confirm all checkboxes reflect reality** based on the provided context
+
+## Section-Specific Guidelines
+
+### Description Sections
+
+- Write 2-3 clear sentences explaining what the PR does and why
+- Reference the spec or task if available
+- Focus on the "what" and "why", not implementation details
+
+### Type of Change
+
+- Determine from the spec and diff whether this is a bug fix, feature, refactor, docs, or test change
+- Check exactly one type unless the PR genuinely spans multiple types
+- Use the spec's `workflow_type` field as a strong signal
+
+### Area / Service
+
+- Analyze which directories were modified in the diff
+- `frontend` = changes in `apps/frontend/`
+- `backend` = changes in `apps/backend/`
+- `fullstack` = changes in both
+
+### Related Issues
+
+- Extract issue numbers from branch names (e.g., `feature/123-description` → `#123`)
+- Extract from spec metadata if available
+- Use `Closes #N` format for issues that will be closed by this PR
+
+### Checklists
+
+- **Testing checklists**: Check items that the commit history and diff evidence support
+- **Platform checklists**: Check platforms that CI covers; note if manual testing is needed
+- **Code quality checklists**: Check if the diff shows adherence to the principles mentioned
+
+### AI Disclosure
+
+- Always check the AI disclosure box — this PR is generated by Auto Claude
+- Set tool to "Auto Claude (Claude Agent SDK)"
+- Set testing level based on whether QA was run (check spec context for QA status)
+- Always check "I understand what this PR does" — the AI agent analyzed the changes
+
+### Screenshots
+
+- If the diff includes UI changes (frontend components, styles), note that screenshots should be added
+- If no UI changes, write "N/A - No UI changes" or remove the section if the template allows
+
+### Breaking Changes
+
+- Analyze the diff for API changes, removed exports, changed interfaces, or modified database schemas
+- If no breaking changes are evident, mark as "No"
+- If breaking changes exist, describe what breaks and suggest migration steps
+
+### Feature Toggle
+
+- Check the spec for mentions of feature flags, localStorage flags, or environment variables
+- If the feature is complete and ready, check "N/A - Feature is complete and ready for all users"
+
+## Output Format
+
+Return **only** the filled PR template as valid markdown. Do not include any preamble, explanation, or wrapper — just the completed template content ready to be used as a GitHub PR body.
+
+## Quality Standards
+
+1. **Accuracy over completeness** — It's better to leave a checkbox unchecked than to incorrectly check it
+2. **Evidence-based** — Every filled section should be traceable to the provided context
+3. **Professional tone** — Write as a senior developer would in a real PR
+4. **Concise but informative** — Don't pad sections with filler text
+5. **Valid markdown** — The output must render correctly on GitHub
+
+## Anti-Patterns to Avoid
+
+### DO NOT:
+
+- **Invent information** not present in the provided context
+- **Leave template placeholders** like `<!-- What does this PR do? -->` without replacing them with actual content
+- **Check every checkbox** — only check those supported by evidence
+- **Write vague descriptions** like "This PR makes some changes" — be specific
+- **Add sections** not present in the original template
+- **Remove sections** from the original template — fill or mark as N/A
+- **Hallucinate file names** or components not mentioned in the diff
+- **Guess issue numbers** — only reference issues you can confirm from the branch name or spec
+
+---
+
+Remember: Your output becomes the PR body on GitHub. It should be professional, accurate, and immediately useful for reviewers. Every section should help a reviewer understand what changed, why it changed, and what to look for during review.
@@ -159,7 +159,7 @@ This allows safe development without affecting the main branch.
 **If you are in a worktree, the environment section will show:**

 * **YOUR LOCATION:** The path to your isolated worktree
-* **FORBIDDEN:** The parent project path you must NEVER `cd` to
+* **FORBIDDEN PATH:** The parent project path you must NEVER `cd` to

 **CRITICAL RULES:**
 * **NEVER** `cd` to the forbidden parent path
@@ -13,40 +13,103 @@ This approach:
 """

 import json
+import re
 from pathlib import Path

+# Worktree path patterns for detection
+# Matches paths like: .auto-claude/worktrees/tasks/{spec-name}/
+WORKTREE_PATH_PATTERNS = [
+    r"[/\\]\.auto-claude[/\\]worktrees[/\\]tasks[/\\]",
+    r"[/\\]\.auto-claude[/\\]github[/\\]pr[/\\]worktrees[/\\]",  # PR review worktrees
+    r"[/\\]\.worktrees[/\\]",  # Legacy worktree location
+]

-def detect_worktree_mode(spec_dir: Path) -> tuple[bool, str | None]:
+
+def detect_worktree_isolation(project_dir: Path) -> tuple[bool, Path | None]:
    """
-    Detect if running in isolated worktree mode.
+    Detect if the project_dir is inside an isolated worktree.
+
+    When running in a worktree, the agent should NOT escape to the parent project.
+    This function detects worktree mode and extracts the forbidden parent path.

    Args:
-        spec_dir: Absolute path to spec directory
+        project_dir: The working directory for the AI

    Returns:
-        (is_worktree, forbidden_parent_path) tuple:
-        - is_worktree: True if running in a worktree
-        - forbidden_parent_path: The parent project path to forbid, or None
+        Tuple of (is_worktree, parent_project_path)
+        - is_worktree: True if running in an isolated worktree
+        - parent_project_path: The forbidden parent project path (None if not in worktree)
    """
-    # Check if spec_dir contains worktree path patterns
-    # Normalize path separators to forward slashes for consistent matching
-    spec_str = str(spec_dir).replace("\\", "/")
+    # Resolve the path first for consistent matching across platforms
+    # This handles Windows drive letters, symlinks, and relative paths
+    resolved_dir = project_dir.resolve()
+    project_str = str(resolved_dir)

-    # New worktree location: .auto-claude/worktrees/tasks/{spec-name}/
-    new_worktree_marker = "/.auto-claude/worktrees/tasks/"
-    if new_worktree_marker in spec_str:
-        parent_path = spec_str.split(new_worktree_marker, 1)[0]
-        return True, parent_path
-
-    # Legacy worktree location: .worktrees/{spec-name}/
-    legacy_worktree_marker = "/.worktrees/"
-    if legacy_worktree_marker in spec_str:
-        parent_path = spec_str.split(legacy_worktree_marker, 1)[0]
-        return True, parent_path
+    for pattern in WORKTREE_PATH_PATTERNS:
+        match = re.search(pattern, project_str)
+        if match:
+            # Extract the parent project path (everything before the worktree marker)
+            parent_path = project_str[: match.start()]
+            # Handle edge case where worktree is at filesystem root
+            if not parent_path:
+                parent_path = resolved_dir.anchor
+            return True, Path(parent_path)

    return False, None


+def generate_worktree_isolation_warning(
+    project_dir: Path, parent_project_path: Path
+) -> str:
+    """
+    Generate the worktree isolation warning section for prompts.
+
+    This warning explicitly tells the agent that it's in an isolated worktree
+    and must NOT escape to the parent project directory.
+
+    Args:
+        project_dir: The worktree directory (agent's working directory)
+        parent_project_path: The forbidden parent project path
+
+    Returns:
+        Markdown string with isolation warning
+    """
+    return f"""## ⛔ ISOLATED WORKTREE - CRITICAL
+
+You are in an **ISOLATED GIT WORKTREE** - a complete copy of the project for safe development.
+
+**YOUR LOCATION:** `{project_dir}`
+**FORBIDDEN PATH:** `{parent_project_path}`
+
+### Rules:
+1. **NEVER** use `cd {parent_project_path}` or any path starting with `{parent_project_path}`
+2. **NEVER** use absolute paths that reference the parent project
+3. **ALL** project files exist HERE via relative paths
+
+### Why This Matters:
+- Git commits made in the parent project go to the WRONG branch
+- File changes in the parent project escape isolation
+- This defeats the entire purpose of safe, isolated development
+
+### Correct Usage:
+```bash
+# ✅ CORRECT - Use relative paths from your worktree
+./prod/src/file.ts
+./apps/frontend/src/component.tsx
+
+# ❌ WRONG - These escape isolation!
+cd {parent_project_path}
+{parent_project_path}/prod/src/file.ts
+```
+
+If you see absolute paths in spec.md or context.json that reference `{parent_project_path}`,
+convert them to relative paths from YOUR current location.
+
+---
+
+"""
+
+
 def get_relative_spec_path(spec_dir: Path, project_dir: Path) -> str:
    """
    Get the spec directory path relative to the project/working directory.
@@ -75,6 +138,7 @@ def generate_environment_context(project_dir: Path, spec_dir: Path) -> str:
    Generate environment context header for prompts.

    This explicitly tells the AI where it is working, preventing path confusion.
+    When running in a worktree, includes an isolation warning to prevent escaping.

    Args:
        project_dir: The working directory for the AI
@@ -85,14 +149,21 @@ def generate_environment_context(project_dir: Path, spec_dir: Path) -> str:
    """
    relative_spec = get_relative_spec_path(spec_dir, project_dir)

-    # Detect worktree mode and get forbidden parent path
-    is_worktree, forbidden_parent = detect_worktree_mode(spec_dir)
+    # Check if we're in an isolated worktree
+    is_worktree, parent_project_path = detect_worktree_isolation(project_dir)

-    # Build the environment context
-    context = f"""## YOUR ENVIRONMENT
+    # Start with worktree isolation warning if applicable
+    sections = []
+    if is_worktree and parent_project_path:
+        sections.append(
+            generate_worktree_isolation_warning(project_dir, parent_project_path)
+        )
+
+    sections.append(f"""## YOUR ENVIRONMENT

 **Working Directory:** `{project_dir}`
 **Spec Location:** `{relative_spec}/`
+{"**Isolation Mode:** WORKTREE (changes are isolated from main project)" if is_worktree else ""}

 Your filesystem is restricted to your working directory. All file paths should be
 relative to this location. Do NOT use absolute paths.
@@ -110,35 +181,9 @@ coder prompt for detailed examples.

 ---

-"""
+""")

-    # Add worktree isolation warning if in worktree mode
-    if is_worktree and forbidden_parent:
-        context += f"""## 🚨 ISOLATED WORKTREE - CRITICAL
-
-You are in an **ISOLATED GIT WORKTREE** - a complete copy of the project.
-
-**YOUR LOCATION:** `{project_dir}`
-**FORBIDDEN:** Do NOT use `cd {forbidden_parent}` or `cd ../..` - this **ESCAPES ISOLATION**
-
-All project files exist HERE via relative paths from your working directory.
-
-**CRITICAL RULES:**
-* **NEVER** `cd {forbidden_parent}` or any path traversal outside your worktree
-* **STAY** within your working directory at all times
-* **ALL** file operations use paths relative to your current location
-* Before any `cd` command, run `pwd` and verify the target is within your worktree
-
-**VIOLATION WARNING:** Escaping the worktree will cause:
-* Git commits going to wrong branch
-* Files created/modified in the wrong location
-* Breaking worktree isolation guarantees
-
---
-
-"""
-
-    return context
+    return "".join(sections)


 def generate_subtask_prompt(
@@ -81,6 +81,31 @@ def get_base_branch_from_metadata(spec_dir: Path) -> str | None:
    return None


+def get_use_local_branch_from_metadata(spec_dir: Path) -> bool:
+    """
+    Read useLocalBranch from task_metadata.json if it exists.
+
+    When True, the worktree should be created from the local branch directly
+    instead of preferring origin/branch. This preserves gitignored files
+    (.env, configs) that may not exist on the remote.
+
+    Args:
+        spec_dir: Directory containing the spec files
+
+    Returns:
+        True if useLocalBranch is set in metadata, False otherwise
+    """
+    metadata_path = spec_dir / "task_metadata.json"
+    if metadata_path.exists():
+        try:
+            with open(metadata_path, encoding="utf-8") as f:
+                metadata = json.load(f)
+                return bool(metadata.get("useLocalBranch", False))
+        except (json.JSONDecodeError, OSError):
+            pass
+    return False
+
+
 # Alias for backwards compatibility (internal use)
 _get_base_branch_from_metadata = get_base_branch_from_metadata

@@ -1,5 +1,7 @@
 # Auto-Build Framework Dependencies
-claude-agent-sdk>=0.1.19
+# SDK 0.1.25+ required for improved tool use concurrency handling
+# Earlier versions had 400 errors when tool_use blocks had partial failures
+claude-agent-sdk>=0.1.25
 python-dotenv>=1.0.0

 # TOML parsing fallback for Python < 3.11
@@ -10,6 +10,8 @@ Key Features:
 - Skips re-reviewing bot commits
 - Implements "cooling off" period to prevent rapid re-reviews
 - Tracks reviewed commits to avoid duplicate reviews
+- In-progress tracking to prevent concurrent reviews
+- Stale review detection with automatic cleanup

 Usage:
    detector = BotDetector(bot_token="ghp_...")
@@ -20,13 +22,22 @@ Usage:
        print(f"Skipping PR: {reason}")
        return

+    # Mark review as started (prevents concurrent reviews)
+    detector.mark_review_started(pr_number)
+
+    # Perform review...
+
    # After successful review, mark as reviewed
    detector.mark_reviewed(pr_number, head_sha)
+
+    # Or if review failed:
+    detector.mark_review_finished(pr_number, success=False)
 """

 from __future__ import annotations

 import json
+import logging
 import os
 import subprocess
 import sys
@@ -36,6 +47,8 @@ from pathlib import Path

 from core.gh_executable import get_gh_executable

+logger = logging.getLogger(__name__)
+
 try:
    from .file_lock import FileLock, atomic_write
 except (ImportError, ValueError, SystemError):
@@ -52,11 +65,15 @@ class BotDetectionState:
    # PR number -> last review timestamp (ISO format)
    last_review_times: dict[int, str] = field(default_factory=dict)

+    # PR number -> in-progress review start time (ISO format)
+    in_progress_reviews: dict[int, str] = field(default_factory=dict)
+
    def to_dict(self) -> dict:
        """Convert to dictionary for JSON serialization."""
        return {
            "reviewed_commits": self.reviewed_commits,
            "last_review_times": self.last_review_times,
+            "in_progress_reviews": self.in_progress_reviews,
        }

    @classmethod
@@ -65,6 +82,7 @@ class BotDetectionState:
        return cls(
            reviewed_commits=data.get("reviewed_commits", {}),
            last_review_times=data.get("last_review_times", {}),
+            in_progress_reviews=data.get("in_progress_reviews", {}),
        )

    def save(self, state_dir: Path) -> None:
@@ -101,11 +119,16 @@ class BotDetector:
        - 1-minute cooling off period between reviews of same PR (for testing)
        - Tracks reviewed commit SHAs to avoid duplicate reviews
        - Identifies bot user from token to skip bot-authored content
+        - In-progress tracking to prevent concurrent reviews
+        - Stale review detection (30-minute timeout)
    """

    # Cooling off period in minutes (reduced to 1 for testing large PRs)
    COOLING_OFF_MINUTES = 1

+    # Timeout for in-progress reviews in minutes (after this, review is considered stale/crashed)
+    IN_PROGRESS_TIMEOUT_MINUTES = 30
+
    def __init__(
        self,
        state_dir: Path,
@@ -298,6 +321,104 @@ class BotDetector:
        reviewed = self.state.reviewed_commits.get(str(pr_number), [])
        return commit_sha in reviewed

+    def is_review_in_progress(self, pr_number: int) -> tuple[bool, str]:
+        """
+        Check if a review is currently in progress for this PR.
+
+        Also detects stale reviews (started > IN_PROGRESS_TIMEOUT_MINUTES ago).
+
+        Args:
+            pr_number: The PR number
+
+        Returns:
+            Tuple of (is_in_progress, reason_message)
+        """
+        pr_key = str(pr_number)
+        start_time_str = self.state.in_progress_reviews.get(pr_key)
+
+        if not start_time_str:
+            return False, ""
+
+        try:
+            start_time = datetime.fromisoformat(start_time_str)
+            time_elapsed = datetime.now() - start_time
+
+            # Check if review is stale (timeout exceeded)
+            if time_elapsed > timedelta(minutes=self.IN_PROGRESS_TIMEOUT_MINUTES):
+                # Mark as stale and clear the in-progress state
+                print(
+                    f"[BotDetector] Review for PR #{pr_number} is stale "
+                    f"(started {int(time_elapsed.total_seconds() / 60)}m ago, "
+                    f"timeout: {self.IN_PROGRESS_TIMEOUT_MINUTES}m) - clearing in-progress state",
+                    file=sys.stderr,
+                )
+                self.mark_review_finished(pr_number, success=False)
+                return False, ""
+
+            # Review is actively in progress
+            minutes_elapsed = int(time_elapsed.total_seconds() / 60)
+            reason = f"Review already in progress (started {minutes_elapsed}m ago)"
+            print(f"[BotDetector] PR #{pr_number}: {reason}", file=sys.stderr)
+            return True, reason
+
+        except (ValueError, TypeError) as e:
+            print(
+                f"[BotDetector] Error parsing in-progress start time: {e}",
+                file=sys.stderr,
+            )
+            # Clear invalid state
+            self.mark_review_finished(pr_number, success=False)
+            return False, ""
+
+    def mark_review_started(self, pr_number: int) -> None:
+        """
+        Mark a review as started for this PR.
+
+        This should be called when beginning a review to prevent concurrent reviews.
+
+        Args:
+            pr_number: The PR number
+        """
+        pr_key = str(pr_number)
+
+        # Record start time
+        self.state.in_progress_reviews[pr_key] = datetime.now().isoformat()
+
+        # Save state
+        self.state.save(self.state_dir)
+
+        logger.info(f"[BotDetector] Marked PR #{pr_number} review as started")
+        print(f"[BotDetector] Started review for PR #{pr_number}", file=sys.stderr)
+
+    def mark_review_finished(self, pr_number: int, success: bool = True) -> None:
+        """
+        Mark a review as finished for this PR.
+
+        This clears the in-progress state. Should be called when review completes
+        (successfully or with error) or when detected as stale.
+
+        Args:
+            pr_number: The PR number
+            success: Whether the review completed successfully
+        """
+        pr_key = str(pr_number)
+
+        # Clear in-progress state
+        if pr_key in self.state.in_progress_reviews:
+            del self.state.in_progress_reviews[pr_key]
+
+            # Save state
+            self.state.save(self.state_dir)
+
+            status = "successfully" if success else "with error/timeout"
+            logger.info(
+                f"[BotDetector] Marked PR #{pr_number} review as finished ({status})"
+            )
+            print(
+                f"[BotDetector] Finished review for PR #{pr_number} ({status})",
+                file=sys.stderr,
+            )
+
    def should_skip_pr_review(
        self,
        pr_number: int,
@@ -332,13 +453,19 @@ class BotDetector:
                print(f"[BotDetector] SKIP PR #{pr_number}: {reason}")
                return True, reason

-        # Check 3: Are we in the cooling off period?
+        # Check 3: Is a review already in progress?
+        is_in_progress, reason = self.is_review_in_progress(pr_number)
+        if is_in_progress:
+            print(f"[BotDetector] SKIP PR #{pr_number}: {reason}")
+            return True, reason
+
+        # Check 4: Are we in the cooling off period?
        is_cooling, reason = self.is_within_cooling_off(pr_number)
        if is_cooling:
            print(f"[BotDetector] SKIP PR #{pr_number}: {reason}")
            return True, reason

-        # Check 4: Have we already reviewed this exact commit?
+        # Check 5: Have we already reviewed this exact commit?
        head_sha = self.get_last_commit_sha(commits) if commits else None
        if head_sha and self.has_reviewed_commit(pr_number, head_sha):
            reason = f"Already reviewed commit {head_sha[:8]}"
@@ -354,6 +481,7 @@ class BotDetector:
        Mark a PR as reviewed at a specific commit.

        This should be called after successfully posting a review.
+        Also clears the in-progress state.

        Args:
            pr_number: The PR number
@@ -371,10 +499,14 @@ class BotDetector:
        # Update last review time
        self.state.last_review_times[pr_key] = datetime.now().isoformat()

+        # Clear in-progress state
+        if pr_key in self.state.in_progress_reviews:
+            del self.state.in_progress_reviews[pr_key]
+
        # Save state
        self.state.save(self.state_dir)

-        print(
+        logger.info(
            f"[BotDetector] Marked PR #{pr_number} as reviewed at {commit_sha[:8]} "
            f"({len(self.state.reviewed_commits[pr_key])} total commits reviewed)"
        )
@@ -394,6 +526,9 @@ class BotDetector:
        if pr_key in self.state.last_review_times:
            del self.state.last_review_times[pr_key]

+        if pr_key in self.state.in_progress_reviews:
+            del self.state.in_progress_reviews[pr_key]
+
        self.state.save(self.state_dir)

        print(f"[BotDetector] Cleared state for PR #{pr_number}")
@@ -409,13 +544,16 @@ class BotDetector:
        total_reviews = sum(
            len(commits) for commits in self.state.reviewed_commits.values()
        )
+        in_progress_count = len(self.state.in_progress_reviews)

        return {
            "bot_username": self.bot_username,
            "review_own_prs": self.review_own_prs,
            "total_prs_tracked": total_prs,
            "total_reviews_performed": total_reviews,
+            "in_progress_reviews": in_progress_count,
            "cooling_off_minutes": self.COOLING_OFF_MINUTES,
+            "in_progress_timeout_minutes": self.IN_PROGRESS_TIMEOUT_MINUTES,
        }

    def cleanup_stale_prs(self, max_age_days: int = 30) -> int:
@@ -425,6 +563,9 @@ class BotDetector:
        This prevents unbounded growth of the state file by cleaning up
        entries for PRs that are likely closed/merged.

+        Also cleans up stale in-progress reviews (reviews that have been
+        in progress for longer than IN_PROGRESS_TIMEOUT_MINUTES).
+
        Args:
            max_age_days: Remove PRs not reviewed in this many days (default: 30)

@@ -432,8 +573,13 @@ class BotDetector:
            Number of PRs cleaned up
        """
        cutoff = datetime.now() - timedelta(days=max_age_days)
+        in_progress_cutoff = datetime.now() - timedelta(
+            minutes=self.IN_PROGRESS_TIMEOUT_MINUTES
+        )
        prs_to_remove: list[str] = []
+        stale_in_progress: list[str] = []

+        # Find stale reviewed PRs
        for pr_key, last_review_str in self.state.last_review_times.items():
            try:
                last_review = datetime.fromisoformat(last_review_str)
@@ -443,18 +589,43 @@ class BotDetector:
                # Invalid timestamp - mark for removal
                prs_to_remove.append(pr_key)

+        # Find stale in-progress reviews
+        for pr_key, start_time_str in self.state.in_progress_reviews.items():
+            try:
+                start_time = datetime.fromisoformat(start_time_str)
+                if start_time < in_progress_cutoff:
+                    stale_in_progress.append(pr_key)
+            except (ValueError, TypeError):
+                # Invalid timestamp - mark for removal
+                stale_in_progress.append(pr_key)
+
        # Remove stale PRs
        for pr_key in prs_to_remove:
            if pr_key in self.state.reviewed_commits:
                del self.state.reviewed_commits[pr_key]
            if pr_key in self.state.last_review_times:
                del self.state.last_review_times[pr_key]
+            if pr_key in self.state.in_progress_reviews:
+                del self.state.in_progress_reviews[pr_key]

-        if prs_to_remove:
+        # Remove stale in-progress reviews
+        for pr_key in stale_in_progress:
+            if pr_key in self.state.in_progress_reviews:
+                del self.state.in_progress_reviews[pr_key]
+
+        total_cleaned = len(prs_to_remove) + len(stale_in_progress)
+
+        if total_cleaned > 0:
            self.state.save(self.state_dir)
-            print(
-                f"[BotDetector] Cleaned up {len(prs_to_remove)} stale PRs "
-                f"(older than {max_age_days} days)"
-            )
+            if prs_to_remove:
+                print(
+                    f"[BotDetector] Cleaned up {len(prs_to_remove)} stale PRs "
+                    f"(older than {max_age_days} days)"
+                )
+            if stale_in_progress:
+                print(
+                    f"[BotDetector] Cleaned up {len(stale_in_progress)} stale in-progress reviews "
+                    f"(older than {self.IN_PROGRESS_TIMEOUT_MINUTES} minutes)"
+                )

-        return len(prs_to_remove)
+        return total_cleaned
@@ -19,7 +19,6 @@ from __future__ import annotations
 import ast
 import asyncio
 import json
-import os
 import re
 from dataclasses import dataclass, field
 from pathlib import Path
@@ -825,41 +824,11 @@ class PRContextGatherer:
        """
        Find files related to the changes.

-        This includes:
-        - Test files for changed source files
-        - Imported modules and dependencies
-        - Configuration files in the same directory
-        - Related type definition files
-        - Reverse dependencies (files that import changed files)
+        DEPRECATED: LLM agents now discover related files themselves using Read, Grep, and Glob tools.
+        This method returns an empty list - agents have domain expertise to find what's relevant.
        """
-        related = set()
-
-        for changed_file in changed_files:
-            path = Path(changed_file.path)
-
-            # Find test files
-            related.update(self._find_test_files(path))
-
-            # Find imported files (for supported languages)
-            if path.suffix in [".ts", ".tsx", ".js", ".jsx", ".py"]:
-                related.update(self._find_imports(changed_file.content, path))
-
-            # Find config files in same directory
-            related.update(self._find_config_files(path.parent))
-
-            # Find type definition files
-            if path.suffix in [".ts", ".tsx"]:
-                related.update(self._find_type_definitions(path))
-
-            # Find reverse dependencies (files that import this file)
-            related.update(self._find_dependents(changed_file.path))
-
-        # Remove files that are already in changed_files
-        changed_paths = {cf.path for cf in changed_files}
-        related = {r for r in related if r not in changed_paths}
-
-        # Use smart prioritization with increased limit (50 instead of 20)
-        return self._prioritize_related_files(related, limit=50)
+        # Return empty list - LLM agents will discover files via their tools
+        return []

    def _find_test_files(self, source_path: Path) -> set[str]:
        """Find test files related to a source file."""
@@ -1071,168 +1040,35 @@ class PRContextGatherer:
        """
        Find files that import the given file (reverse dependencies).

-        Uses pure Python to search for import statements referencing this file.
-        Cross-platform compatible (Windows, macOS, Linux).
-        Limited to prevent performance issues on large codebases.
+        DEPRECATED: LLM agents now discover reverse dependencies themselves using Grep and Read tools.
+        Returns empty set - agents can search the codebase with their domain expertise.

        Args:
            file_path: Path of the file to find dependents for
            max_results: Maximum number of dependents to return

        Returns:
-            Set of file paths that import this file.
+            Empty set - LLM agents will discover dependents via Grep tool.
        """
-        dependents: set[str] = set()
-        path_obj = Path(file_path)
-        stem = path_obj.stem  # e.g., 'helpers' from 'utils/helpers.ts'
-
-        # Skip if stem is too generic (would match too many files)
-        if stem in ["index", "main", "app", "utils", "helpers", "types", "constants"]:
-            return dependents
-
-        # Build regex patterns and file extensions based on file type
-        pattern = None
-        file_extensions = []
-
-        if path_obj.suffix in [".ts", ".tsx", ".js", ".jsx"]:
-            # Match various import styles for JS/TS
-            # from './helpers', from '../utils/helpers', from '@/utils/helpers'
-            # Escape stem for regex safety
-            escaped_stem = re.escape(stem)
-            pattern = re.compile(rf"['\"].*{escaped_stem}['\"]")
-            file_extensions = [".ts", ".tsx", ".js", ".jsx"]
-        elif path_obj.suffix == ".py":
-            # Match Python imports: from .helpers import, import helpers
-            escaped_stem = re.escape(stem)
-            pattern = re.compile(rf"(from.*{escaped_stem}|import.*{escaped_stem})")
-            file_extensions = [".py"]
-        else:
-            return dependents
-
-        # Directories to exclude
-        exclude_dirs = {
-            "node_modules",
-            ".git",
-            "dist",
-            "build",
-            "__pycache__",
-            ".venv",
-            "venv",
-        }
-
-        # Walk the project directory
-        project_path = Path(self.project_dir)
-        files_checked = 0
-        max_files_to_check = 2000  # Prevent infinite scanning on large codebases
-
-        try:
-            for root, dirs, files in os.walk(project_path):
-                # Modify dirs in-place to exclude certain directories
-                dirs[:] = [d for d in dirs if d not in exclude_dirs]
-
-                for filename in files:
-                    # Check if we've hit the file limit
-                    if files_checked >= max_files_to_check:
-                        safe_print(
-                            f"[Context] File limit reached finding dependents for {file_path}"
-                        )
-                        return dependents
-
-                    # Check if file has the right extension
-                    if not any(filename.endswith(ext) for ext in file_extensions):
-                        continue
-
-                    file_full_path = Path(root) / filename
-                    files_checked += 1
-
-                    # Get relative path from project root
-                    try:
-                        relative_path = file_full_path.relative_to(project_path)
-                        relative_path_str = str(relative_path).replace("\\", "/")
-
-                        # Don't include the file itself
-                        if relative_path_str == file_path:
-                            continue
-
-                        # Search for the pattern in the file
-                        try:
-                            with open(
-                                file_full_path, encoding="utf-8", errors="ignore"
-                            ) as f:
-                                content = f.read()
-                                if pattern.search(content):
-                                    dependents.add(relative_path_str)
-                                    if len(dependents) >= max_results:
-                                        return dependents
-                        except (OSError, UnicodeDecodeError):
-                            # Skip files that can't be read
-                            continue
-
-                    except ValueError:
-                        # File is not relative to project_path, skip it
-                        continue
-
-        except Exception as e:
-            safe_print(f"[Context] Error finding dependents: {e}")
-
-        return dependents
+        # Return empty set - LLM agents will use Grep to find importers when needed
+        return set()

    def _prioritize_related_files(self, files: set[str], limit: int = 50) -> list[str]:
        """
        Prioritize related files by relevance.

-        Priority order:
-        1. Test files (most important for review context)
-        2. Type definition files (.d.ts)
-        3. Configuration files
-        4. Direct imports/dependents
-        5. Other files
+        DEPRECATED: LLM agents now prioritize exploration based on their domain expertise.
+        Returns empty list since _find_related_files no longer populates files.

        Args:
            files: Set of file paths to prioritize
            limit: Maximum number of files to return

        Returns:
-            List of files sorted by priority, limited to `limit`.
+            Empty list - LLM agents handle prioritization via their tools.
        """
-        test_files = []
-        type_files = []
-        config_files = []
-        other_files = []
-
-        for f in files:
-            path = Path(f)
-            name_lower = path.name.lower()
-
-            # Test files
-            if (
-                ".test." in name_lower
-                or ".spec." in name_lower
-                or name_lower.startswith("test_")
-                or name_lower.endswith("_test.py")
-                or "__tests__" in f
-            ):
-                test_files.append(f)
-            # Type definition files
-            elif name_lower.endswith(".d.ts") or "types" in name_lower:
-                type_files.append(f)
-            # Config files
-            elif name_lower in [
-                n.lower() for n in CONFIG_FILE_NAMES
-            ] or name_lower.endswith((".config.js", ".config.ts", "rc", "rc.json")):
-                config_files.append(f)
-            else:
-                other_files.append(f)
-
-        # Sort within each category alphabetically for consistency, then combine
-        prioritized = (
-            sorted(test_files)
-            + sorted(type_files)
-            + sorted(config_files)
-            + sorted(other_files)
-        )
-
-        return prioritized[:limit]
+        # Return empty list - LLM agents will prioritize exploration themselves
+        return []

    def _load_json_safe(self, filename: str) -> dict | None:
        """
@@ -1460,59 +1296,18 @@ class PRContextGatherer:
        """
        Find files related to the changes using a specific project root.

-        This static method allows finding related files AFTER a worktree
-        has been created, ensuring files exist in the worktree filesystem.
+        DEPRECATED: LLM agents now discover related files themselves using Read, Grep, and Glob tools.
+        This method returns an empty list - agents have domain expertise to find what's relevant.

        Args:
            changed_files: List of changed files from the PR
            project_root: Path to search for related files (e.g., worktree path)

        Returns:
-            List of related file paths (relative to project root)
+            Empty list - LLM agents will discover files via their tools.
        """
-        related: set[str] = set()
-
-        for changed_file in changed_files:
-            path = Path(changed_file.path)
-
-            # Find test files
-            test_patterns = [
-                # Jest/Vitest patterns
-                path.parent / f"{path.stem}.test{path.suffix}",
-                path.parent / f"{path.stem}.spec{path.suffix}",
-                path.parent / "__tests__" / f"{path.name}",
-                # Python patterns
-                path.parent / f"test_{path.stem}.py",
-                path.parent / f"{path.stem}_test.py",
-                # Go patterns
-                path.parent / f"{path.stem}_test.go",
-            ]
-
-            for test_path in test_patterns:
-                full_path = project_root / test_path
-                if full_path.exists() and full_path.is_file():
-                    related.add(str(test_path))
-
-            # Find config files in same directory
-            for name in CONFIG_FILE_NAMES:
-                config_path = path.parent / name
-                full_path = project_root / config_path
-                if full_path.exists() and full_path.is_file():
-                    related.add(str(config_path))
-
-            # Find type definition files
-            if path.suffix in [".ts", ".tsx"]:
-                type_def = path.parent / f"{path.stem}.d.ts"
-                full_path = project_root / type_def
-                if full_path.exists() and full_path.is_file():
-                    related.add(str(type_def))
-
-        # Remove files that are already in changed_files
-        changed_paths = {cf.path for cf in changed_files}
-        related = {r for r in related if r not in changed_paths}
-
-        # Limit to 50 most relevant files (increased from 20)
-        return sorted(related)[:50]
+        # Return empty list - LLM agents will discover files via their tools
+        return []


 class FollowupContextGatherer:
@@ -367,12 +367,21 @@ class PRReviewFinding:
    validation_evidence: str | None = None  # Code snippet examined during validation
    validation_explanation: str | None = None  # Why finding was validated/dismissed

-    # Cross-validation and confidence routing fields
-    confidence: float = 0.5  # Confidence score (0.0-1.0), defaults to medium confidence
+    # Cross-validation fields
+    # NOTE: confidence field is DEPRECATED - we use evidence-based validation, not confidence scores
+    # The finding-validator determines validity by examining actual code, not by confidence thresholds
+    confidence: float = 0.5  # DEPRECATED: No longer used for filtering
    source_agents: list[str] = field(
        default_factory=list
    )  # Which agents reported this finding
-    cross_validated: bool = False  # Whether multiple agents agreed on this finding
+    cross_validated: bool = (
+        False  # Whether multiple agents agreed on this finding (signal, not filter)
+    )
+
+    # Impact finding flag - indicates this finding is about code OUTSIDE the PR's changed files
+    # (e.g., callers affected by contract changes). Used by _is_finding_in_scope() to allow
+    # findings about related files that aren't directly in the PR diff.
+    is_impact_finding: bool = False

    def to_dict(self) -> dict:
        return {
@@ -398,6 +407,8 @@ class PRReviewFinding:
            "confidence": self.confidence,
            "source_agents": self.source_agents,
            "cross_validated": self.cross_validated,
+            # Impact finding flag
+            "is_impact_finding": self.is_impact_finding,
        }

    @classmethod
@@ -425,6 +436,8 @@ class PRReviewFinding:
            confidence=data.get("confidence", 0.5),
            source_agents=data.get("source_agents", []),
            cross_validated=data.get("cross_validated", False),
+            # Impact finding flag
+            is_impact_finding=data.get("is_impact_finding", False),
        )


@@ -396,9 +396,15 @@ class GitHubOrchestrator:
                        # No existing review found, create skip result
                        return await self._create_skip_result(pr_number, skip_reason)
                else:
-                    # For other skip reasons (bot-authored, cooling off), create a skip result
+                    # For other skip reasons (bot-authored, cooling off, in-progress), create a skip result
                    return await self._create_skip_result(pr_number, skip_reason)

+            # Mark review as started (prevents concurrent reviews)
+            self.bot_detector.mark_review_started(pr_number)
+            safe_print(
+                f"[BOT DETECTION] Marked PR #{pr_number} review as started", flush=True
+            )
+
            self._report_progress(
                "analyzing", 30, "Running multi-pass review...", pr_number=pr_number
            )
@@ -572,6 +578,13 @@ class GitHubOrchestrator:
        except Exception as e:
            import traceback

+            # Mark review as finished with error
+            self.bot_detector.mark_review_finished(pr_number, success=False)
+            safe_print(
+                f"[BOT DETECTION] Marked PR #{pr_number} review as finished (error)",
+                flush=True,
+            )
+
            # Log full exception details for debugging
            error_details = f"{type(e).__name__}: {e}"
            full_traceback = traceback.format_exc()
@@ -634,6 +647,13 @@ class GitHubOrchestrator:
            pr_number=pr_number,
        )

+        # Mark review as started (prevents concurrent reviews)
+        self.bot_detector.mark_review_started(pr_number)
+        safe_print(
+            f"[BOT DETECTION] Marked PR #{pr_number} follow-up review as started",
+            flush=True,
+        )
+
        try:
            # Import here to avoid circular imports at module level
            try:
@@ -956,6 +976,13 @@ class GitHubOrchestrator:
            return result

        except Exception as e:
+            # Mark review as finished with error
+            self.bot_detector.mark_review_finished(pr_number, success=False)
+            safe_print(
+                f"[BOT DETECTION] Marked PR #{pr_number} follow-up review as finished (error)",
+                flush=True,
+            )
+
            result = PRReviewResult(
                pr_number=pr_number,
                repo=self.config.repo,
@@ -22,30 +22,6 @@ except (ImportError, ValueError, SystemError):
 class FindingValidator:
    """Validates and filters AI-generated PR review findings."""

-    # Vague patterns that indicate low-quality findings
-    VAGUE_PATTERNS = [
-        "could be improved",
-        "consider using",
-        "might want to",
-        "you may want",
-        "it would be better",
-        "possibly consider",
-        "perhaps use",
-        "potentially add",
-        "you should consider",
-        "it might be good",
-    ]
-
-    # Generic suggestions without specifics
-    GENERIC_PATTERNS = [
-        "improve this",
-        "fix this",
-        "change this",
-        "update this",
-        "refactor this",
-        "review this",
-    ]
-
    # Minimum lengths for quality checks
    MIN_DESCRIPTION_LENGTH = 30
    MIN_SUGGESTED_FIX_LENGTH = 20
@@ -123,10 +99,6 @@ class FindingValidator:
            # Update the finding with corrected line
            finding.line = corrected.line

-        # Check for false positives
-        if self._is_false_positive(finding):
-            return False
-
        # Check confidence threshold
        if not self._meets_confidence_threshold(finding):
            return False
@@ -294,51 +266,6 @@ class FindingValidator:

        return finding

-    def _is_false_positive(self, finding: PRReviewFinding) -> bool:
-        """
-        Detect likely false positives.
-
-        Args:
-            finding: Finding to check
-
-        Returns:
-            True if likely a false positive, False otherwise
-        """
-        description_lower = finding.description.lower()
-
-        # Check for vague descriptions
-        for pattern in self.VAGUE_PATTERNS:
-            if pattern in description_lower:
-                # Vague low/medium findings are likely FPs
-                if finding.severity in [ReviewSeverity.LOW, ReviewSeverity.MEDIUM]:
-                    return True
-
-        # Check for generic suggestions
-        for pattern in self.GENERIC_PATTERNS:
-            if pattern in description_lower:
-                if finding.severity == ReviewSeverity.LOW:
-                    return True
-
-        # Check for generic suggestions without specifics
-        if (
-            not finding.suggested_fix
-            or len(finding.suggested_fix) < self.MIN_SUGGESTED_FIX_LENGTH
-        ):
-            if finding.severity == ReviewSeverity.LOW:
-                return True
-
-        # Check for style findings without clear justification
-        if finding.category.value == "style":
-            # Style findings should have good suggestions
-            if not finding.suggested_fix or len(finding.suggested_fix) < 30:
-                return True
-
-        # Check for overly short descriptions
-        if len(finding.description) < 50 and finding.severity == ReviewSeverity.LOW:
-            return True
-
-        return False
-
    def _score_actionability(self, finding: PRReviewFinding) -> float:
        """
        Score how actionable a finding is (0.0 to 1.0).
@@ -0,0 +1,33 @@
+"""
+Agent Utilities
+===============
+
+Shared utility functions for GitHub PR review agents.
+"""
+
+from pathlib import Path
+
+
+def create_working_dir_injector(working_dir: Path):
+    """Factory that creates a prompt injector with working directory context.
+
+    Args:
+        working_dir: The working directory path to inject into prompts
+
+    Returns:
+        A function that takes (prompt, fallback) and returns the prompt with
+        working directory prefix prepended.
+    """
+    working_dir_prefix = (
+        f"## Working Directory\n\n"
+        f"Your working directory is: `{working_dir.resolve()}`\n"
+        f"All file paths should be relative to this directory.\n"
+        f"Use the Read, Grep, and Glob tools to examine files.\n\n"
+    )
+
+    def with_working_dir(prompt: str | None, fallback: str) -> str:
+        """Inject working directory context into agent prompt."""
+        base = prompt or fallback
+        return f"{working_dir_prefix}{base}"
+
+    return with_working_dir
@@ -43,6 +43,7 @@ try:
        PRReviewResult,
        ReviewSeverity,
    )
+    from .agent_utils import create_working_dir_injector
    from .category_utils import map_category
    from .io_utils import safe_print
    from .pr_worktree_manager import PRWorktreeManager
@@ -62,6 +63,7 @@ except (ImportError, ValueError, SystemError):
        ReviewSeverity,
    )
    from phase_config import get_thinking_budget, resolve_model_id
+    from services.agent_utils import create_working_dir_injector
    from services.category_utils import map_category
    from services.io_utils import safe_print
    from services.pr_worktree_manager import PRWorktreeManager
@@ -183,22 +185,35 @@ class ParallelFollowupReviewer:
        """
        self.worktree_manager.remove_worktree(worktree_path)

-    def _define_specialist_agents(self) -> dict[str, AgentDefinition]:
+    def _define_specialist_agents(
+        self, project_root: Path | None = None
+    ) -> dict[str, AgentDefinition]:
        """
        Define specialist agents for follow-up review.

        Each agent has:
        - description: When the orchestrator should invoke this agent
-        - prompt: System prompt for the agent
+        - prompt: System prompt for the agent (includes working directory)
        - tools: Tools the agent can use (read-only for PR review)
        - model: "inherit" = use same model as orchestrator (user's choice)
+
+        Args:
+            project_root: Working directory for the agents (worktree path).
+                         If None, falls back to self.project_dir.
        """
+        # Use provided project_root or fall back to default
+        working_dir = project_root or self.project_dir
+
        # Load agent prompts from files
        resolution_prompt = self._load_prompt("pr_followup_resolution_agent.md")
        newcode_prompt = self._load_prompt("pr_followup_newcode_agent.md")
        comment_prompt = self._load_prompt("pr_followup_comment_agent.md")
        validator_prompt = self._load_prompt("pr_finding_validator.md")

+        # CRITICAL: Inject working directory into all prompts
+        # Subagents don't inherit cwd from parent, so they need explicit path info
+        with_working_dir = create_working_dir_injector(working_dir)
+
        return {
            "resolution-verifier": AgentDefinition(
                description=(
@@ -207,8 +222,10 @@ class ParallelFollowupReviewer:
                    "are truly fixed, partially fixed, or still unresolved. "
                    "Invoke when: There are previous findings to verify."
                ),
-                prompt=resolution_prompt
-                or "You verify whether previous findings are resolved.",
+                prompt=with_working_dir(
+                    resolution_prompt,
+                    "You verify whether previous findings are resolved.",
+                ),
                tools=["Read", "Grep", "Glob"],
                model="inherit",
            ),
@@ -219,7 +236,9 @@ class ParallelFollowupReviewer:
                    "Invoke when: There are substantial code changes (>50 lines diff) or "
                    "changes to security-sensitive areas."
                ),
-                prompt=newcode_prompt or "You review new code for issues.",
+                prompt=with_working_dir(
+                    newcode_prompt, "You review new code for issues."
+                ),
                tools=["Read", "Grep", "Glob"],
                model="inherit",
            ),
@@ -230,7 +249,9 @@ class ParallelFollowupReviewer:
                    "unanswered questions and valid concerns. "
                    "Invoke when: There are comments or formal reviews since last review."
                ),
-                prompt=comment_prompt or "You analyze comments and feedback.",
+                prompt=with_working_dir(
+                    comment_prompt, "You analyze comments and feedback."
+                ),
                tools=["Read", "Grep", "Glob"],
                model="inherit",
            ),
@@ -243,8 +264,10 @@ class ParallelFollowupReviewer:
                    "CRITICAL: Invoke for ALL unresolved findings after resolution-verifier runs. "
                    "Invoke when: There are findings marked as unresolved that need validation."
                ),
-                prompt=validator_prompt
-                or "You validate whether unresolved findings are real issues.",
+                prompt=with_working_dir(
+                    validator_prompt,
+                    "You validate whether unresolved findings are real issues.",
+                ),
                tools=["Read", "Grep", "Glob"],
                model="inherit",
            ),
@@ -487,6 +510,9 @@ The SDK will run invoked agents in parallel automatically.
                    "using local checkout"
                )

+            # Capture agent definitions for debug logging (AFTER worktree creation)
+            agent_defs = self._define_specialist_agents(project_root)
+
            # Use model and thinking level from config (user settings)
            # Resolve model shorthand via environment variable override if configured
            model_shorthand = self.config.model or "sonnet"
@@ -499,14 +525,14 @@ The SDK will run invoked agents in parallel automatically.
                f"thinking_level={thinking_level}, thinking_budget={thinking_budget}"
            )

-            # Create client with subagents defined
+            # Create client with subagents defined (using worktree path)
            client = create_client(
                project_dir=project_root,
                spec_dir=self.github_dir,
                model=model,
                agent_type="pr_followup_parallel",
                max_thinking_tokens=thinking_budget,
-                agents=self._define_specialist_agents(),
+                agents=self._define_specialist_agents(project_root),
                output_format={
                    "type": "json_schema",
                    "schema": ParallelFollowupResponse.model_json_schema(),
@@ -534,6 +560,8 @@ The SDK will run invoked agents in parallel automatically.
                    client=client,
                    context_name="ParallelFollowup",
                    model=model,
+                    system_prompt=prompt,
+                    agent_definitions=agent_defs,
                )

                # Check for stream processing errors
@@ -883,6 +911,7 @@ The SDK will run invoked agents in parallel automatically.
                                    validation_status=validation_status,
                                    validation_evidence=validation_evidence,
                                    validation_explanation=validation_explanation,
+                                    is_impact_finding=original.is_impact_finding,
                                )
                            )

@@ -903,6 +932,7 @@ The SDK will run invoked agents in parallel automatically.
                        line=nf.line,
                        suggested_fix=nf.suggested_fix,
                        fixable=nf.fixable,
+                        is_impact_finding=getattr(nf, "is_impact_finding", False),
                    )
                )

@@ -139,18 +139,9 @@ class PRReviewEngine:
            files_list.append(f"- ... and {len(context.changed_files) - 20} more files")
        files_str = "\n".join(files_list)

-        # NEW: Format related files (imports, tests, etc.)
+        # Removed: Related files section
+        # LLM agents now discover relevant files themselves via Read, Grep, Glob tools
        related_files_str = ""
-        if context.related_files:
-            related_files_list = [f"- `{f}`" for f in context.related_files[:10]]
-            if len(context.related_files) > 10:
-                related_files_list.append(
-                    f"- ... and {len(context.related_files) - 10} more"
-                )
-            related_files_str = f"""
-### Related Files (imports, tests, configs)
-{chr(10).join(related_files_list)}
-"""

        # NEW: Format commits for context
        commits_str = ""
@@ -28,6 +28,50 @@ from typing import Literal

 from pydantic import BaseModel, Field

+# =============================================================================
+# Verification Evidence (Required for All Findings)
+# =============================================================================
+
+
+class VerificationEvidence(BaseModel):
+    """Evidence that a finding was verified against actual code.
+
+    All fields are required - schema enforcement guarantees evidence exists.
+    This shifts quality control from programmatic filters to schema enforcement.
+    """
+
+    code_examined: str = Field(
+        min_length=1,
+        description=(
+            "REQUIRED: Exact code snippet that was examined. "
+            "Must be actual code from the file, not a description of code. "
+            "Copy-paste the relevant lines directly."
+        ),
+    )
+    line_range_examined: list[int] = Field(
+        min_length=2,
+        max_length=2,
+        description=(
+            "Start and end line numbers [start, end] of the examined code. "
+            "Must match the code in code_examined."
+        ),
+    )
+    verification_method: Literal[
+        "direct_code_inspection",
+        "cross_file_trace",
+        "test_verification",
+        "dependency_analysis",
+    ] = Field(
+        description=(
+            "How the issue was verified: "
+            "direct_code_inspection = found issue directly in the code shown; "
+            "cross_file_trace = traced through imports/calls to find the issue; "
+            "test_verification = verified through examination of test code; "
+            "dependency_analysis = verified through analyzing dependencies"
+        )
+    )
+
+
 # =============================================================================
 # Common Finding Types
 # =============================================================================
@@ -48,7 +92,10 @@ class BaseFinding(BaseModel):
    fixable: bool = Field(False, description="Whether this can be auto-fixed")
    evidence: str | None = Field(
        None,
-        description="Actual code snippet proving the issue exists. Required for validation.",
+        description="DEPRECATED: Use verification.code_examined instead. Will be removed in Phase 5.",
+    )
+    verification: VerificationEvidence = Field(
+        description="Evidence that this finding was verified against actual code"
    )


@@ -155,6 +202,9 @@ class FollowupFinding(BaseModel):
    line: int = Field(0, description="Line number of the issue")
    suggested_fix: str | None = Field(None, description="How to fix this issue")
    fixable: bool = Field(False, description="Whether this can be auto-fixed")
+    verification: VerificationEvidence = Field(
+        description="Evidence that this finding was verified against actual code"
+    )


 class FollowupReviewResponse(BaseModel):
@@ -323,7 +373,10 @@ class OrchestratorFinding(BaseModel):
    suggestion: str | None = Field(None, description="How to fix this issue")
    evidence: str | None = Field(
        None,
-        description="Actual code snippet proving the issue exists. Required for validation.",
+        description="DEPRECATED: Use verification.code_examined instead. Will be removed in Phase 5.",
+    )
+    verification: VerificationEvidence = Field(
+        description="Evidence that this finding was verified against actual code"
    )


@@ -399,7 +452,25 @@ class ParallelOrchestratorFinding(BaseModel):
    )
    evidence: str | None = Field(
        None,
-        description="Actual code snippet proving the issue exists. Required for validation.",
+        description="DEPRECATED: Use verification.code_examined instead. Will be removed in Phase 5.",
+    )
+    verification: VerificationEvidence = Field(
+        description="Evidence that this finding was verified against actual code"
+    )
+    is_impact_finding: bool = Field(
+        False,
+        description=(
+            "True if this finding is about impact on OTHER files (not the changed file). "
+            "Impact findings may reference files outside the PR's changed files list."
+        ),
+    )
+    checked_for_handling_elsewhere: bool = Field(
+        False,
+        description=(
+            "For 'missing X' claims (missing error handling, missing validation, etc.), "
+            "True if the agent verified X is not handled elsewhere in the codebase. "
+            "False if this is a 'missing X' claim but other locations were not checked."
+        ),
    )
    suggested_fix: str | None = Field(None, description="How to fix this issue")
    fixable: bool = Field(False, description="Whether this can be auto-fixed")
@@ -428,6 +499,89 @@ class AgentAgreement(BaseModel):
    )


+class DismissedFinding(BaseModel):
+    """A finding that was validated and dismissed as a false positive.
+
+    Included in output for transparency - users can see what was investigated and why it was dismissed.
+    """
+
+    id: str = Field(description="Original finding ID")
+    original_title: str = Field(description="Original finding title")
+    original_severity: Literal["critical", "high", "medium", "low"] = Field(
+        description="Original severity assigned by specialist"
+    )
+    original_file: str = Field(description="File where issue was claimed")
+    original_line: int = Field(0, description="Line where issue was claimed")
+    dismissal_reason: str = Field(
+        description="Why this finding was dismissed as a false positive"
+    )
+    validation_evidence: str = Field(
+        description="Actual code examined that disproved the finding"
+    )
+
+
+class ValidationSummary(BaseModel):
+    """Summary of validation results for transparency."""
+
+    total_findings_from_specialists: int = Field(
+        description="Total findings reported by all specialist agents"
+    )
+    confirmed_valid: int = Field(
+        description="Findings confirmed as real issues by validator"
+    )
+    dismissed_false_positive: int = Field(
+        description="Findings dismissed as false positives by validator"
+    )
+    needs_human_review: int = Field(
+        0, description="Findings that couldn't be definitively validated"
+    )
+
+
+class SpecialistFinding(BaseModel):
+    """A finding from a specialist agent (used in parallel SDK sessions)."""
+
+    severity: Literal["critical", "high", "medium", "low"] = Field(
+        description="Issue severity level"
+    )
+    category: Literal[
+        "security", "quality", "logic", "performance", "pattern", "test", "docs"
+    ] = Field(description="Issue category")
+    title: str = Field(description="Brief issue title (max 80 chars)")
+    description: str = Field(description="Detailed explanation of the issue")
+    file: str = Field(description="File path where issue was found")
+    line: int = Field(0, description="Line number of the issue")
+    end_line: int | None = Field(None, description="End line number if multi-line")
+    suggested_fix: str | None = Field(None, description="How to fix this issue")
+    evidence: str = Field(
+        min_length=1,
+        description="Actual code snippet examined that shows the issue. Required.",
+    )
+    is_impact_finding: bool = Field(
+        False,
+        description="True if this is about affected code outside the PR (callers, dependencies)",
+    )
+
+
+class SpecialistResponse(BaseModel):
+    """Response schema for individual specialist agent (parallel SDK sessions).
+
+    Used when each specialist runs as its own SDK session rather than via Task tool.
+    """
+
+    specialist_name: str = Field(
+        description="Name of the specialist (security, quality, logic, codebase-fit)"
+    )
+    analysis_summary: str = Field(description="Brief summary of what was analyzed")
+    files_examined: list[str] = Field(
+        default_factory=list,
+        description="List of files that were examined",
+    )
+    findings: list[SpecialistFinding] = Field(
+        default_factory=list,
+        description="Issues found during analysis",
+    )
+
+
 class ParallelOrchestratorResponse(BaseModel):
    """Complete response schema for parallel orchestrator PR review."""

@@ -438,8 +592,20 @@ class ParallelOrchestratorResponse(BaseModel):
        default_factory=list,
        description="List of agent names that were invoked",
    )
+    validation_summary: ValidationSummary | None = Field(
+        None,
+        description="Summary of validation results (total, confirmed, dismissed, needs_review)",
+    )
    findings: list[ParallelOrchestratorFinding] = Field(
-        default_factory=list, description="All findings from synthesis"
+        default_factory=list,
+        description="Validated findings only (confirmed_valid or needs_human_review)",
+    )
+    dismissed_findings: list[DismissedFinding] = Field(
+        default_factory=list,
+        description=(
+            "Findings that were validated and dismissed as false positives. "
+            "Included for transparency - users can see what was investigated."
+        ),
    )
    agent_agreement: AgentAgreement = Field(
        default_factory=AgentAgreement,
@@ -495,7 +661,10 @@ class ParallelFollowupFinding(BaseModel):
    )
    evidence: str | None = Field(
        None,
-        description="Actual code snippet proving the issue exists. Required for validation.",
+        description="DEPRECATED: Use verification.code_examined instead. Will be removed in Phase 5.",
+    )
+    verification: VerificationEvidence = Field(
+        description="Evidence that this finding was verified against actual code"
    )
    suggested_fix: str | None = Field(None, description="How to fix this issue")
    fixable: bool = Field(False, description="Whether this can be auto-fixed")
@@ -505,6 +674,14 @@ class ParallelFollowupFinding(BaseModel):
    related_to_previous: str | None = Field(
        None, description="ID of related previous finding if this is a regression"
    )
+    is_impact_finding: bool = Field(
+        False,
+        description=(
+            "True if this finding is about impact on OTHER files (callers, dependents) "
+            "outside the PR's changed files. Used by _is_finding_in_scope() to allow "
+            "findings about related files that aren't directly in the PR diff."
+        ),
+    )


 class CommentAnalysis(BaseModel):
@@ -594,7 +594,7 @@ Focus on:
 - Insecure cryptography
 - Input validation issues

-Output findings in JSON format with high confidence (>80%) only.
+Output findings in JSON format with evidence from the actual code.
 """


@@ -611,5 +611,5 @@ Focus on:
 - Pattern adherence
 - Maintainability

-Output findings in JSON format with high confidence (>80%) only.
+Output findings in JSON format with evidence from the actual code.
 """
@@ -124,6 +124,49 @@ def _get_tool_detail(tool_name: str, tool_input: dict[str, Any]) -> str:
    return f"Using tool: {tool_name}"


+# Circuit breaker threshold - abort if message count exceeds this
+# Prevents runaway retry loops from consuming unbounded resources
+MAX_MESSAGE_COUNT = 500
+
+
+def _is_tool_concurrency_error(text: str) -> bool:
+    """
+    Detect the specific tool use concurrency error pattern.
+
+    This error occurs when Claude makes multiple parallel tool_use blocks
+    and some fail, corrupting the tool_use/tool_result message pairing.
+
+    Args:
+        text: Text to check for error pattern
+
+    Returns:
+        True if this is the tool concurrency error, False otherwise
+    """
+    text_lower = text.lower()
+    # Check for the specific error message pattern
+    # Pattern 1: Explicit concurrency or tool_use errors with 400
+    has_400 = "400" in text_lower
+    has_tool = "tool" in text_lower
+
+    if has_400 and has_tool:
+        # Look for specific keywords indicating tool concurrency issues
+        error_keywords = [
+            "concurrency",
+            "tool_use",
+            "tool use",
+            "tool_result",
+            "tool result",
+        ]
+        if any(keyword in text_lower for keyword in error_keywords):
+            return True
+
+    # Pattern 2: API error with 400 and tool mention
+    if "api error" in text_lower and has_400 and has_tool:
+        return True
+
+    return False
+
+
 async def process_sdk_stream(
    client: Any,
    on_thinking: Callable[[str], None] | None = None,
@@ -133,6 +176,10 @@ async def process_sdk_stream(
    on_structured_output: Callable[[dict[str, Any]], None] | None = None,
    context_name: str = "SDK",
    model: str | None = None,
+    max_messages: int | None = None,
+    # Deprecated parameters (kept for backwards compatibility, no longer used)
+    system_prompt: str | None = None,  # noqa: ARG001
+    agent_definitions: dict | None = None,  # noqa: ARG001
 ) -> dict[str, Any]:
    """
    Process SDK response stream with customizable callbacks.
@@ -153,6 +200,7 @@ async def process_sdk_stream(
        on_structured_output: Callback for structured output - receives dict
        context_name: Name for logging (e.g., "ParallelOrchestrator", "ParallelFollowup")
        model: Model name for logging (e.g., "claude-sonnet-4-5-20250929")
+        max_messages: Optional override for max message count circuit breaker (default: MAX_MESSAGE_COUNT)

    Returns:
        Dictionary with:
@@ -171,6 +219,11 @@ async def process_sdk_stream(
    # Track subagent tool IDs to log their results
    subagent_tool_ids: dict[str, str] = {}  # tool_id -> agent_name
    completed_agent_tool_ids: set[str] = set()  # tool_ids of completed agents
+    # Track tool concurrency errors for retry logic
+    detected_concurrency_error = False
+
+    # Circuit breaker: max messages before aborting
+    message_limit = max_messages if max_messages is not None else MAX_MESSAGE_COUNT

    safe_print(f"[{context_name}] Processing SDK stream...")
    if DEBUG_MODE:
@@ -186,6 +239,17 @@ async def process_sdk_stream(
                msg_type = type(msg).__name__
                msg_count += 1

+                # CIRCUIT BREAKER: Abort if message count exceeds threshold
+                # This prevents runaway retry loops (e.g., 400 errors causing infinite retries)
+                if msg_count > message_limit:
+                    stream_error = (
+                        f"Circuit breaker triggered: message count ({msg_count}) "
+                        f"exceeded limit ({message_limit}). Possible retry loop detected."
+                    )
+                    logger.error(f"[{context_name}] {stream_error}")
+                    safe_print(f"[{context_name}] ERROR: {stream_error}")
+                    break
+
                # Log progress periodically so user knows AI is working
                if msg_count - last_progress_log >= PROGRESS_LOG_INTERVAL:
                    if subagent_tool_ids:
@@ -258,6 +322,16 @@ async def process_sdk_stream(
                        safe_print(
                            f"[{context_name}] Invoking agent: {agent_name}{model_info}"
                        )
+                        # Log delegation prompt for debugging trigger system
+                        delegation_prompt = tool_input.get("prompt", "")
+                        if delegation_prompt:
+                            # Show first 300 chars of delegation prompt
+                            prompt_preview = delegation_prompt[:300]
+                            if len(delegation_prompt) > 300:
+                                prompt_preview += "..."
+                            safe_print(
+                                f"[{context_name}] Delegation prompt for {agent_name}: {prompt_preview}"
+                            )
                    elif tool_name != "StructuredOutput":
                        # Log meaningful tool info (not just tool name)
                        tool_detail = _get_tool_detail(tool_name, tool_input)
@@ -349,6 +423,15 @@ async def process_sdk_stream(
                        block_type = type(block).__name__
                        if block_type == "TextBlock" and hasattr(block, "text"):
                            result_text += block.text
+                            # Check for tool concurrency error pattern in text output
+                            if _is_tool_concurrency_error(block.text):
+                                detected_concurrency_error = True
+                                logger.warning(
+                                    f"[{context_name}] Detected tool use concurrency error in response"
+                                )
+                                safe_print(
+                                    f"[{context_name}] WARNING: Tool concurrency error detected"
+                                )
                            # Always print text content preview (not just in DEBUG_MODE)
                            text_preview = block.text[:500].replace("\n", " ").strip()
                            if text_preview:
@@ -465,6 +548,13 @@ async def process_sdk_stream(

    safe_print(f"[{context_name}] Session ended. Total messages: {msg_count}")

+    # Set error flag if tool concurrency error was detected
+    if detected_concurrency_error and not stream_error:
+        stream_error = "tool_use_concurrency_error"
+        logger.warning(
+            f"[{context_name}] Tool use concurrency error detected - caller should retry"
+        )
+
    return {
        "result_text": result_text,
        "structured_output": structured_output,
@@ -532,5 +532,176 @@ class TestGhExecutableDetection:
            mock_run.assert_not_called()


+class TestInProgressTracking:
+    """Test in-progress review tracking."""
+
+    def test_mark_review_started(self, mock_bot_detector, temp_state_dir):
+        """Test marking review as started."""
+        mock_bot_detector.mark_review_started(123)
+
+        # Check state
+        assert "123" in mock_bot_detector.state.in_progress_reviews
+        start_time_str = mock_bot_detector.state.in_progress_reviews["123"]
+        start_time = datetime.fromisoformat(start_time_str)
+
+        # Should be very recent (within last 5 seconds)
+        time_diff = datetime.now() - start_time
+        assert time_diff.total_seconds() < 5
+
+        # Check persistence
+        loaded = BotDetectionState.load(temp_state_dir)
+        assert "123" in loaded.in_progress_reviews
+
+    def test_mark_review_finished_success(self, mock_bot_detector, temp_state_dir):
+        """Test marking review as finished successfully."""
+        mock_bot_detector.mark_review_started(123)
+        assert "123" in mock_bot_detector.state.in_progress_reviews
+
+        mock_bot_detector.mark_review_finished(123, success=True)
+
+        # In-progress state should be cleared
+        assert "123" not in mock_bot_detector.state.in_progress_reviews
+
+        # Check persistence
+        loaded = BotDetectionState.load(temp_state_dir)
+        assert "123" not in loaded.in_progress_reviews
+
+    def test_mark_review_finished_error(self, mock_bot_detector):
+        """Test marking review as finished with error."""
+        mock_bot_detector.mark_review_started(123)
+        mock_bot_detector.mark_review_finished(123, success=False)
+
+        # In-progress state should be cleared
+        assert "123" not in mock_bot_detector.state.in_progress_reviews
+
+    def test_is_review_in_progress_active(self, mock_bot_detector):
+        """Test detecting active in-progress review."""
+        mock_bot_detector.mark_review_started(123)
+
+        is_in_progress, reason = mock_bot_detector.is_review_in_progress(123)
+
+        assert is_in_progress is True
+        assert "already in progress" in reason.lower()
+
+    def test_is_review_in_progress_not_started(self, mock_bot_detector):
+        """Test checking in-progress when review not started."""
+        is_in_progress, reason = mock_bot_detector.is_review_in_progress(999)
+
+        assert is_in_progress is False
+        assert reason == ""
+
+    def test_is_review_in_progress_stale(self, mock_bot_detector):
+        """Test detecting stale in-progress review."""
+        # Set review start time to 31 minutes ago (past timeout)
+        stale_time = datetime.now() - timedelta(minutes=31)
+        mock_bot_detector.state.in_progress_reviews["123"] = stale_time.isoformat()
+
+        is_in_progress, reason = mock_bot_detector.is_review_in_progress(123)
+
+        # Should detect as stale and clear it
+        assert is_in_progress is False
+        assert reason == ""
+        # Should be removed from state
+        assert "123" not in mock_bot_detector.state.in_progress_reviews
+
+    def test_is_review_in_progress_invalid_timestamp(self, mock_bot_detector):
+        """Test handling invalid timestamp in in-progress state."""
+        mock_bot_detector.state.in_progress_reviews["123"] = "invalid-timestamp"
+
+        is_in_progress, reason = mock_bot_detector.is_review_in_progress(123)
+
+        # Should clear invalid state
+        assert is_in_progress is False
+        assert reason == ""
+        assert "123" not in mock_bot_detector.state.in_progress_reviews
+
+    def test_should_skip_review_in_progress(self, mock_bot_detector):
+        """Test skipping PR when review is in progress."""
+        mock_bot_detector.mark_review_started(123)
+
+        pr_data = {"author": {"login": "alice"}}
+        commits = [{"author": {"login": "alice"}, "oid": "abc123"}]
+
+        should_skip, reason = mock_bot_detector.should_skip_pr_review(
+            pr_number=123,
+            pr_data=pr_data,
+            commits=commits,
+        )
+
+        assert should_skip is True
+        assert "already in progress" in reason.lower()
+
+    def test_mark_reviewed_clears_in_progress(self, mock_bot_detector):
+        """Test that mark_reviewed also clears in-progress state."""
+        mock_bot_detector.mark_review_started(123)
+        assert "123" in mock_bot_detector.state.in_progress_reviews
+
+        mock_bot_detector.mark_reviewed(123, "abc123")
+
+        # In-progress should be cleared
+        assert "123" not in mock_bot_detector.state.in_progress_reviews
+        # Reviewed state should be set
+        assert "123" in mock_bot_detector.state.reviewed_commits
+        assert "abc123" in mock_bot_detector.state.reviewed_commits["123"]
+
+    def test_clear_pr_state_clears_in_progress(self, mock_bot_detector):
+        """Test that clear_pr_state also clears in-progress state."""
+        mock_bot_detector.mark_review_started(123)
+        mock_bot_detector.mark_reviewed(123, "abc123")
+
+        assert (
+            "123" in mock_bot_detector.state.in_progress_reviews or True
+        )  # May be cleared by mark_reviewed
+        assert "123" in mock_bot_detector.state.reviewed_commits
+
+        # Start another review
+        mock_bot_detector.mark_review_started(123)
+        assert "123" in mock_bot_detector.state.in_progress_reviews
+
+        mock_bot_detector.clear_pr_state(123)
+
+        # Everything should be cleared
+        assert "123" not in mock_bot_detector.state.in_progress_reviews
+        assert "123" not in mock_bot_detector.state.reviewed_commits
+        assert "123" not in mock_bot_detector.state.last_review_times
+
+    def test_get_stats_includes_in_progress(self, mock_bot_detector):
+        """Test that get_stats includes in-progress count."""
+        mock_bot_detector.mark_review_started(123)
+        mock_bot_detector.mark_review_started(456)
+        mock_bot_detector.mark_reviewed(789, "abc123")
+
+        stats = mock_bot_detector.get_stats()
+
+        assert stats["in_progress_reviews"] == 2
+        assert stats["total_prs_tracked"] == 1  # Only 789 is tracked as reviewed
+        assert stats["in_progress_timeout_minutes"] == 30
+
+    def test_cleanup_stale_prs_removes_stale_in_progress(self, mock_bot_detector):
+        """Test that cleanup_stale_prs removes stale in-progress reviews."""
+        # Add a stale in-progress review (32 minutes ago)
+        stale_time = datetime.now() - timedelta(minutes=32)
+        mock_bot_detector.state.in_progress_reviews["123"] = stale_time.isoformat()
+
+        # Add an active in-progress review (5 minutes ago)
+        active_time = datetime.now() - timedelta(minutes=5)
+        mock_bot_detector.state.in_progress_reviews["456"] = active_time.isoformat()
+
+        # Add a stale reviewed PR (40 days ago)
+        stale_review_time = datetime.now() - timedelta(days=40)
+        mock_bot_detector.state.reviewed_commits["789"] = ["abc123"]
+        mock_bot_detector.state.last_review_times["789"] = stale_review_time.isoformat()
+
+        cleaned = mock_bot_detector.cleanup_stale_prs(max_age_days=30)
+
+        # Should remove stale in-progress and stale reviewed PR
+        assert cleaned == 2  # 1 stale in-progress + 1 stale reviewed
+        assert "123" not in mock_bot_detector.state.in_progress_reviews
+        assert (
+            "456" in mock_bot_detector.state.in_progress_reviews
+        )  # Active one remains
+        assert "789" not in mock_bot_detector.state.reviewed_commits
+
+
 if __name__ == "__main__":
    pytest.main([__file__, "-v"])
@@ -46,6 +46,7 @@ if sys.version_info < (3, 10):  # noqa: UP036

 import asyncio
 import io
+import json
 import os
 import subprocess
 from pathlib import Path
@@ -252,9 +253,17 @@ Examples:
    # Find project root (look for auto-claude folder)
    project_dir = args.project_dir

-    # Auto-detect if running from within auto-claude directory (the source code)
-    if project_dir.name == "auto-claude" and (project_dir / "run.py").exists():
-        # Running from within auto-claude/ source directory, go up 1 level
+    # Auto-detect if running from within auto-claude/apps/backend/ source directory.
+    # This must be specific: check for run.py FILE (not dir) AND core/client.py to confirm
+    # we're in the actual backend source tree, not just a project named "auto-claude".
+    run_py_path = project_dir / "run.py"
+    if (
+        project_dir.name == "auto-claude"
+        and run_py_path.exists()
+        and run_py_path.is_file()
+        and (project_dir / "core" / "client.py").exists()
+    ):
+        # Running from within auto-claude/apps/backend/ source directory, go up 1 level
        project_dir = project_dir.parent
    elif not (project_dir / ".auto-claude").exists():
        # No .auto-claude folder found - try to find project root
@@ -351,6 +360,36 @@ Examples:
                "--auto-continue",  # Non-interactive mode for chained execution
            ]

+            # Bypass approval re-validation when all conditions are met:
+            # 1. Spec was auto-approved (no human review required)
+            # 2. Spec creation succeeded (we're past the success check above)
+            # 3. No review-before-coding gate was requested
+            # This prevents hash mismatch failures when spec files are
+            # touched between auto-approval and run.py startup.
+            if args.auto_approve:
+                # Default to requiring review (fail-closed) - only skip if explicitly disabled
+                require_review = True
+                task_meta_path = orchestrator.spec_dir / "task_metadata.json"
+                if task_meta_path.exists():
+                    try:
+                        with open(task_meta_path, encoding="utf-8") as f:
+                            task_meta = json.load(f)
+                        require_review = task_meta.get(
+                            "requireReviewBeforeCoding", False
+                        )
+                    except (json.JSONDecodeError, OSError) as e:
+                        # On parse error, keep require_review=True (fail-closed)
+                        debug(
+                            "spec_runner",
+                            f"Failed to parse task_metadata.json, not adding --force: {e}",
+                        )
+                if not require_review:
+                    run_cmd.append("--force")
+                    debug(
+                        "spec_runner",
+                        "Adding --force: auto-approved, no review required, spec completed",
+                    )
+
            # Pass base branch if specified (for worktree creation)
            if args.base_branch:
                run_cmd.extend(["--base-branch", args.base_branch])
@@ -38,7 +38,7 @@ async def bash_security_hook(
        context: Optional context

    Returns:
-        Empty dict to allow, or {"decision": "block", "reason": "..."} to block
+        Empty dict to allow, or hookSpecificOutput with permissionDecision "deny" to block
    """
    if input_data.get("tool_name") != "Bash":
        return {}
@@ -49,15 +49,21 @@ async def bash_security_hook(
    # Check if tool_input is None (malformed tool call)
    if tool_input is None:
        return {
-            "decision": "block",
-            "reason": "Bash tool_input is None - malformed tool call from SDK",
+            "hookSpecificOutput": {
+                "hookEventName": "PreToolUse",
+                "permissionDecision": "deny",
+                "permissionDecisionReason": "Bash tool_input is None - malformed tool call from SDK",
+            }
        }

    # Check if tool_input is a dict
    if not isinstance(tool_input, dict):
        return {
-            "decision": "block",
-            "reason": f"Bash tool_input must be dict, got {type(tool_input).__name__}",
+            "hookSpecificOutput": {
+                "hookEventName": "PreToolUse",
+                "permissionDecision": "deny",
+                "permissionDecisionReason": f"Bash tool_input must be dict, got {type(tool_input).__name__}",
+            }
        }

    # Now safe to access command
@@ -97,8 +103,11 @@ async def bash_security_hook(
    if not commands:
        # Could not parse - fail safe by blocking
        return {
-            "decision": "block",
-            "reason": f"Could not parse command for security validation: {command}",
+            "hookSpecificOutput": {
+                "hookEventName": "PreToolUse",
+                "permissionDecision": "deny",
+                "permissionDecisionReason": f"Could not parse command for security validation: {command}",
+            }
        }

    # Split into segments for per-command validation
@@ -114,8 +123,11 @@ async def bash_security_hook(

        if not is_allowed:
            return {
-                "decision": "block",
-                "reason": reason,
+                "hookSpecificOutput": {
+                    "hookEventName": "PreToolUse",
+                    "permissionDecision": "deny",
+                    "permissionDecisionReason": reason,
+                }
            }

        # Additional validation for sensitive commands
@@ -127,7 +139,13 @@ async def bash_security_hook(
            validator = VALIDATORS[cmd]
            allowed, reason = validator(cmd_segment)
            if not allowed:
-                return {"decision": "block", "reason": reason}
+                return {
+                    "hookSpecificOutput": {
+                        "hookEventName": "PreToolUse",
+                        "permissionDecision": "deny",
+                        "permissionDecisionReason": reason,
+                    }
+                }

    return {}

@@ -10,6 +10,7 @@ from collections.abc import Callable
 from pathlib import Path

 from analysis.analyzers import analyze_project
+from core.task_event import TaskEventEmitter
 from core.workspace.models import SpecNumberLock
 from phase_config import get_thinking_budget
 from prompts_pkg.project_context import should_refresh_project_index
@@ -234,6 +235,7 @@ class SpecOrchestrator:
        # Initialize task logger for planning phase
        task_logger = get_task_logger(self.spec_dir)
        task_logger.start_phase(LogPhase.PLANNING, "Starting spec creation process")
+        TaskEventEmitter.from_spec_dir(self.spec_dir).emit("PLANNING_STARTED")

        print(
            box(
@@ -408,6 +410,28 @@ class SpecOrchestrator:
            LogPhase.PLANNING, success=True, message="Spec creation complete"
        )

+        # Load task metadata to check requireReviewBeforeCoding setting
+        task_metadata_file = self.spec_dir / "task_metadata.json"
+        require_review_before_coding = False
+        if task_metadata_file.exists():
+            with open(task_metadata_file, encoding="utf-8") as f:
+                task_metadata = json.load(f)
+                require_review_before_coding = task_metadata.get(
+                    "requireReviewBeforeCoding", False
+                )
+
+        # Emit PLANNING_COMPLETE event for XState machine transition
+        # This signals the frontend that spec creation is done
+        task_emitter = TaskEventEmitter.from_spec_dir(self.spec_dir)
+        task_emitter.emit(
+            "PLANNING_COMPLETE",
+            {
+                "hasSubtasks": False,  # Spec creation doesn't have subtasks yet
+                "subtaskCount": 0,
+                "requireReviewBeforeCoding": require_review_before_coding,
+            },
+        )
+
        # === HUMAN REVIEW CHECKPOINT ===
        return self._run_review_checkpoint(auto_approve)

@@ -0,0 +1,172 @@
+# XState Task State Machine Migration - Summary
+
+**Issue:** #1338
+**PR:** #1575
+**Date:** 2026-01-28
+**Branch:** fix/1524-xstate-clean
+
+## Overview
+
+Migrated task status management from scattered decision logic across multiple handler files to a centralized XState v5 state machine. This eliminates race conditions, inconsistent status updates, and makes the task lifecycle formally defined and testable.
+
+## Critical Dependencies & Blockers
+
+### 1. Windows Credential Manager Fix (Required for Testing)
+**PR:** #1569 - fix(windows): fix Windows Credential Manager authentication
+**Issue:** #1525
+
+This PR includes changes that depend on the Windows authentication fix. We could not complete end-to-end testing without this fix in place. If a different solution is implemented for #1525, we can remove these changes and resubmit.
+
+### 2. spec_runner.py Project Detection Fix
+**Issue:** #1570 - spec_runner.py incorrectly detects auto-claude project as source directory
+
+We encountered and fixed this bug during development as it was blocking our test workflow. The fix is included in this PR.
+
+## Implementation Phases
+
+| Phase | Description | Status |
+|-------|-------------|--------|
+| Phase 1 | Create XState machine definition (task-machine.ts) | ✅ Complete |
+| Phase 2 | Create TaskStateManager singleton wrapper | ✅ Complete |
+| Phase 3 | Integrate into agent-events-handlers.ts | ⏸️ Partially done |
+| Phase 4 | Remove legacy TaskStateMachine class | ❌ Not started |
+
+### Why We Stopped at Phase 2
+
+The original scope was to introduce XState as the new state management approach. Full integration (Phase 3-4) requires:
+
+- Extensive refactoring of agent-events-handlers.ts to remove all legacy decision logic
+- Removing the old TaskStateMachine class entirely
+- Migration of all status persistence to go through XState
+
+We delivered Phases 1-2 to establish the foundation. The current state has both systems running in parallel with XState as primary:
+
+- **XState is primary:** When TaskStateManager returns a valid state transition, that decision is used
+- **Legacy as fallback:** The old TaskStateMachine logic only applies when XState doesn't produce a decision
+- **Safe rollback:** If XState causes issues, the legacy system is still present and can take over
+
+This dual-system approach allows:
+- Validation that XState produces correct state transitions in production
+- Safe rollback if issues arise
+- Incremental adoption path for Phase 3-4
+
+## What Changed
+
+### Before (Old Architecture)
+- Status decisions scattered across agent-events-handlers.ts, execution-handlers.ts, worktree-handlers.ts
+- `validateStatusTransition()` function with complex conditional logic
+- TaskStateMachine class that was essentially an event emitter wrapper
+- Multiple places persisting status to implementation_plan.json
+- Race conditions possible when multiple handlers tried to update status
+
+### After (New Architecture)
+- **Single source of truth:** TaskStateManager (XState-based singleton)
+- **Formal state machine:** taskMachine with explicit states and transitions
+- **Centralized persistence:** Status written to JSON from one place
+- **Testable:** Unit tests verify all state transitions
+- **Observable:** XState actors can be inspected/visualized
+
+## State Machine States
+
+```
+backlog → planning → coding → qa_review → qa_fixing → human_review → done
+                  ↘ plan_review ↗              ↓
+                                             error
+```
+
+| State | Maps to Legacy Status | reviewReason |
+|-------|----------------------|--------------|
+| backlog | backlog | - |
+| planning | in_progress | - |
+| coding | in_progress | - |
+| plan_review | human_review | plan_review |
+| qa_review | ai_review | - |
+| qa_fixing | ai_review | - |
+| human_review | human_review | completed or stopped |
+| creating_pr | human_review | completed |
+| pr_created | pr_created | - |
+| error | human_review | errors |
+| done | done | - |
+
+## Key Files
+
+| File | Purpose |
+|------|---------|
+| `apps/frontend/src/shared/state-machines/task-machine.ts` | XState machine definition |
+| `apps/frontend/src/main/task-state-manager.ts` | Singleton service wrapping XState actors |
+| `apps/frontend/src/shared/state-machines/__tests__/task-machine.test.ts` | State machine unit tests (35 tests) |
+| `apps/frontend/src/main/__tests__/task-state-manager.test.ts` | Manager service unit tests (20 tests) |
+| `apps/frontend/src/main/ipc-handlers/agent-events-handlers.ts` | Refactored to call TaskStateManager |
+
+## Events
+
+The state machine responds to these events:
+
+| Event | Triggered By |
+|-------|-------------|
+| PLANNING_STARTED | Execution progress phase=planning |
+| PLANNING_COMPLETE | Execution progress moving past planning |
+| PLAN_APPROVED | User clicks "Proceed to Coding" from plan_review |
+| CODING_STARTED | Execution progress phase=coding |
+| QA_STARTED | Execution progress phase=qa_review |
+| QA_PASSED | Execution progress phase=complete |
+| QA_FAILED | Execution progress phase=qa_fixing |
+| PROCESS_EXITED | Agent process exit event |
+| USER_STOPPED | User clicks stop |
+| USER_RESUMED | User resumes task |
+| MARK_DONE | User marks task as done |
+| CREATE_PR | User initiates PR creation |
+| PR_CREATED | PR successfully created |
+
+## Rollback Plan
+
+If issues arise post-merge:
+
+1. **Quick rollback:** `git revert <merge-commit>`
+2. **Restore point:** Commit 3e5f004a has old code intact
+3. **Legacy persistence still works:** implementation_plan.json continues to store status
+
+## Testing
+
+| Test Suite | Result |
+|------------|--------|
+| Frontend unit tests | ✅ 2579 passed |
+| TypeScript strict mode | ✅ Pass |
+| Biome lint | ✅ Pass |
+| XState machine tests | ✅ 35 passed |
+| TaskStateManager tests | ✅ 20 passed |
+| Python backend tests | ✅ Pass |
+
+## Session Fixes (2026-01-28)
+
+### Fixed Issues
+
+1. **Badge showing "Needs Review" instead of "Complete"** - Added `effectiveReviewReason` logic in TaskCard.tsx that sets 'completed' when phase === 'complete'
+
+2. **Task showing "Incomplete" badge for plan_review** - Added 'plan_review' to exclusion list in `isIncompleteHumanReview`
+
+3. **Missing "Proceed to Coding" button** - Restored in WorkspaceMessages.tsx for plan_review flow
+
+4. **Wrong XState event for plan_review → coding** - Fixed to send PLAN_APPROVED instead of PLANNING_STARTED when starting from plan_review state
+
+5. **Stuck detection logic** - Reverted useTaskDetail.ts to simpler logic from working branch (only skip 'planning' phase, 2s timeout)
+
+## Outstanding Items (Requires PM Input)
+
+### 1. Future: Subtask XState Migration
+- **Issue:** `subtask.status` is checked directly in UI code
+- **Recommendation:** Should be managed by state machine for consistency
+- **Status:** Out of scope for current PR, document for future work
+
+## Future Improvements
+
+- Add @stately-ai/inspect for runtime devtools
+- **Subtask state management** - Track individual subtask states within the machine using XState parallel states
+- Add more granular QA states (qa_round_1, qa_round_2, etc.)
+- Complete Phase 3-4: Full integration and removal of legacy TaskStateMachine class
+
+## Visualization
+
+The state machine can be visualized at [Stately.ai Editor](https://stately.ai/editor):
+1. Paste the contents of task-machine.ts
+2. Click "Visualize" to see the state diagram
@@ -1,6 +1,10 @@
 import { defineConfig, externalizeDepsPlugin } from 'electron-vite';
 import react from '@vitejs/plugin-react';
 import { resolve } from 'path';
+import { config as dotenvConfig } from 'dotenv';
+
+// Load .env file for build-time constants (Sentry DSN, etc.)
+dotenvConfig({ path: resolve(__dirname, '.env') });

 /**
 * Sentry configuration embedded at build time.
@@ -43,7 +47,9 @@ export default defineConfig({
        'debug',
        'ms',
        // Minimatch for glob pattern matching in worktree handlers
-        'minimatch'
+        'minimatch',
+        // XState for task state machine
+        'xstate'
      ]
    })],
    build: {
@@ -1,6 +1,6 @@
 {
  "name": "auto-claude-ui",
-  "version": "2.7.5",
+  "version": "2.7.6-beta.2",
  "type": "module",
  "description": "Desktop UI for Auto Claude autonomous coding framework",
  "homepage": "https://github.com/AndyMik90/Auto-Claude",
@@ -36,6 +36,8 @@
    "package:win": "node scripts/package-with-python.cjs --win",
    "package:linux": "node scripts/package-with-python.cjs --linux",
    "package:flatpak": "node scripts/package-with-python.cjs --linux flatpak",
+    "verify:linux": "node scripts/verify-linux-packages.cjs dist",
+    "test:verify-linux": "node --test scripts/verify-linux-packages.test.mjs",
    "start:packaged:mac": "open dist/mac-arm64/Auto-Claude.app || open dist/mac/Auto-Claude.app",
    "start:packaged:win": "start \"\" \"dist\\win-unpacked\\Auto-Claude.exe\"",
    "start:packaged:linux": "./dist/linux-unpacked/auto-claude",
@@ -94,10 +96,13 @@
    "react-i18next": "^16.5.0",
    "react-markdown": "^10.1.0",
    "react-resizable-panels": "^4.2.0",
+    "rehype-raw": "^7.0.0",
+    "rehype-sanitize": "^6.0.0",
    "remark-gfm": "^4.0.1",
    "semver": "^7.7.3",
    "tailwind-merge": "^3.4.0",
    "uuid": "^13.0.0",
+    "xstate": "^5.26.0",
    "zod": "^4.2.1",
    "zustand": "^5.0.9"
  },
@@ -165,14 +170,6 @@
        "from": "resources/icon.ico",
        "to": "icon.ico"
      },
-      {
-        "from": "python-runtime/${os}-${arch}/python",
-        "to": "python"
-      },
-      {
-        "from": "python-runtime/${os}-${arch}/site-packages",
-        "to": "python-site-packages"
-      },
      {
        "from": "../backend",
        "to": "backend",
@@ -202,6 +199,16 @@
      "target": [
        "dmg",
        "zip"
+      ],
+      "extraResources": [
+        {
+          "from": "python-runtime/${os}-${arch}/python",
+          "to": "python"
+        },
+        {
+          "from": "python-runtime/${os}-${arch}/site-packages",
+          "to": "python-site-packages"
+        }
      ]
    },
    "win": {
@@ -209,6 +216,16 @@
      "target": [
        "nsis",
        "zip"
+      ],
+      "extraResources": [
+        {
+          "from": "python-runtime/${os}-${arch}/python",
+          "to": "python"
+        },
+        {
+          "from": "python-runtime/${os}-${arch}/site-packages",
+          "to": "python-site-packages"
+        }
      ]
    },
    "linux": {
@@ -218,7 +235,17 @@
        "deb",
        "flatpak"
      ],
-      "category": "Development"
+      "category": "Development",
+      "extraResources": [
+        {
+          "from": "python-runtime/${os}-${arch}/python",
+          "to": "python"
+        },
+        {
+          "from": "python-runtime/${os}-${arch}/site-packages",
+          "to": "python-site-packages"
+        }
+      ]
    },
    "flatpak": {
      "runtime": "org.freedesktop.Platform",
@@ -0,0 +1,406 @@
+#!/usr/bin/env node
+/**
+ * Verify Linux package contents to ensure alignment between AppImage, deb, and Flatpak.
+ *
+ * This script extracts and inspects each Linux package format to verify that critical
+ * files (Python binary, backend code, Python packages) are present and correctly bundled.
+ *
+ * Usage: node scripts/verify-linux-packages.cjs [dist-dir]
+ */
+
+const fs = require('fs');
+const path = require('path');
+const { spawnSync } = require('child_process');
+
+// Critical Python packages that must be present
+const CRITICAL_PACKAGES = [
+  'secretstorage', // Linux OAuth token storage
+  'pydantic_core',
+  'claude_agent_sdk',
+  'dotenv',
+];
+
+// Minimum expected Flatpak file size (50 MB)
+// Flatpak files are large OCI archives; anything smaller is suspicious
+// Based on observed minimum sizes of valid builds
+const FLATPAK_MIN_SIZE_MB = 50;
+
+// Colors for terminal output
+const colors = {
+  reset: '\x1b[0m',
+  red: '\x1b[31m',
+  green: '\x1b[32m',
+  yellow: '\x1b[33m',
+  blue: '\x1b[34m',
+  cyan: '\x1b[36m',
+};
+
+function log(message, color = colors.reset) {
+  console.log(`${color}${message}${colors.reset}`);
+}
+
+function logSuccess(message) {
+  log(`✓ ${message}`, colors.green);
+}
+
+function logError(message) {
+  log(`✗ ${message}`, colors.red);
+}
+
+function logWarning(message) {
+  log(`⚠ ${message}`, colors.yellow);
+}
+
+function logInfo(message) {
+  log(`ℹ ${message}`, colors.cyan);
+}
+
+/**
+ * Check if a command exists
+ * Uses 'which' directly without shell interpolation to prevent command injection
+ */
+function commandExists(cmd) {
+  const result = spawnSync('which', [cmd], { stdio: 'ignore' });
+  return result.status === 0;
+}
+
+/**
+ * Find all Linux packages in the dist directory
+ */
+function findPackages(distDir) {
+  const packages = {
+    appImage: null,
+    deb: null,
+    flatpak: null,
+  };
+
+  if (!fs.existsSync(distDir)) {
+    logError(`Distribution directory not found: ${distDir}`);
+    return packages;
+  }
+
+  const files = fs.readdirSync(distDir);
+
+  for (const file of files) {
+    const fullPath = path.join(distDir, file);
+
+    if (file.endsWith('.AppImage')) {
+      if (!packages.appImage) {
+        packages.appImage = fullPath;
+      } else {
+        logWarning(`Multiple AppImage files found, using first: ${path.basename(packages.appImage)}`);
+      }
+    } else if (file.endsWith('.deb')) {
+      if (!packages.deb) {
+        packages.deb = fullPath;
+      } else {
+        logWarning(`Multiple deb files found, using first: ${path.basename(packages.deb)}`);
+      }
+    } else if (file.endsWith('.flatpak')) {
+      if (!packages.flatpak) {
+        packages.flatpak = fullPath;
+      } else {
+        logWarning(`Multiple Flatpak files found, using first: ${path.basename(packages.flatpak)}`);
+      }
+    }
+  }
+
+  return packages;
+}
+
+/**
+ * Common file list verification logic
+ * @param {string[]} files - List of files from package
+ * @param {string} packageType - Type of package (for error messages)
+ * @returns {Object} Verification result with verified flag and issues array
+ *
+ * File formats:
+ * - AppImage (bsdtar): './resources/python', './resources/backend/file.py'
+ * - deb (dpkg-deb -c): 'resources/python', 'resources/backend/file.py' (in last column)
+ */
+function verifyFileList(files, packageType) {
+  const issues = [];
+
+  // Normalize paths by removing trailing slashes (archive tools commonly add these)
+  const normalizePath = (p) => p.replace(/\/+$/, '');
+
+  // Check for Python binary directory
+  // AppImage: './resources/python' or './resources/python/' (with trailing slash)
+  // deb: 'resources/python' or 'resources/python/' (with trailing slash)
+  // Must NOT match 'resources/python-site-packages'
+  const pythonBinFound = files.some((f) => {
+    const normalized = normalizePath(f);
+    return (
+      (normalized === './resources/python' ||
+        normalized === 'resources/python' ||
+        normalized.endsWith('/resources/python')) &&
+      !f.includes('python-site-packages')
+    );
+  });
+  if (!pythonBinFound) {
+    issues.push(`Python binary directory not found in ${packageType}`);
+  }
+
+  // Check for backend directory (must be under resources/)
+  const backendFound = files.some((f) => {
+    const normalized = normalizePath(f);
+    return (
+      f.includes('./resources/backend/') ||
+      f.includes('resources/backend/') ||
+      normalized === './resources/backend' ||
+      normalized === 'resources/backend'
+    );
+  });
+  if (!backendFound) {
+    issues.push(`Backend directory not found in ${packageType}`);
+  }
+
+  // Check for critical Python packages (must be under python-site-packages/)
+  for (const pkg of CRITICAL_PACKAGES) {
+    // Match: './resources/python-site-packages/secretstorage/__init__.py'
+    // Match: 'resources/python-site-packages/secretstorage/__init__.py'
+    // Don't match: '/some/other/path/secretstorage/'
+    const found = files.some(
+      (f) => f.includes(`python-site-packages/${pkg}/`) || f.includes(`python-site-packages/${pkg}.`),
+    );
+    if (!found) {
+      issues.push(`Python package not found: ${pkg}`);
+    }
+  }
+
+  return {
+    verified: issues.length === 0,
+    issues,
+    fileCount: files.filter((f) => f.trim()).length,
+  };
+}
+
+/**
+ * Verify AppImage contents using bsdtar (libarchive)
+ */
+function verifyAppImage(appImagePath) {
+  logInfo(`Verifying AppImage: ${path.basename(appImagePath)}`);
+
+  // Check if bsdtar is available
+  if (!commandExists('bsdtar')) {
+    logWarning('bsdtar not found. Install with: sudo apt-get install libarchive-tools');
+    logWarning('Skipping AppImage verification');
+    return { verified: false, reason: 'bsdtar not available', critical: true };
+  }
+
+  // Extract file list from AppImage using bsdtar
+  const result = spawnSync('bsdtar', ['-t', '-f', appImagePath], {
+    stdio: 'pipe',
+    encoding: 'utf-8',
+    maxBuffer: 50 * 1024 * 1024, // 50MB buffer for large file listings
+  });
+
+  // Check for spawn errors (e.g., permission denied, memory issues)
+  if (result.error) {
+    logError(`Failed to execute bsdtar: ${result.error.message}`);
+    return { verified: false, reason: `Command execution failed: ${result.error.message}` };
+  }
+
+  if (result.status !== 0) {
+    logError(`Failed to read AppImage: ${result.stderr}`);
+    return { verified: false, reason: 'Failed to extract file list' };
+  }
+
+  const files = result.stdout.split('\n');
+  return verifyFileList(files, 'AppImage');
+}
+
+/**
+ * Verify deb package contents
+ */
+function verifyDeb(debPath) {
+  logInfo(`Verifying deb package: ${path.basename(debPath)}`);
+
+  // Check if dpkg is available
+  if (!commandExists('dpkg-deb')) {
+    logWarning('dpkg-deb not found. Skipping deb verification');
+    return { verified: false, reason: 'dpkg-deb not available', critical: true };
+  }
+
+  // List contents of deb package
+  const result = spawnSync('dpkg-deb', ['-c', debPath], {
+    stdio: 'pipe',
+    encoding: 'utf-8',
+    maxBuffer: 50 * 1024 * 1024, // 50MB buffer for large file listings
+  });
+
+  // Check for spawn errors (e.g., permission denied, memory issues)
+  if (result.error) {
+    logError(`Failed to execute dpkg-deb: ${result.error.message}`);
+    return { verified: false, reason: `Command execution failed: ${result.error.message}` };
+  }
+
+  if (result.status !== 0) {
+    logError(`Failed to read deb package: ${result.stderr}`);
+    return { verified: false, reason: 'Failed to extract file list' };
+  }
+
+  const files = result.stdout.split('\n');
+  return verifyFileList(files, 'deb package');
+}
+
+/**
+ * Verify Flatpak package contents
+ * Note: Flatpak is more complex to inspect, so we do basic validation
+ */
+function verifyFlatpak(flatpakPath) {
+  logInfo(`Verifying Flatpak package: ${path.basename(flatpakPath)}`);
+
+  const issues = [];
+
+  // Check if flatpak command is available for detailed validation
+  const hasFlatpakCli = commandExists('flatpak');
+  if (!hasFlatpakCli) {
+    logWarning('flatpak command not found. Skipping detailed Flatpak verification');
+    // Continue with basic file existence/size checks
+  }
+
+  // Check if file exists and is not empty
+  if (!fs.existsSync(flatpakPath)) {
+    return { verified: false, issues: ['Flatpak file does not exist'] };
+  }
+
+  const stats = fs.statSync(flatpakPath);
+  if (stats.size === 0) {
+    return { verified: false, issues: ['Flatpak file is empty'] };
+  }
+
+  // Flatpak files are large OCI archives, so we just verify file size and basic structure
+  // Detailed content inspection would require mounting or extracting the flatpak
+  if (stats.size < FLATPAK_MIN_SIZE_MB * 1024 * 1024) {
+    // Less than minimum size is suspicious
+    issues.push(
+      `Flatpak file seems too small (${(stats.size / 1024 / 1024).toFixed(2)} MB, expected at least ${FLATPAK_MIN_SIZE_MB} MB)`,
+    );
+  }
+
+  return {
+    verified: issues.length === 0,
+    issues,
+    size: stats.size,
+  };
+}
+
+/**
+ * Main verification function
+ */
+function main() {
+  const distDir = process.argv[2] || path.join(__dirname, '..', 'dist');
+
+  log('\n=== Linux Package Verification ===\n', colors.blue);
+  logInfo(`Distribution directory: ${distDir}\n`);
+
+  const packages = findPackages(distDir);
+
+  // Report found packages
+  if (packages.appImage) {
+    logSuccess(`Found AppImage: ${path.basename(packages.appImage)}`);
+  } else {
+    logWarning('No AppImage found');
+  }
+
+  if (packages.deb) {
+    logSuccess(`Found deb: ${path.basename(packages.deb)}`);
+  } else {
+    logWarning('No deb package found');
+  }
+
+  if (packages.flatpak) {
+    logSuccess(`Found Flatpak: ${path.basename(packages.flatpak)}`);
+  } else {
+    logWarning('No Flatpak package found');
+  }
+
+  if (!packages.appImage && !packages.deb && !packages.flatpak) {
+    logError('\nNo Linux packages found to verify!');
+    process.exit(1);
+  }
+
+  log('');
+
+  // Verify each package
+  const results = {};
+
+  if (packages.appImage) {
+    results.appImage = verifyAppImage(packages.appImage);
+  }
+
+  if (packages.deb) {
+    results.deb = verifyDeb(packages.deb);
+  }
+
+  if (packages.flatpak) {
+    results.flatpak = verifyFlatpak(packages.flatpak);
+  }
+
+  // Print results
+  log('\n=== Verification Results ===\n', colors.blue);
+
+  let hasFailures = false;
+  let hasCriticalSkips = false;
+
+  for (const [type, result] of Object.entries(results)) {
+    if (result.reason) {
+      if (result.critical) {
+        logError(`${type}: CRITICAL - SKIPPED (${result.reason})`);
+        hasCriticalSkips = true;
+      } else {
+        logWarning(`${type}: SKIPPED (${result.reason})`);
+      }
+    } else if (result.verified) {
+      logSuccess(`${type}: VERIFIED`);
+      if (result.fileCount) {
+        logInfo(`  Files: ${result.fileCount}`);
+      }
+      if (result.size) {
+        logInfo(`  Size: ${(result.size / 1024 / 1024).toFixed(2)} MB`);
+      }
+    } else {
+      logError(`${type}: FAILED`);
+      hasFailures = true;
+      for (const issue of result.issues || []) {
+        logError(`  - ${issue}`);
+      }
+    }
+  }
+
+  log('');
+
+  if (hasFailures || hasCriticalSkips) {
+    logError('\n=== VERIFICATION FAILED ===\n');
+    if (hasFailures) {
+      log('Some packages are missing critical files. This will cause runtime errors.\n', colors.red);
+    }
+    if (hasCriticalSkips) {
+      log('Some packages could not be verified due to missing required tools.\n', colors.red);
+      log('Install required tools:\n', colors.red);
+      log('  - bsdtar: sudo apt-get install libarchive-tools\n', colors.red);
+      log('  - dpkg-deb: sudo apt-get install dpkg\n', colors.red);
+    }
+    process.exit(1);
+  } else {
+    logSuccess('\n=== ALL PACKAGES VERIFIED ===\n');
+    log('All Linux packages contain the required files.\n', colors.green);
+    process.exit(0);
+  }
+}
+
+// Only run main if this file is executed directly (not imported)
+if (require.main === module) {
+  main();
+}
+
+// Export for testing
+module.exports = {
+  CRITICAL_PACKAGES,
+  findPackages,
+  verifyFileList,
+  verifyAppImage,
+  verifyDeb,
+  verifyFlatpak,
+};
@@ -0,0 +1,533 @@
+/**
+ * Tests for verify-linux-packages.cjs
+ *
+ * These tests cover the core logic by calling the actual exported functions.
+ */
+
+import { describe, it, mock } from 'node:test';
+import assert from 'node:assert';
+import fs from 'node:fs';
+import { createRequire } from 'node:module';
+import { fileURLToPath } from 'node:url';
+import { dirname } from 'node:path';
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = dirname(__filename);
+
+const require = createRequire(import.meta.url);
+
+// Get child_process and save original spawnSync
+const childProcess = require('child_process');
+const originalSpawnSync = childProcess.spawnSync;
+
+// Helper to reload the verification module with a mocked spawnSync
+function loadWithMockedSpawnSync(mockFn) {
+  // Set the mock before requiring
+  childProcess.spawnSync = mockFn;
+  // Clear the module cache
+  delete require.cache[require.resolve('./verify-linux-packages.cjs')];
+  // Re-require the module
+  return require('./verify-linux-packages.cjs');
+}
+
+function restoreSpawnSync() {
+  childProcess.spawnSync = originalSpawnSync;
+  delete require.cache[require.resolve('./verify-linux-packages.cjs')];
+}
+
+// Load the module normally for tests that don't need spawnSync mocking
+const {
+  CRITICAL_PACKAGES,
+  findPackages,
+  verifyFileList,
+  verifyFlatpak,
+} = require('./verify-linux-packages.cjs');
+
+describe('verify-linux-packages', () => {
+  describe('package finding logic', () => {
+    it('should identify all three Linux package types', () => {
+      // Test that findPackages correctly identifies .AppImage, .deb, and .flatpak files
+      const mockFiles = [
+        'Auto-Claude-2.7.5-linux-x86_64.AppImage',
+        'auto-claude_2.7.5_amd64.deb',
+        'com.autoclaude.ui_2.7.5_linux_x86_64.flatpak',
+        'latest-mac.yml',
+        'latest.yml',
+      ];
+
+      const distDir = '/test/dist';
+
+      // Mock fs.existsSync to return true (directory exists)
+      const existsSync = mock.method(fs, 'existsSync', mock.fn(() => true));
+      // Mock fs.readdirSync to return our test files
+      const readdirSync = mock.method(fs, 'readdirSync', mock.fn(() => mockFiles));
+
+      try {
+        const result = findPackages(distDir);
+
+        // Verify the expected results
+        assert.equal(result.appImage, '/test/dist/Auto-Claude-2.7.5-linux-x86_64.AppImage');
+        assert.equal(result.deb, '/test/dist/auto-claude_2.7.5_amd64.deb');
+        assert.equal(result.flatpak, '/test/dist/com.autoclaude.ui_2.7.5_linux_x86_64.flatpak');
+      } finally {
+        existsSync.mock.restore();
+        readdirSync.mock.restore();
+      }
+    });
+
+    it('should handle missing packages gracefully', () => {
+      // Test behavior when packages are missing
+      const mockFiles = ['latest-mac.yml', 'latest.yml'];
+      const distDir = '/test/dist';
+
+      const existsSync = mock.method(fs, 'existsSync', mock.fn(() => true));
+      const readdirSync = mock.method(fs, 'readdirSync', mock.fn(() => mockFiles));
+
+      try {
+        const result = findPackages(distDir);
+
+        assert.equal(result.appImage, null);
+        assert.equal(result.deb, null);
+        assert.equal(result.flatpak, null);
+      } finally {
+        existsSync.mock.restore();
+        readdirSync.mock.restore();
+      }
+    });
+
+    it('should handle missing dist directory', () => {
+      // Test behavior when dist directory doesn't exist
+      const distDir = '/test/dist';
+
+      const existsSync = mock.method(fs, 'existsSync', mock.fn(() => false));
+
+      try {
+        const result = findPackages(distDir);
+
+        // Should return empty packages object without error
+        assert.equal(result.appImage, null);
+        assert.equal(result.deb, null);
+        assert.equal(result.flatpak, null);
+      } finally {
+        existsSync.mock.restore();
+      }
+    });
+
+    it('should warn about duplicate packages', () => {
+      // Test behavior when multiple packages of same type exist
+      const mockFiles = [
+        'Auto-Claude-2.7.5-linux-x86_64.AppImage',
+        'Auto-Claude-2.7.5-linux-x86_64.AppImage', // Duplicate
+        'auto-claude_2.7.5_amd64.deb',
+        'auto-claude_2.7.5_amd64.deb', // Duplicate
+        'com.autoclaude.ui_2.7.5_linux_x86_64.flatpak',
+      ];
+      const distDir = '/test/dist';
+
+      const existsSync = mock.method(fs, 'existsSync', mock.fn(() => true));
+      const readdirSync = mock.method(fs, 'readdirSync', mock.fn(() => mockFiles));
+
+      try {
+        const result = findPackages(distDir);
+
+        // Should still find packages, using first occurrence
+        assert.equal(result.appImage, '/test/dist/Auto-Claude-2.7.5-linux-x86_64.AppImage');
+        assert.equal(result.deb, '/test/dist/auto-claude_2.7.5_amd64.deb');
+        assert.equal(result.flatpak, '/test/dist/com.autoclaude.ui_2.7.5_linux_x86_64.flatpak');
+      } finally {
+        existsSync.mock.restore();
+        readdirSync.mock.restore();
+      }
+    });
+  });
+
+  describe('critical packages list', () => {
+    it('should contain all required Linux packages', () => {
+      assert.ok(CRITICAL_PACKAGES.includes('secretstorage'), 'secretstorage must be present for Linux OAuth');
+      assert.ok(CRITICAL_PACKAGES.includes('pydantic_core'), 'pydantic_core must be present');
+      assert.ok(CRITICAL_PACKAGES.includes('claude_agent_sdk'), 'claude_agent_sdk must be present');
+      assert.ok(CRITICAL_PACKAGES.includes('dotenv'), 'dotenv must be present');
+    });
+  });
+
+  describe('file content verification logic', () => {
+    it('should detect Python binary in file list', () => {
+      // AppImage format uses './' prefix
+      const mockFiles = [
+        'usr/bin/auto-claude',
+        './resources/python',
+        './resources/backend/core/client.py',
+        './resources/python-site-packages/secretstorage/__init__.py',
+        './resources/python-site-packages/pydantic_core/__init__.py',
+        './resources/python-site-packages/claude_agent_sdk/__init__.py',
+        './resources/python-site-packages/dotenv/__init__.py',
+      ];
+
+      const result = verifyFileList(mockFiles, 'test-package');
+      assert.ok(result.verified, 'Should detect Python binary directory');
+      assert.equal(result.issues.length, 0);
+    });
+
+    it('should detect Python binary with trailing slashes', () => {
+      // Archive tools like bsdtar/dpkg-deb commonly output directories with trailing slashes
+      const mockFiles = [
+        'usr/bin/auto-claude',
+        './resources/python/',  // Trailing slash
+        'resources/backend/',   // Trailing slash
+        './resources/python-site-packages/secretstorage/__init__.py',
+        './resources/python-site-packages/pydantic_core/__init__.py',
+        './resources/python-site-packages/claude_agent_sdk/__init__.py',
+        './resources/python-site-packages/dotenv/__init__.py',
+      ];
+
+      const result = verifyFileList(mockFiles, 'test-package');
+      assert.ok(result.verified, 'Should detect Python binary directory with trailing slash');
+      assert.equal(result.issues.length, 0);
+    });
+
+    it('should detect backend directory in file list', () => {
+      const mockFiles = [
+        'usr/bin/auto-claude',
+        './resources/python',
+        './resources/backend/core/client.py',
+        './resources/python-site-packages/secretstorage/__init__.py',
+        './resources/python-site-packages/pydantic_core/__init__.py',
+        './resources/python-site-packages/claude_agent_sdk/__init__.py',
+        './resources/python-site-packages/dotenv/__init__.py',
+      ];
+
+      const result = verifyFileList(mockFiles, 'test-package');
+      assert.ok(result.verified, 'Should detect backend directory');
+      assert.equal(result.issues.length, 0);
+    });
+
+    it('should detect critical Python packages', () => {
+      const mockFiles = [
+        'usr/bin/auto-claude',
+        './resources/python',
+        './resources/backend/core/client.py',
+        './resources/python-site-packages/secretstorage/__init__.py',
+        './resources/python-site-packages/pydantic_core/__init__.py',
+        './resources/python-site-packages/claude_agent_sdk/__init__.py',
+        './resources/python-site-packages/dotenv/__init__.py',
+      ];
+
+      const result = verifyFileList(mockFiles, 'test-package');
+      assert.ok(result.verified, 'Should detect all critical packages');
+      assert.equal(result.issues.length, 0);
+    });
+
+    it('should report missing packages', () => {
+      const mockFiles = [
+        'usr/bin/auto-claude',
+        './resources/python',
+        './resources/backend/core/client.py',
+        './resources/python-site-packages/dotenv/__init__.py',
+      ];
+
+      const result = verifyFileList(mockFiles, 'test-package');
+
+      assert.ok(!result.verified, 'Should fail verification');
+      assert.ok(result.issues.includes('Python package not found: secretstorage'));
+      assert.ok(result.issues.includes('Python package not found: pydantic_core'));
+      assert.ok(result.issues.includes('Python package not found: claude_agent_sdk'));
+      assert.ok(!result.issues.some((i) => i.includes('dotenv')));
+    });
+
+    it('should not match python-site-packages when looking for python binary', () => {
+      const mockFiles = [
+        'usr/bin/auto-claude',
+        './resources/python-site-packages/secretstorage/__init__.py',
+        './resources/python-site-packages/pydantic_core/__init__.py',
+        './resources/python-site-packages/claude_agent_sdk/__init__.py',
+        './resources/python-site-packages/dotenv/__init__.py',
+        // Note: NO './resources/python' entry
+      ];
+
+      const result = verifyFileList(mockFiles, 'test-package');
+
+      assert.ok(!result.verified, 'Should fail verification');
+      assert.ok(result.issues.some((i) => i.includes('Python binary directory not found')));
+    });
+
+    it('should not match unrelated paths when looking for packages', () => {
+      const mockFiles = [
+        'usr/bin/auto-claude',
+        './resources/python',
+        './resources/backend/core/client.py',
+        // These paths end with package names but are NOT under python-site-packages
+        './some/other/path/secretstorage/file.txt',
+        './unrelated/dotenv/config',
+        './another/pydantic_core/standalone/__init__.py',
+      ];
+
+      const result = verifyFileList(mockFiles, 'test-package');
+
+      assert.ok(!result.verified, 'Should fail verification');
+      assert.ok(result.issues.some((i) => i.includes('Python package not found: secretstorage')));
+    });
+  });
+
+  describe('Flatpak file validation', () => {
+    it('should reject empty Flatpak files', () => {
+      const flatpakPath = '/test/app.flatpak';
+      const mockStat = { size: 0 };
+
+      // Mock fs.existsSync to return true
+      const existsSync = mock.method(fs, 'existsSync', mock.fn(() => true));
+      // Mock fs.statSync to return empty file stats
+      const statSync = mock.method(fs, 'statSync', mock.fn(() => mockStat));
+
+      try {
+        const result = verifyFlatpak(flatpakPath);
+
+        assert.ok(!result.verified, 'Should reject empty Flatpak files');
+        assert.ok(result.issues.includes('Flatpak file is empty'));
+      } finally {
+        existsSync.mock.restore();
+        statSync.mock.restore();
+      }
+    });
+
+    it('should warn about suspiciously small Flatpak files', () => {
+      const flatpakPath = '/test/app.flatpak';
+      const mockStat = { size: 10 * 1024 * 1024 }; // 10 MB
+
+      const existsSync = mock.method(fs, 'existsSync', mock.fn(() => true));
+      const statSync = mock.method(fs, 'statSync', mock.fn(() => mockStat));
+
+      try {
+        const result = verifyFlatpak(flatpakPath);
+
+        assert.ok(!result.verified, 'Should fail verification for too-small files');
+        assert.ok(result.issues.some((i) => i.includes('too small')));
+      } finally {
+        existsSync.mock.restore();
+        statSync.mock.restore();
+      }
+    });
+
+    it('should accept reasonable Flatpak file sizes', () => {
+      const flatpakPath = '/test/app.flatpak';
+      const mockStat = { size: 133 * 1024 * 1024 }; // 133 MB (typical size)
+
+      const existsSync = mock.method(fs, 'existsSync', mock.fn(() => true));
+      const statSync = mock.method(fs, 'statSync', mock.fn(() => mockStat));
+
+      try {
+        const result = verifyFlatpak(flatpakPath);
+
+        assert.ok(result.verified, 'Should accept reasonable Flatpak file sizes');
+        assert.equal(result.issues.length, 0);
+      } finally {
+        existsSync.mock.restore();
+        statSync.mock.restore();
+      }
+    });
+
+    it('should handle non-existent Flatpak files', () => {
+      const flatpakPath = '/test/nonexistent.flatpak';
+
+      const existsSync = mock.method(fs, 'existsSync', mock.fn(() => false));
+
+      try {
+        const result = verifyFlatpak(flatpakPath);
+
+        assert.ok(!result.verified, 'Should reject non-existent Flatpak files');
+        assert.ok(result.issues.includes('Flatpak file does not exist'));
+      } finally {
+        existsSync.mock.restore();
+      }
+    });
+  });
+
+  describe('AppImage verification', () => {
+    const appImagePath = '/test/Auto-Claude-2.7.5-linux-x86_64.AppImage';
+
+    it('should successfully verify valid AppImage', () => {
+      const mockFiles = [
+        './resources/python',
+        './resources/backend/core/client.py',
+        './resources/python-site-packages/secretstorage/__init__.py',
+        './resources/python-site-packages/pydantic_core/__init__.py',
+        './resources/python-site-packages/claude_agent_sdk/__init__.py',
+        './resources/python-site-packages/dotenv/__init__.py',
+      ];
+
+      const mockFn = (cmd, args) => {
+        if (cmd === 'which' && args[0] === 'bsdtar') {
+          return { status: 0, stdout: '/usr/bin/bsdtar', stderr: '' };
+        }
+        if (cmd === 'bsdtar') {
+          return { status: 0, stdout: mockFiles.join('\n'), stderr: '', error: undefined };
+        }
+        return { status: 1, stderr: 'Unknown command' };
+      };
+
+      const { verifyAppImage } = loadWithMockedSpawnSync(mockFn);
+      const result = verifyAppImage(appImagePath);
+
+      restoreSpawnSync();
+
+      assert.ok(result.verified, 'Should verify valid AppImage');
+      assert.equal(result.issues.length, 0);
+      assert.equal(result.fileCount, mockFiles.length);
+    });
+
+    it('should handle spawn errors (OS-level failures)', () => {
+      const spawnError = new Error('EACCES: permission denied');
+
+      const mockFn = (cmd) => {
+        if (cmd === 'which') {
+          return { status: 0, stdout: '/usr/bin/bsdtar', stderr: '', error: undefined };
+        }
+        if (cmd === 'bsdtar') {
+          return { status: null, stdout: '', stderr: '', error: spawnError };
+        }
+        return { status: 1, stderr: 'Unknown command' };
+      };
+
+      const { verifyAppImage } = loadWithMockedSpawnSync(mockFn);
+      const result = verifyAppImage(appImagePath);
+
+      restoreSpawnSync();
+
+      assert.ok(!result.verified, 'Should fail on spawn error');
+      assert.ok(result.reason.includes('Command execution failed'));
+      assert.ok(result.reason.includes('permission denied'));
+    });
+
+    it('should handle non-zero exit status from bsdtar', () => {
+      const mockFn = (cmd) => {
+        if (cmd === 'which') {
+          return { status: 0, stdout: '/usr/bin/bsdtar', stderr: '', error: undefined };
+        }
+        if (cmd === 'bsdtar') {
+          return { status: 1, stdout: '', stderr: 'bsdtar: Error: Not an AppImage file', error: undefined };
+        }
+        return { status: 1, stderr: 'Unknown command' };
+      };
+
+      const { verifyAppImage } = loadWithMockedSpawnSync(mockFn);
+      const result = verifyAppImage(appImagePath);
+
+      restoreSpawnSync();
+
+      assert.ok(!result.verified, 'Should fail on non-zero status');
+      assert.equal(result.reason, 'Failed to extract file list');
+    });
+
+    it('should handle missing bsdtar tool', () => {
+      const mockFn = (cmd) => {
+        if (cmd === 'which') {
+          return { status: 1, stdout: '', stderr: '', error: undefined };
+        }
+        return { status: 1, stderr: 'Unknown command' };
+      };
+
+      const { verifyAppImage } = loadWithMockedSpawnSync(mockFn);
+      const result = verifyAppImage(appImagePath);
+
+      restoreSpawnSync();
+
+      assert.ok(!result.verified, 'Should fail when bsdtar is missing');
+      assert.equal(result.reason, 'bsdtar not available');
+      assert.ok(result.critical, 'Should be marked as critical');
+    });
+  });
+
+  describe('deb package verification', () => {
+    const debPath = '/test/auto-claude_2.7.5_amd64.deb';
+
+    it('should successfully verify valid deb package', () => {
+      const mockFiles = [
+        'drwxr-xr-x root/root         0 2025-01-01 00:00 ./resources/python',
+        '-rw-r--r-- root/root      1234 2025-01-01 00:00 ./resources/backend/core/client.py',
+        '-rw-r--r-- root/root       567 2025-01-01 00:00 ./resources/python-site-packages/secretstorage/__init__.py',
+        '-rw-r--r-- root/root       789 2025-01-01 00:00 ./resources/python-site-packages/pydantic_core/__init__.py',
+        '-rw-r--r-- root/root       456 2025-01-01 00:00 ./resources/python-site-packages/claude_agent_sdk/__init__.py',
+        '-rw-r--r-- root/root       321 2025-01-01 00:00 ./resources/python-site-packages/dotenv/__init__.py',
+      ];
+
+      const mockFn = (cmd, args) => {
+        if (cmd === 'which' && args[0] === 'dpkg-deb') {
+          return { status: 0, stdout: '/usr/bin/dpkg-deb', stderr: '', error: undefined };
+        }
+        if (cmd === 'dpkg-deb') {
+          return { status: 0, stdout: mockFiles.join('\n'), stderr: '', error: undefined };
+        }
+        return { status: 1, stderr: 'Unknown command' };
+      };
+
+      const { verifyDeb } = loadWithMockedSpawnSync(mockFn);
+      const result = verifyDeb(debPath);
+
+      restoreSpawnSync();
+
+      assert.ok(result.verified, 'Should verify valid deb package');
+      assert.equal(result.issues.length, 0);
+      assert.equal(result.fileCount, mockFiles.length);
+    });
+
+    it('should handle spawn errors (OS-level failures)', () => {
+      const spawnError = new Error('ENOMEM: Cannot allocate memory');
+
+      const mockFn = (cmd) => {
+        if (cmd === 'which') {
+          return { status: 0, stdout: '/usr/bin/dpkg-deb', stderr: '', error: undefined };
+        }
+        if (cmd === 'dpkg-deb') {
+          return { status: null, stdout: '', stderr: '', error: spawnError };
+        }
+        return { status: 1, stderr: 'Unknown command' };
+      };
+
+      const { verifyDeb } = loadWithMockedSpawnSync(mockFn);
+      const result = verifyDeb(debPath);
+
+      restoreSpawnSync();
+
+      assert.ok(!result.verified, 'Should fail on spawn error');
+      assert.ok(result.reason.includes('Command execution failed'));
+      assert.ok(result.reason.includes('Cannot allocate memory'));
+    });
+
+    it('should handle non-zero exit status from dpkg-deb', () => {
+      const mockFn = (cmd) => {
+        if (cmd === 'which') {
+          return { status: 0, stdout: '/usr/bin/dpkg-deb', stderr: '', error: undefined };
+        }
+        if (cmd === 'dpkg-deb') {
+          return { status: 2, stdout: '', stderr: 'dpkg-deb: error: cannot read archive', error: undefined };
+        }
+        return { status: 1, stderr: 'Unknown command' };
+      };
+
+      const { verifyDeb } = loadWithMockedSpawnSync(mockFn);
+      const result = verifyDeb(debPath);
+
+      restoreSpawnSync();
+
+      assert.ok(!result.verified, 'Should fail on non-zero status');
+      assert.equal(result.reason, 'Failed to extract file list');
+    });
+
+    it('should handle missing dpkg-deb tool', () => {
+      const mockFn = (cmd) => {
+        if (cmd === 'which') {
+          return { status: 1, stdout: '', stderr: '', error: undefined };
+        }
+        return { status: 1, stderr: 'Unknown command' };
+      };
+
+      const { verifyDeb } = loadWithMockedSpawnSync(mockFn);
+      const result = verifyDeb(debPath);
+
+      restoreSpawnSync();
+
+      assert.ok(!result.verified, 'Should fail when dpkg-deb is missing');
+      assert.equal(result.reason, 'dpkg-deb not available');
+      assert.ok(result.critical, 'Should be marked as critical');
+    });
+  });
+});
@@ -396,7 +396,7 @@ describe('E2E Smoke Tests', () => {
        statusHandler({}, 'task-001', 'in_progress');
      }

-      expect(statusCallback).toHaveBeenCalledWith('task-001', 'in_progress', undefined);
+      expect(statusCallback).toHaveBeenCalledWith('task-001', 'in_progress', undefined, undefined);

      // Cleanup listeners
      cleanupProgress();
@@ -656,6 +656,7 @@ describe('E2E Smoke Tests', () => {
          index + 1,
          'task-001',
          status,
+          undefined,
          undefined
        );
      });
@@ -6,7 +6,7 @@ import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
 import { mkdirSync, rmSync, existsSync, mkdtempSync } from 'fs';
 import { tmpdir } from 'os';
 import path from 'path';
-import type { ClaudeProfile, IPCResult, TerminalCreateOptions } from '../../shared/types';
+import type { ClaudeProfile, IPCResult, } from '../../shared/types';

 // Test directories - use secure temp directory with random suffix
 let TEST_DIR: string;
@@ -297,7 +297,7 @@ describe('Subprocess Spawn Integration', () => {
      // Simulate stdout data (must include newline for buffered output processing)
      mockStdout.emit('data', Buffer.from('Test log output\n'));

-      expect(logHandler).toHaveBeenCalledWith('task-1', 'Test log output\n');
+      expect(logHandler).toHaveBeenCalledWith('task-1', 'Test log output\n', undefined);
    }, 30000);  // Increase timeout for Windows CI (dynamic imports are slow)

    it('should emit log events from stderr', async () => {
@@ -313,7 +313,7 @@ describe('Subprocess Spawn Integration', () => {
      // Simulate stderr data (must include newline for buffered output processing)
      mockStderr.emit('data', Buffer.from('Progress: 50%\n'));

-      expect(logHandler).toHaveBeenCalledWith('task-1', 'Progress: 50%\n');
+      expect(logHandler).toHaveBeenCalledWith('task-1', 'Progress: 50%\n', undefined);
    }, 30000);  // Increase timeout for Windows CI (dynamic imports are slow)

    it('should emit exit event when process exits', async () => {
@@ -329,8 +329,8 @@ describe('Subprocess Spawn Integration', () => {
      // Simulate process exit
      mockProcess.emit('exit', 0);

-      // Exit event includes taskId, exit code, and process type
-      expect(exitHandler).toHaveBeenCalledWith('task-1', 0, expect.any(String));
+      // Exit event includes taskId, exit code, process type, and optional projectId
+      expect(exitHandler).toHaveBeenCalledWith('task-1', 0, expect.any(String), undefined);
    }, 30000);  // Increase timeout for Windows CI (dynamic imports are slow)

    it('should emit error event when process errors', async () => {
@@ -346,7 +346,7 @@ describe('Subprocess Spawn Integration', () => {
      // Simulate process error
      mockProcess.emit('error', new Error('Spawn failed'));

-      expect(errorHandler).toHaveBeenCalledWith('task-1', 'Spawn failed');
+      expect(errorHandler).toHaveBeenCalledWith('task-1', 'Spawn failed', undefined);
    }, 30000);  // Increase timeout for Windows CI (dynamic imports are slow)

    it('should kill task and remove from tracking', async () => {
@@ -395,19 +395,20 @@ describe('Subprocess Spawn Integration', () => {
        expect(manager.getRunningTasks()).toHaveLength(2);
      }, { timeout: 5000 });

-      // Wait for spawn to complete (ensures exit handlers are attached)
-      await new Promise(resolve => setImmediate(resolve));
+      // Wait for both spawn promises to fully resolve — this ensures the exit
+      // handlers are attached to mockProcess. A single setImmediate is NOT enough
+      // on Windows CI because spawnProcess has async operations (getAPIProfileEnv,
+      // getRecoveryCoordinator) between addProcess and the .on('exit') listener.
+      // Waiting for the promises guarantees spawnProcess has completed fully.
+      await Promise.allSettled([promise1, promise2]);

-      // Emit exit for task-1 (first task's handler)
+      // Both tasks share the same mockProcess, so one emit fires both exit handlers
      mockProcess.emit('exit', 0);
-      await promise1;

-      // Emit exit for task-2 (second task's handler replaces first due to shared mock process)
-      mockProcess.emit('exit', 0);
-      await promise2;
-
-      // Tasks should be removed from tracking after exit
-      expect(manager.getRunningTasks()).toHaveLength(0);
+      // Wait for tasks to be removed from tracking (cleanup may be async)
+      await vi.waitFor(() => {
+        expect(manager.getRunningTasks()).toHaveLength(0);
+      }, { timeout: 5000 });
    }, 15000);

    it('should use configured Python path', async () => {
@@ -241,12 +241,12 @@ describe('Task Lifecycle Integration', () => {
      )?.[1];

      if (eventHandler) {
-        eventHandler({}, 'task-001', 'spec_complete');
+        eventHandler({}, 'task-001', 'spec_complete', undefined, undefined);
      }

-      // Verify callback was invoked with correct parameters (taskId, status, projectId)
-      // Note: projectId is optional and undefined when not provided
-      expect(callback).toHaveBeenCalledWith('task-001', 'spec_complete', undefined);
+      // Verify callback was invoked with correct parameters (taskId, status, projectId, reviewReason)
+      // Note: projectId/reviewReason are optional and undefined when not provided
+      expect(callback).toHaveBeenCalledWith('task-001', 'spec_complete', undefined, undefined);
    });

    it('should emit task:progress event with updated plan during spec creation', async () => {
@@ -379,4 +379,4 @@ describe('Task Lifecycle Integration', () => {
    });
  });

-});
+});
@@ -69,7 +69,7 @@ vi.mock('child_process', () => {
  // so when tests call vi.mocked(execFileSync).mockReturnValue(), it affects execSync too
  const sharedSyncMock = vi.fn();

-const mockExecFile = vi.fn((cmd: unknown, args: unknown, options: unknown, callback: unknown) => {
+const mockExecFile = vi.fn((_cmd: unknown, _args: unknown, _options: unknown, callback: unknown) => {
    // Return a minimal ChildProcess-like object
    const childProcess = {
      stdout: { on: vi.fn() },
@@ -86,7 +86,7 @@ const mockExecFile = vi.fn((cmd: unknown, args: unknown, options: unknown, callb
    return childProcess as unknown as import('child_process').ChildProcess;
  });

-  const mockExec = vi.fn((cmd: unknown, options: unknown, callback: unknown) => {
+  const mockExec = vi.fn((_cmd: unknown, _options: unknown, callback: unknown) => {
    // Return a minimal ChildProcess-like object
    const childProcess = {
      stdout: { on: vi.fn() },
@@ -61,15 +61,17 @@ import path from 'path';
 import { isValidConfigDir } from '../utils/config-path-validator';

 describe('isValidConfigDir - Security Validation', () => {
-  let originalHomedir: string;
-  let consoleWarnSpy: any;
+  let _originalHomedir: string;
+  let consoleWarnSpy: ReturnType<typeof vi.spyOn>;

  beforeEach(() => {
    // Store original homedir for restoration
-    originalHomedir = os.homedir();
+    _originalHomedir = os.homedir();

    // Spy on console.warn to suppress warning output during tests
-    consoleWarnSpy = vi.spyOn(console, 'warn').mockImplementation(() => {});
+    consoleWarnSpy = vi.spyOn(console, 'warn').mockImplementation(() => {
+      /* intentionally empty - suppress console output during tests */
+    });
  });

  afterEach(() => {
@@ -564,7 +564,7 @@ describe("IPC Handlers", { timeout: 30000 }, () => {

      expect(result).toHaveProperty("success", true);
      const data = (result as { data: { theme: string } }).data;
-      expect(data).toHaveProperty("theme", "system");
+      expect(data).toHaveProperty("theme", "dark");
    });
  });

@@ -689,7 +689,8 @@ describe("IPC Handlers", { timeout: 30000 }, () => {
        "task:statusChange",
        "task-1",
        "human_review",
-        expect.any(String) // projectId for multi-project filtering
+        expect.any(String), // projectId for multi-project filtering
+        "errors"
      );
    });
  });
@@ -292,6 +292,7 @@ describe('ProjectStore', () => {
        feature: 'Test Feature',
        workflow_type: 'feature',
        services_involved: [],
+        status: 'in_progress',
        phases: [
          {
            phase: 1,
@@ -338,6 +339,7 @@ describe('ProjectStore', () => {
        feature: 'Pending Feature',
        workflow_type: 'feature',
        services_involved: [],
+        status: 'backlog',
        phases: [
          {
            phase: 1,
@@ -377,6 +379,7 @@ describe('ProjectStore', () => {
        feature: 'Complete Feature',
        workflow_type: 'feature',
        services_involved: [],
+        status: 'ai_review',
        phases: [
          {
            phase: 1,
@@ -408,7 +411,7 @@ describe('ProjectStore', () => {
      expect(tasks[0].status).toBe('ai_review');
    });

-    it('should determine status as human_review when QA report rejected', async () => {
+    it('should determine status as human_review when plan status is human_review', async () => {
      const specsDir = path.join(TEST_PROJECT_PATH, '.auto-claude', 'specs', '004-rejected');
      mkdirSync(specsDir, { recursive: true });

@@ -416,6 +419,8 @@ describe('ProjectStore', () => {
        feature: 'Rejected Feature',
        workflow_type: 'feature',
        services_involved: [],
+        status: 'human_review',
+        reviewReason: 'qa_rejected',
        phases: [
          {
            phase: 1,
@@ -437,11 +442,6 @@ describe('ProjectStore', () => {
        JSON.stringify(plan)
      );

-      writeFileSync(
-        path.join(specsDir, 'qa_report.md'),
-        '# QA Report\n\nStatus: REJECTED\n'
-      );
-
      const { ProjectStore } = await import('../project-store');
      const store = new ProjectStore();

@@ -451,8 +451,7 @@ describe('ProjectStore', () => {
      expect(tasks[0].status).toBe('human_review');
    });

-    it('should determine status as human_review when QA report approved', async () => {
-      // QA approval moves task to human_review (user needs to review before marking done)
+    it('should determine reviewReason from plan when status is human_review', async () => {
      const specsDir = path.join(TEST_PROJECT_PATH, '.auto-claude', 'specs', '005-approved');
      mkdirSync(specsDir, { recursive: true });

@@ -460,6 +459,8 @@ describe('ProjectStore', () => {
        feature: 'Approved Feature',
        workflow_type: 'feature',
        services_involved: [],
+        status: 'human_review',
+        reviewReason: 'completed',
        phases: [
          {
            phase: 1,
@@ -481,11 +482,6 @@ describe('ProjectStore', () => {
        JSON.stringify(plan)
      );

-      writeFileSync(
-        path.join(specsDir, 'qa_report.md'),
-        '# QA Report\n\nStatus: APPROVED\n'
-      );
-
      const { ProjectStore } = await import('../project-store');
      const store = new ProjectStore();

@@ -93,7 +93,7 @@ describe('PythonEnvManager', () => {
      const sitePackagesPath = 'C:\\test\\site-packages';

      // Access private property for testing
-      (manager as any).sitePackagesPath = sitePackagesPath;
+      (manager as unknown as { sitePackagesPath: string }).sitePackagesPath = sitePackagesPath;

      const env = manager.getPythonEnv();

@@ -106,7 +106,7 @@ describe('PythonEnvManager', () => {
      const sitePackagesPath = 'C:\\test\\site-packages';

      // Access private property for testing
-      (manager as any).sitePackagesPath = sitePackagesPath;
+      (manager as unknown as { sitePackagesPath: string }).sitePackagesPath = sitePackagesPath;

      const env = manager.getPythonEnv();

@@ -125,7 +125,7 @@ describe('PythonEnvManager', () => {
      const sitePackagesPath = '/test/site-packages';

      // Access private property for testing
-      (manager as any).sitePackagesPath = sitePackagesPath;
+      (manager as unknown as { sitePackagesPath: string }).sitePackagesPath = sitePackagesPath;

      const env = manager.getPythonEnv();

@@ -144,7 +144,7 @@ describe('PythonEnvManager', () => {
      const sitePackagesPath = 'C:\\test\\site-packages';

      // Access private property for testing
-      (manager as any).sitePackagesPath = sitePackagesPath;
+      (manager as unknown as { sitePackagesPath: string }).sitePackagesPath = sitePackagesPath;

      // Save and clear existing PATH, then set lowercase 'Path'
      // This simulates a Windows environment where the system has 'Path' instead of 'PATH'
@@ -507,7 +507,7 @@ describe('Rate Limit Edge Cases', () => {
      ];

      for (const msg of falsePositives) {
-        const result = detectRateLimit(msg);
+        const _result = detectRateLimit(msg);
        // Note: Some may still match secondary indicators - that's intentional
        // The primary pattern should NOT match these
        const primaryPattern = /Limit reached\s*[·•]\s*resets/i;
@@ -99,7 +99,7 @@ describe('Rate Limit Detector', () => {
    it('should return false for non-rate-limit errors', async () => {
      const { isRateLimitError } = await import('../rate-limit-detector');

-      expect(isRateLimitError('authentication required')).toBe(false);
+      expect(isRateLimitError('[CLI] authentication required')).toBe(false);
      expect(isRateLimitError('Task completed')).toBe(false);
    });
  });
@@ -135,10 +135,10 @@ describe('Auth Failure Detection', () => {
  });

  describe('detectAuthFailure', () => {
-    it('should detect "authentication required" pattern', async () => {
+    it('should detect "authentication required" pattern with bracket prefix', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Error: authentication required';
+      const output = '[CLI] authentication required';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -146,40 +146,40 @@ describe('Auth Failure Detection', () => {
      expect(result.message).toContain('authentication required');
    });

-    it('should detect "authentication is required" pattern', async () => {
+    it('should detect "authentication is required" pattern with bracket prefix', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Authentication is required to proceed';
+      const output = '[Auth] Authentication is required to proceed';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
      expect(result.failureType).toBe('missing');
    });

-    it('should detect "not authenticated" pattern', async () => {
+    it('should detect "not authenticated" pattern with bracket prefix', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Error: not authenticated';
+      const output = '[Error] not authenticated';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
      expect(result.failureType).toBe('missing');
    });

-    it('should detect "not yet authenticated" pattern', async () => {
+    it('should detect "not yet authenticated" pattern with bracket prefix', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'You are not yet authenticated';
+      const output = '[CLI] not yet authenticated';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
      expect(result.failureType).toBe('missing');
    });

-    it('should detect "login required" pattern', async () => {
+    it('should detect "login required" pattern with bracket prefix', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Login required';
+      const output = '[CLI] Login required';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -189,7 +189,7 @@ describe('Auth Failure Detection', () => {
    it('should detect "oauth token invalid" pattern', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'OAuth token is invalid';
+      const output = 'Error: invalid token';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -199,7 +199,7 @@ describe('Auth Failure Detection', () => {
    it('should detect "oauth token expired" pattern', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'OAuth token expired';
+      const output = 'OAuth token has expired';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -209,7 +209,7 @@ describe('Auth Failure Detection', () => {
    it('should detect "oauth token missing" pattern', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'OAuth token missing';
+      const output = '[CLI] authentication required - OAuth token missing';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -219,39 +219,39 @@ describe('Auth Failure Detection', () => {
    it('should detect "unauthorized" pattern', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Error: Unauthorized';
+      const output = 'Error: unauthorized access';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
      expect(result.failureType).toBe('invalid');
    });

-    it('should detect "please log in" pattern', async () => {
+    it('should detect "please log in" pattern with CLI format', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Please log in to continue';
+      const output = '· Please run /login to continue';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
-      // "please log in" doesn't contain 'required' keyword, so classified as 'unknown'
+      // Contains 'login' pattern
      expect(result.failureType).toBeDefined();
    });

-    it('should detect "please authenticate" pattern', async () => {
+    it('should detect "please authenticate" pattern with Error prefix', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Please authenticate before proceeding';
+      const output = 'Error: authentication required before proceeding';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
-      // "please authenticate" doesn't contain 'required' keyword, so classified as 'unknown'
+      // "Error: ... authentication" matches the pattern
      expect(result.failureType).toBeDefined();
    });

    it('should detect "invalid credentials" pattern', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Invalid credentials provided';
+      const output = 'Error: invalid token credentials provided';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -261,26 +261,26 @@ describe('Auth Failure Detection', () => {
    it('should detect "invalid token" pattern', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Invalid token';
+      const output = 'Error: invalid token';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
      expect(result.failureType).toBe('invalid');
    });

-    it('should detect "auth failed" pattern', async () => {
+    it('should detect "auth failed" pattern with authentication_error type', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Auth failed';
+      const output = '{"type":"authentication_error","message":"Auth failed"}';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
    });

-    it('should detect "authentication error" pattern', async () => {
+    it('should detect "authentication error" pattern with JSON type', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Authentication error occurred';
+      const output = '{"type": "authentication_error", "message": "Authentication error occurred"}';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -289,7 +289,7 @@ describe('Auth Failure Detection', () => {
    it('should detect "session expired" pattern', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Your session expired';
+      const output = 'Please obtain a new token - your session expired';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -299,7 +299,7 @@ describe('Auth Failure Detection', () => {
    it('should detect "access denied" pattern', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Access denied';
+      const output = 'status: 401 - Access denied';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -309,7 +309,7 @@ describe('Auth Failure Detection', () => {
    it('should detect "permission denied" pattern', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Permission denied';
+      const output = 'HTTP 401 - Permission denied';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -329,7 +329,7 @@ describe('Auth Failure Detection', () => {
    it('should detect "credentials missing" pattern', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Credentials are missing';
+      const output = '[CLI] authentication required - credentials are missing';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -339,7 +339,7 @@ describe('Auth Failure Detection', () => {
    it('should detect "credentials expired" pattern', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = 'Credentials expired';
+      const output = 'Please refresh your existing token - credentials expired';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -375,7 +375,7 @@ describe('Auth Failure Detection', () => {
    it('should include profile ID in result', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const result = detectAuthFailure('authentication required', 'custom-profile');
+      const result = detectAuthFailure('[CLI] authentication required', 'custom-profile');

      expect(result.isAuthFailure).toBe(true);
      expect(result.profileId).toBe('custom-profile');
@@ -384,7 +384,7 @@ describe('Auth Failure Detection', () => {
    it('should use active profile ID when not specified', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const result = detectAuthFailure('authentication required');
+      const result = detectAuthFailure('[Auth] authentication required');

      expect(result.isAuthFailure).toBe(true);
      expect(result.profileId).toBe('test-profile-id');
@@ -403,7 +403,7 @@ describe('Auth Failure Detection', () => {
    it('should provide user-friendly message for missing auth', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const result = detectAuthFailure('authentication required');
+      const result = detectAuthFailure('[CLI] authentication required');

      expect(result.isAuthFailure).toBe(true);
      expect(result.message).toContain('Settings');
@@ -413,7 +413,7 @@ describe('Auth Failure Detection', () => {
    it('should provide user-friendly message for expired auth', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const result = detectAuthFailure('session expired');
+      const result = detectAuthFailure('Please obtain a new token - session expired');

      expect(result.isAuthFailure).toBe(true);
      expect(result.message).toContain('expired');
@@ -423,7 +423,7 @@ describe('Auth Failure Detection', () => {
    it('should provide user-friendly message for invalid auth', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const result = detectAuthFailure('unauthorized');
+      const result = detectAuthFailure('Error: unauthorized');

      expect(result.isAuthFailure).toBe(true);
      expect(result.message).toContain('Invalid');
@@ -434,10 +434,10 @@ describe('Auth Failure Detection', () => {
    it('should return true for auth failure errors', async () => {
      const { isAuthFailureError } = await import('../rate-limit-detector');

-      expect(isAuthFailureError('authentication required')).toBe(true);
-      expect(isAuthFailureError('not authenticated')).toBe(true);
-      expect(isAuthFailureError('unauthorized')).toBe(true);
-      expect(isAuthFailureError('invalid token')).toBe(true);
+      expect(isAuthFailureError('[CLI] authentication required')).toBe(true);
+      expect(isAuthFailureError('[Auth] not authenticated')).toBe(true);
+      expect(isAuthFailureError('Error: unauthorized')).toBe(true);
+      expect(isAuthFailureError('Error: invalid token')).toBe(true);
    });

    it('should return false for non-auth-failure errors', async () => {
@@ -454,12 +454,12 @@ describe('Auth Failure Detection', () => {
      const { detectRateLimit } = await import('../rate-limit-detector');

      const authErrors = [
-        'authentication required',
-        'not authenticated',
-        'unauthorized',
-        'invalid token',
-        'session expired',
-        'please log in'
+        '[CLI] authentication required',
+        '[Auth] not authenticated',
+        'Error: unauthorized',
+        'Error: invalid token',
+        'Please obtain a new token - session expired',
+        '· Please run /login'
      ];

      for (const error of authErrors) {
@@ -503,12 +503,12 @@ Please authenticate and try again.`;
      const { detectAuthFailure } = await import('../rate-limit-detector');

      const testCases = [
-        'AUTHENTICATION REQUIRED',
-        'Authentication Required',
-        'UNAUTHORIZED',
-        'Unauthorized',
-        'NOT AUTHENTICATED',
-        'Not Authenticated'
+        '[CLI] AUTHENTICATION REQUIRED',
+        '[Auth] Authentication Required',
+        'ERROR: UNAUTHORIZED',
+        'Error: Unauthorized',
+        '[API] NOT AUTHENTICATED',
+        '[Error] Not Authenticated'
      ];

      for (const output of testCases) {
@@ -538,7 +538,7 @@ Please authenticate and try again.`;
    it('should handle JSON error responses', async () => {
      const { detectAuthFailure } = await import('../rate-limit-detector');

-      const output = '{"error": "unauthorized", "message": "Please authenticate"}';
+      const output = '{"type":"authentication_error", "error": "unauthorized", "message": "Please authenticate"}';
      const result = detectAuthFailure(output);

      expect(result.isAuthFailure).toBe(true);
@@ -609,3 +609,392 @@ Please authenticate and try again.`;
    });
  });
 });
+
+describe('Billing Failure Detection', () => {
+  beforeEach(() => {
+    vi.resetModules();
+  });
+
+  afterEach(() => {
+    vi.clearAllMocks();
+  });
+
+  describe('detectBillingFailure - spec appendix messages', () => {
+    it('should detect "Your credit balance is too low to access the Anthropic API"', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('Your credit balance is too low to access the Anthropic API');
+
+      expect(result.isBillingFailure).toBe(true);
+      expect(result.failureType).toBe('insufficient_credits');
+      expect(result.message).toBeDefined();
+    });
+
+    it('should detect "insufficient credits"', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('insufficient credits');
+
+      expect(result.isBillingFailure).toBe(true);
+      expect(result.failureType).toBe('insufficient_credits');
+    });
+
+    it('should detect "billing error"', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('billing error');
+
+      expect(result.isBillingFailure).toBe(true);
+      expect(result.failureType).toBe('payment_required');
+    });
+
+    it('should detect "extra_usage exceeded"', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('extra_usage exceeded');
+
+      expect(result.isBillingFailure).toBe(true);
+      expect(result.failureType).toBe('insufficient_credits');
+    });
+
+    it('should detect standalone "extra_usage"', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('extra_usage');
+
+      expect(result.isBillingFailure).toBe(true);
+    });
+
+    it('should detect "payment required"', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('payment required');
+
+      expect(result.isBillingFailure).toBe(true);
+      expect(result.failureType).toBe('payment_required');
+    });
+
+    it('should detect "subscription expired"', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('subscription expired');
+
+      expect(result.isBillingFailure).toBe(true);
+      expect(result.failureType).toBe('subscription_inactive');
+    });
+
+    it('should detect credit balance variations', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const messages = [
+        'credit balance is insufficient',
+        'credit balance is empty',
+        'credit balance is zero',
+        'credit balance is exhausted',
+        'credit balance is too low',
+      ];
+
+      for (const msg of messages) {
+        const result = detectBillingFailure(msg);
+        expect(result.isBillingFailure).toBe(true);
+      }
+    });
+  });
+
+  describe('detectBillingFailure - negative cases', () => {
+    it('should return false for normal output', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const normalMessages = [
+        'Task completed successfully',
+        'Processing 100 files',
+        'Build succeeded',
+        'Deploying to production',
+        '',
+      ];
+
+      for (const msg of normalMessages) {
+        const result = detectBillingFailure(msg);
+        expect(result.isBillingFailure).toBe(false);
+      }
+    });
+
+    it('should NOT detect rate limit errors as billing failures', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const rateLimitMessages = [
+        'Limit reached · resets Dec 17 at 6am',
+        'rate limit exceeded',
+        'too many requests',
+      ];
+
+      for (const msg of rateLimitMessages) {
+        const result = detectBillingFailure(msg);
+        expect(result.isBillingFailure).toBe(false);
+      }
+    });
+
+    it('should NOT detect auth errors as billing failures', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const authMessages = [
+        'authentication required',
+        'not authenticated',
+        'unauthorized',
+        'invalid token',
+        'session expired',
+      ];
+
+      for (const msg of authMessages) {
+        const result = detectBillingFailure(msg);
+        expect(result.isBillingFailure).toBe(false);
+      }
+    });
+
+    it('should NOT match "line 402" as billing failure (false positive check)', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const falsePositives = [
+        'line 402',
+        'Processing line 402 of 1000',
+        'Found 4020 records',
+        'Error on line 402',
+        'The user has 402 files',
+      ];
+
+      for (const msg of falsePositives) {
+        const result = detectBillingFailure(msg);
+        expect(result.isBillingFailure).toBe(false);
+      }
+    });
+  });
+
+  describe('detectBillingFailure - 402 with proper context', () => {
+    it('should detect "HTTP 402 Payment Required"', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('HTTP 402 Payment Required');
+
+      expect(result.isBillingFailure).toBe(true);
+    });
+
+    it('should detect "API Error: 402"', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('API Error: 402');
+
+      expect(result.isBillingFailure).toBe(true);
+    });
+
+    it('should detect "status: 402"', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('status: 402');
+
+      expect(result.isBillingFailure).toBe(true);
+    });
+
+    it('should detect "error 402"', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('error 402');
+
+      expect(result.isBillingFailure).toBe(true);
+    });
+
+    it('should detect "402 payment required" (lowercase)', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('402 payment required');
+
+      expect(result.isBillingFailure).toBe(true);
+    });
+  });
+
+  describe('isBillingFailureError', () => {
+    it('should return true for billing failure errors', async () => {
+      const { isBillingFailureError } = await import('../rate-limit-detector');
+
+      expect(isBillingFailureError('credit balance is too low')).toBe(true);
+      expect(isBillingFailureError('insufficient credits')).toBe(true);
+      expect(isBillingFailureError('billing error')).toBe(true);
+      expect(isBillingFailureError('extra_usage exceeded')).toBe(true);
+    });
+
+    it('should return false for non-billing errors', async () => {
+      const { isBillingFailureError } = await import('../rate-limit-detector');
+
+      expect(isBillingFailureError('Task completed')).toBe(false);
+      expect(isBillingFailureError('')).toBe(false);
+      expect(isBillingFailureError('rate limit exceeded')).toBe(false);
+    });
+  });
+
+  describe('classifyBillingFailureType via detectBillingFailure', () => {
+    it('should classify credit-related failures as insufficient_credits', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const creditMessages = [
+        'Your credit balance is too low',
+        'insufficient credits',
+        'out of credits',
+        'extra_usage exceeded',
+      ];
+
+      for (const msg of creditMessages) {
+        const result = detectBillingFailure(msg);
+        expect(result.isBillingFailure).toBe(true);
+        expect(result.failureType).toBe('insufficient_credits');
+      }
+    });
+
+    it('should classify payment failures as payment_required', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const paymentMessages = [
+        'payment required',
+        'billing error',
+        'billing failure',
+      ];
+
+      for (const msg of paymentMessages) {
+        const result = detectBillingFailure(msg);
+        expect(result.isBillingFailure).toBe(true);
+        expect(result.failureType).toBe('payment_required');
+      }
+    });
+
+    it('should classify subscription failures as subscription_inactive', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const subscriptionMessages = [
+        'subscription expired',
+        'subscription inactive',
+        'subscription cancelled',
+        'account suspended',
+      ];
+
+      for (const msg of subscriptionMessages) {
+        const result = detectBillingFailure(msg);
+        expect(result.isBillingFailure).toBe(true);
+        expect(result.failureType).toBe('subscription_inactive');
+      }
+    });
+  });
+
+  describe('detectBillingFailure - result structure', () => {
+    it('should include profile ID in result', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('credit balance is too low', 'custom-profile');
+
+      expect(result.isBillingFailure).toBe(true);
+      expect(result.profileId).toBe('custom-profile');
+    });
+
+    it('should use active profile ID when not specified', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('credit balance is too low');
+
+      expect(result.isBillingFailure).toBe(true);
+      expect(result.profileId).toBe('test-profile-id');
+    });
+
+    it('should include original error in result', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const output = 'Error: credit balance is too low to access the API';
+      const result = detectBillingFailure(output);
+
+      expect(result.isBillingFailure).toBe(true);
+      expect(result.originalError).toBe(output);
+    });
+
+    it('should include user-friendly message', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const result = detectBillingFailure('credit balance is too low');
+
+      expect(result.isBillingFailure).toBe(true);
+      expect(result.message).toContain('credit');
+      expect(result.message).toContain('Settings');
+    });
+  });
+
+  describe('cross-detection tests', () => {
+    it('should not detect billing errors as auth failures', async () => {
+      const { detectAuthFailure } = await import('../rate-limit-detector');
+
+      const billingMessages = [
+        'credit balance is too low',
+        'insufficient credits',
+        'billing error',
+        'extra_usage exceeded',
+        'payment required',
+      ];
+
+      for (const msg of billingMessages) {
+        const result = detectAuthFailure(msg);
+        expect(result.isAuthFailure).toBe(false);
+      }
+    });
+
+    it('should not detect billing errors as rate limits', async () => {
+      const { detectRateLimit } = await import('../rate-limit-detector');
+
+      const billingMessages = [
+        'credit balance is too low',
+        'insufficient credits',
+        'billing error',
+        'extra_usage exceeded',
+      ];
+
+      for (const msg of billingMessages) {
+        const result = detectRateLimit(msg);
+        expect(result.isRateLimited).toBe(false);
+      }
+    });
+  });
+
+  describe('edge cases', () => {
+    it('should handle case-insensitive billing messages', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const testCases = [
+        'CREDIT BALANCE IS TOO LOW',
+        'Credit Balance Is Too Low',
+        'BILLING ERROR',
+        'Insufficient Credits',
+        'EXTRA_USAGE EXCEEDED',
+      ];
+
+      for (const output of testCases) {
+        const result = detectBillingFailure(output);
+        expect(result.isBillingFailure).toBe(true);
+      }
+    });
+
+    it('should handle multiline output with billing failure', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const output = `Starting API call...
+Processing request...
+Error: Your credit balance is too low to access the Anthropic API
+Please add credits to continue.`;
+
+      const result = detectBillingFailure(output);
+
+      expect(result.isBillingFailure).toBe(true);
+    });
+
+    it('should handle JSON error responses with billing failure', async () => {
+      const { detectBillingFailure } = await import('../rate-limit-detector');
+
+      const output = '{"type":"billing_error","message":"Insufficient credits"}';
+      const result = detectBillingFailure(output);
+
+      expect(result.isBillingFailure).toBe(true);
+    });
+  });
+});
@@ -243,24 +243,31 @@ describe('SETTINGS_CLAUDE_CODE_GET_ONBOARDING_STATUS handler', () => {
      // Get the mocked functions (vitest stores the implementation)
      const { existsSync, readFileSync } = await import('fs');

+      // Cast to vi.Mock for accessing mock methods
+      type MockFn = ReturnType<typeof vi.fn>;
+      const existsSyncMock = existsSync as unknown as MockFn;
+      const readFileSyncMock = readFileSync as unknown as MockFn;
+
      // Save original mock implementations to restore after test
-      const originalExistsSync = (existsSync as any).getMockImplementation();
-      const originalReadFileSync = (readFileSync as any).getMockImplementation();
+      type ExistsSyncFn = (path: string) => boolean;
+      type ReadFileSyncFn = (path: string) => string;
+      const originalExistsSync: ExistsSyncFn = (existsSyncMock.getMockImplementation() as ExistsSyncFn | undefined) ?? (() => false);
+      const originalReadFileSync: ReadFileSyncFn = (readFileSyncMock.getMockImplementation() as ReadFileSyncFn | undefined) ?? (() => '');

      // Override existsSync to make file appear to exist
-      (existsSync as any).mockImplementation((path: string) => {
+      existsSyncMock.mockImplementation((path: string) => {
        if (path === claudeJsonPath) {
          return true; // File appears to exist
        }
-        return originalExistsSync ? originalExistsSync(path) : false;
+        return originalExistsSync(path);
      });

      // Override readFileSync to throw error for our specific file
-      (readFileSync as any).mockImplementation((path: string) => {
+      readFileSyncMock.mockImplementation((path: string) => {
        if (path === claudeJsonPath) {
          throw new Error('EACCES: permission denied, open \'' + path + '\'');
        }
-        return originalReadFileSync ? originalReadFileSync(path) : '';
+        return originalReadFileSync(path);
      });

      const result = await onboardingStatusHandler({}, null) as {
@@ -272,8 +279,8 @@ describe('SETTINGS_CLAUDE_CODE_GET_ONBOARDING_STATUS handler', () => {
      expect(result.data?.hasCompletedOnboarding).toBe(false);

      // Restore original mocks
-      (existsSync as any).mockImplementation(originalExistsSync);
-      (readFileSync as any).mockImplementation(originalReadFileSync);
+      existsSyncMock.mockImplementation(originalExistsSync);
+      readFileSyncMock.mockImplementation(originalReadFileSync);
    });
  });
 });
@@ -0,0 +1,521 @@
+import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
+import { TaskStateManager } from '../task-state-manager';
+import type { Task, Project } from '../../shared/types';
+
+// Mock dependencies
+vi.mock('../ipc-handlers/utils', () => ({
+  safeSendToRenderer: vi.fn()
+}));
+
+vi.mock('../ipc-handlers/task/plan-file-utils', () => ({
+  getPlanPath: vi.fn(() => '/mock/path/implementation_plan.json'),
+  persistPlanStatusAndReasonSync: vi.fn()
+}));
+
+vi.mock('../worktree-paths', () => ({
+  findTaskWorktree: vi.fn(() => null)
+}));
+
+vi.mock('fs', () => ({
+  existsSync: vi.fn(() => false)
+}));
+
+// Create mock task and project
+function createMockTask(overrides: Partial<Task> = {}): Task {
+  return {
+    id: 'test-task-id',
+    specId: '001-test-spec',
+    projectId: 'test-project-id',
+    title: 'Test Task',
+    description: 'Test description',
+    status: 'backlog',
+    subtasks: [],
+    logs: [],
+    createdAt: new Date(),
+    updatedAt: new Date(),
+    ...overrides
+  };
+}
+
+function createMockProject(overrides: Partial<Project> = {}): Project {
+  return {
+    id: 'test-project-id',
+    name: 'Test Project',
+    path: '/mock/project/path',
+    createdAt: new Date().toISOString(),
+    lastOpenedAt: new Date().toISOString(),
+    ...overrides
+  } as Project;
+}
+
+describe('TaskStateManager', () => {
+  let manager: TaskStateManager;
+  let mockTask: Task;
+  let mockProject: Project;
+
+  beforeEach(() => {
+    manager = new TaskStateManager();
+    mockTask = createMockTask();
+    mockProject = createMockProject();
+    vi.clearAllMocks();
+  });
+
+  afterEach(() => {
+    manager.clearAllTasks();
+  });
+
+  describe('handleTaskEvent', () => {
+    it('should accept events with increasing sequence numbers', () => {
+      const event1 = {
+        type: 'PLANNING_STARTED',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 0
+      };
+
+      const event2 = {
+        type: 'PLANNING_COMPLETE',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-2',
+        sequence: 1,
+        hasSubtasks: true,
+        subtaskCount: 1,
+        requireReviewBeforeCoding: false
+      };
+
+      const result1 = manager.handleTaskEvent(mockTask.id, event1, mockTask, mockProject);
+      const result2 = manager.handleTaskEvent(mockTask.id, event2, mockTask, mockProject);
+
+      expect(result1).toBe(true);
+      expect(result2).toBe(true);
+    });
+
+    it('should reject events with stale sequence numbers', () => {
+      const event1 = {
+        type: 'PLANNING_STARTED',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 5
+      };
+
+      const event2 = {
+        type: 'PLANNING_COMPLETE',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-2',
+        sequence: 3, // Older than event1
+        hasSubtasks: true,
+        subtaskCount: 1,
+        requireReviewBeforeCoding: false
+      };
+
+      const result1 = manager.handleTaskEvent(mockTask.id, event1, mockTask, mockProject);
+      const result2 = manager.handleTaskEvent(mockTask.id, event2, mockTask, mockProject);
+
+      expect(result1).toBe(true);
+      expect(result2).toBe(false); // Should be rejected
+    });
+
+    it('should accept events with equal sequence numbers (edge case)', () => {
+      // This handles reload scenarios where we might see the same sequence
+      const event1 = {
+        type: 'PLANNING_STARTED',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 5
+      };
+
+      const event2 = {
+        type: 'PLANNING_COMPLETE',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-2',
+        sequence: 5, // Same as event1
+        hasSubtasks: true,
+        subtaskCount: 1,
+        requireReviewBeforeCoding: false
+      };
+
+      const result1 = manager.handleTaskEvent(mockTask.id, event1, mockTask, mockProject);
+      const result2 = manager.handleTaskEvent(mockTask.id, event2, mockTask, mockProject);
+
+      expect(result1).toBe(true);
+      expect(result2).toBe(true); // Should be accepted (>= comparison)
+    });
+  });
+
+  describe('handleUiEvent', () => {
+    it('should send PLAN_APPROVED event correctly', () => {
+      // First, set up the task in plan_review state
+      const planningEvent = {
+        type: 'PLANNING_STARTED',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 0
+      };
+
+      const planCompleteEvent = {
+        type: 'PLANNING_COMPLETE',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-2',
+        sequence: 1,
+        hasSubtasks: true,
+        subtaskCount: 1,
+        requireReviewBeforeCoding: true // This will cause plan_review state
+      };
+
+      manager.handleTaskEvent(mockTask.id, planningEvent, mockTask, mockProject);
+      manager.handleTaskEvent(mockTask.id, planCompleteEvent, mockTask, mockProject);
+
+      // Now send PLAN_APPROVED
+      manager.handleUiEvent(mockTask.id, { type: 'PLAN_APPROVED' }, mockTask, mockProject);
+
+      // The actor should now be in 'coding' state
+      // We can't easily check the state directly, but we can verify no errors occurred
+    });
+
+    it('should send USER_STOPPED event correctly', () => {
+      const event = {
+        type: 'PLANNING_STARTED',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 0
+      };
+
+      manager.handleTaskEvent(mockTask.id, event, mockTask, mockProject);
+      manager.handleUiEvent(mockTask.id, { type: 'USER_STOPPED', hasPlan: false }, mockTask, mockProject);
+
+      // Should not throw
+    });
+  });
+
+  describe('handleManualStatusChange', () => {
+    it('should handle done status', () => {
+      const result = manager.handleManualStatusChange(mockTask.id, 'done', mockTask, mockProject);
+      expect(result).toBe(true);
+    });
+
+    it('should handle pr_created status', () => {
+      const taskWithPrUrl = createMockTask({ metadata: { prUrl: 'https://github.com/test/pr/1' } });
+      const result = manager.handleManualStatusChange(mockTask.id, 'pr_created', taskWithPrUrl, mockProject);
+      expect(result).toBe(true);
+    });
+
+    it('should handle in_progress status with plan_review', () => {
+      const taskInPlanReview = createMockTask({
+        status: 'human_review',
+        reviewReason: 'plan_review'
+      });
+      const result = manager.handleManualStatusChange(mockTask.id, 'in_progress', taskInPlanReview, mockProject);
+      expect(result).toBe(true);
+    });
+
+    it('should handle in_progress status without plan_review', () => {
+      const stoppedTask = createMockTask({
+        status: 'human_review',
+        reviewReason: 'stopped'
+      });
+      const result = manager.handleManualStatusChange(mockTask.id, 'in_progress', stoppedTask, mockProject);
+      expect(result).toBe(true);
+    });
+
+    it('should handle backlog status', () => {
+      const result = manager.handleManualStatusChange(mockTask.id, 'backlog', mockTask, mockProject);
+      expect(result).toBe(true);
+    });
+
+    it('should handle human_review status (stage-only merge keeps task in review)', () => {
+      const taskInReview = createMockTask({
+        status: 'human_review',
+        reviewReason: 'completed'
+      });
+      const result = manager.handleManualStatusChange(mockTask.id, 'human_review', taskInReview, mockProject);
+      expect(result).toBe(true);
+    });
+
+    it('should handle human_review with default reviewReason when task has none', () => {
+      const taskNoReason = createMockTask({
+        status: 'human_review'
+        // no reviewReason set
+      });
+      const result = manager.handleManualStatusChange(mockTask.id, 'human_review', taskNoReason, mockProject);
+      expect(result).toBe(true);
+    });
+
+    it('should return false for unhandled status', () => {
+      const result = manager.handleManualStatusChange(mockTask.id, 'ai_review', mockTask, mockProject);
+      expect(result).toBe(false);
+    });
+  });
+
+  describe('sequence management', () => {
+    it('should set and get last sequence', () => {
+      manager.setLastSequence(mockTask.id, 42);
+      expect(manager.getLastSequence(mockTask.id)).toBe(42);
+    });
+
+    it('should return undefined for unknown task', () => {
+      expect(manager.getLastSequence('unknown-task')).toBeUndefined();
+    });
+  });
+
+  describe('clearTask', () => {
+    it('should clear task state', () => {
+      // Set up some state
+      manager.setLastSequence(mockTask.id, 10);
+
+      const event = {
+        type: 'PLANNING_STARTED',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 0
+      };
+      manager.handleTaskEvent(mockTask.id, event, mockTask, mockProject);
+
+      // Clear
+      manager.clearTask(mockTask.id);
+
+      // Verify cleared
+      expect(manager.getLastSequence(mockTask.id)).toBeUndefined();
+    });
+  });
+
+  describe('clearAllTasks', () => {
+    it('should clear all task state', () => {
+      // Set up state for multiple tasks
+      manager.setLastSequence('task-1', 10);
+      manager.setLastSequence('task-2', 20);
+
+      const event1 = {
+        type: 'PLANNING_STARTED',
+        taskId: 'task-1',
+        specId: '001',
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 0
+      };
+      const event2 = {
+        type: 'PLANNING_STARTED',
+        taskId: 'task-2',
+        specId: '002',
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-2',
+        sequence: 0
+      };
+
+      manager.handleTaskEvent('task-1', event1, createMockTask({ id: 'task-1' }), mockProject);
+      manager.handleTaskEvent('task-2', event2, createMockTask({ id: 'task-2' }), mockProject);
+
+      // Clear all
+      manager.clearAllTasks();
+
+      // Verify actors and state are cleared, but sequence tracking is preserved
+      // (to prevent duplicate event processing during refresh window)
+      expect(manager.getLastSequence('task-1')).toBe(10);
+      expect(manager.getLastSequence('task-2')).toBe(20);
+    });
+  });
+
+  describe('handleProcessExited', () => {
+    it('should NOT mark as error if terminal event was already seen', () => {
+      // First send a terminal event (like QA_PASSED)
+      const qaPassedEvent = {
+        type: 'QA_PASSED',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 0,
+        iteration: 1,
+        testsRun: {}
+      };
+
+      manager.handleTaskEvent(mockTask.id, qaPassedEvent, mockTask, mockProject);
+
+      // Now process exits - this should NOT trigger error state
+      manager.handleProcessExited(mockTask.id, 0, mockTask, mockProject);
+
+      // Should not throw and should not transition to error
+      // (We can't easily verify the state, but the important thing is no crash)
+    });
+
+    it('should mark as error on unexpected exit when no terminal event seen', () => {
+      // Start a task
+      const planningEvent = {
+        type: 'PLANNING_STARTED',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 0
+      };
+
+      manager.handleTaskEvent(mockTask.id, planningEvent, mockTask, mockProject);
+
+      // Process exits unexpectedly (no terminal event seen)
+      manager.handleProcessExited(mockTask.id, 1, mockTask, mockProject);
+
+      // Should have sent PROCESS_EXITED event with unexpected=true
+      // This should transition to error state
+    });
+
+    it('should NOT mark exit code 0 as unexpected (plan_review stays intact)', () => {
+      // Simulate: PLANNING_STARTED → PLANNING_COMPLETE (requireReview) → process exits code 0
+      const planningStarted = {
+        type: 'PLANNING_STARTED',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 0
+      };
+
+      const planningComplete = {
+        type: 'PLANNING_COMPLETE',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-2',
+        sequence: 1,
+        hasSubtasks: false,
+        subtaskCount: 0,
+        requireReviewBeforeCoding: true
+      };
+
+      manager.handleTaskEvent(mockTask.id, planningStarted, mockTask, mockProject);
+      manager.handleTaskEvent(mockTask.id, planningComplete, mockTask, mockProject);
+
+      // XState should be in plan_review now
+      expect(manager.getCurrentState(mockTask.id)).toBe('plan_review');
+
+      // Process exits with code 0 - should NOT transition to error
+      manager.handleProcessExited(mockTask.id, 0, mockTask, mockProject);
+
+      // PLANNING_COMPLETE is a terminal event, so handleProcessExited should skip entirely
+      // Task should remain in plan_review
+      expect(manager.getCurrentState(mockTask.id)).toBe('plan_review');
+    });
+
+    it('should treat PLANNING_COMPLETE as a terminal event', () => {
+      // PLANNING_COMPLETE should prevent handleProcessExited from running
+      const planningStarted = {
+        type: 'PLANNING_STARTED',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 0
+      };
+
+      const planningComplete = {
+        type: 'PLANNING_COMPLETE',
+        taskId: mockTask.id,
+        specId: mockTask.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-2',
+        sequence: 1,
+        hasSubtasks: true,
+        subtaskCount: 3,
+        requireReviewBeforeCoding: false
+      };
+
+      manager.handleTaskEvent(mockTask.id, planningStarted, mockTask, mockProject);
+      manager.handleTaskEvent(mockTask.id, planningComplete, mockTask, mockProject);
+
+      // XState should be in coding (no review required)
+      expect(manager.getCurrentState(mockTask.id)).toBe('coding');
+
+      // Process exits with code 1 - should still skip because PLANNING_COMPLETE is terminal
+      manager.handleProcessExited(mockTask.id, 1, mockTask, mockProject);
+
+      // Task should remain in coding, NOT transition to error
+      expect(manager.getCurrentState(mockTask.id)).toBe('coding');
+    });
+  });
+
+  describe('actor state restoration', () => {
+    it('should restore actor state from task with in_progress status', () => {
+      const taskInProgress = createMockTask({
+        status: 'in_progress',
+        executionProgress: { phase: 'coding', phaseProgress: 50, overallProgress: 50 }
+      });
+
+      const event = {
+        type: 'QA_STARTED',
+        taskId: taskInProgress.id,
+        specId: taskInProgress.specId,
+        projectId: mockProject.id,
+        timestamp: new Date().toISOString(),
+        eventId: 'evt-1',
+        sequence: 0,
+        iteration: 1,
+        maxIterations: 3
+      };
+
+      // This should create an actor restored to 'coding' state, then transition to 'qa_review'
+      manager.handleTaskEvent(taskInProgress.id, event, taskInProgress, mockProject);
+
+      // No error should occur
+    });
+
+    it('should restore actor state from task with human_review/plan_review', () => {
+      const taskInPlanReview = createMockTask({
+        status: 'human_review',
+        reviewReason: 'plan_review'
+      });
+
+      // Actor should be created in plan_review state
+      manager.handleUiEvent(taskInPlanReview.id, { type: 'PLAN_APPROVED' }, taskInPlanReview, mockProject);
+
+      // Should transition from plan_review to coding without error
+    });
+
+    it('should restore actor state from task with error status', () => {
+      const taskInError = createMockTask({
+        status: 'error',
+        reviewReason: 'errors'
+      });
+
+      // Actor should be created in error state
+      manager.handleUiEvent(taskInError.id, { type: 'USER_RESUMED' }, taskInError, mockProject);
+
+      // Should transition from error to coding without error
+    });
+  });
+});
@@ -178,7 +178,9 @@ describe("safeSendToRenderer", () => {

  describe("error handling - non-disposal errors", () => {
    it("catches other errors and returns false", () => {
-      const consoleErrorSpy = vi.spyOn(console, "error").mockImplementation(() => {});
+      const consoleErrorSpy = vi.spyOn(console, "error").mockImplementation(() => {
+        /* intentionally empty - suppress console output during tests */
+      });

      mockSend.mockImplementation(() => {
        throw new Error("Some other IPC error");
@@ -216,7 +218,9 @@ describe("safeSendToRenderer", () => {
    });

    it("logs console.warn only once for multiple consecutive calls to same channel", () => {
-      const consoleWarnSpy = vi.spyOn(console, "warn").mockImplementation(() => {});
+      const consoleWarnSpy = vi.spyOn(console, "warn").mockImplementation(() => {
+        /* intentionally empty - suppress console output during tests */
+      });

      mockWindow = {
        isDestroyed: vi.fn(() => true),
@@ -242,7 +246,9 @@ describe("safeSendToRenderer", () => {
    });

    it("logs console.warn separately for different channels", () => {
-      const consoleWarnSpy = vi.spyOn(console, "warn").mockImplementation(() => {});
+      const consoleWarnSpy = vi.spyOn(console, "warn").mockImplementation(() => {
+        /* intentionally empty - suppress console output during tests */
+      });

      mockWindow = {
        isDestroyed: vi.fn(() => true),
@@ -307,7 +313,9 @@ describe("safeSendToRenderer", () => {

  describe("warning pruning logic - 100-entry hard cap", () => {
    it("enforces 100-entry cap by removing oldest entries when exceeded", async () => {
-      const consoleWarnSpy = vi.spyOn(console, "warn").mockImplementation(() => {});
+      const consoleWarnSpy = vi.spyOn(console, "warn").mockImplementation(() => {
+        /* intentionally empty - suppress console output during tests */
+      });

      mockWindow = {
        isDestroyed: vi.fn(() => true),
@@ -340,7 +348,9 @@ describe("safeSendToRenderer", () => {
    });

    it("handles many unique channels without throwing errors", async () => {
-      const consoleWarnSpy = vi.spyOn(console, "warn").mockImplementation(() => {});
+      const consoleWarnSpy = vi.spyOn(console, "warn").mockImplementation(() => {
+        /* intentionally empty - suppress console output during tests */
+      });

      mockWindow = {
        isDestroyed: vi.fn(() => true),
@@ -3,6 +3,7 @@ import { parsePhaseEvent } from './phase-event-parser';
 import {
  wouldPhaseRegress,
  isTerminalPhase,
+  isPausePhase,
  isValidExecutionPhase,
  type ExecutionPhase
 } from '../../shared/constants/phase-protocol';
@@ -13,18 +14,48 @@ export class AgentEvents {
    log: string,
    currentPhase: ExecutionProgressData['phase'],
    isSpecRunner: boolean
-  ): { phase: ExecutionProgressData['phase']; message?: string; currentSubtask?: string } | null {
+  ): {
+    phase: ExecutionProgressData['phase'];
+    message?: string;
+    currentSubtask?: string;
+    resetTimestamp?: number;
+    profileId?: string;
+  } | null {
    const structuredEvent = parsePhaseEvent(log);
    if (structuredEvent) {
-      return {
-        phase: structuredEvent.phase as ExecutionProgressData['phase'],
+      // structuredEvent.phase is validated as BackendPhase (via Zod schema),
+      // which is a subset of ExecutionPhase, so this assertion is safe
+      const result: {
+        phase: ExecutionProgressData['phase'];
+        message?: string;
+        currentSubtask?: string;
+        resetTimestamp?: number;
+        profileId?: string;
+      } = {
+        phase: structuredEvent.phase as ExecutionPhase,
        message: structuredEvent.message,
        currentSubtask: structuredEvent.subtask
      };
+
+      // Include pause phase metadata if present
+      if (structuredEvent.reset_timestamp !== undefined) {
+        result.resetTimestamp = structuredEvent.reset_timestamp;
+      }
+      if (structuredEvent.profile_id !== undefined) {
+        result.profileId = structuredEvent.profile_id;
+      }
+
+      return result;
    }

    // Terminal states can't be changed by fallback matching
-    if (isTerminalPhase(currentPhase as ExecutionPhase)) {
+    if (isTerminalPhase(currentPhase)) {
+      return null;
+    }
+
+    // Pause phases should only be changed by structured events
+    // Don't allow fallback text matching to transition out of pause phases
+    if (isPausePhase(currentPhase)) {
      return null;
    }

@@ -43,6 +74,7 @@ export class AgentEvents {
    const lowerLog = log.toLowerCase();

    // Spec runner phase detection (all part of "planning")
+    // IMPORTANT: Spec runner should NEVER transition to coding/qa phases via fallback matching
    if (isSpecRunner) {
      if (lowerLog.includes('discovering') || lowerLog.includes('discovery')) {
        return { phase: 'planning', message: 'Discovering project context...' };
@@ -59,6 +91,8 @@ export class AgentEvents {
      if (lowerLog.includes('spec complete') || lowerLog.includes('specification complete')) {
        return { phase: 'planning', message: 'Specification complete' };
      }
+      // Spec runner: don't fall through to run.py patterns (would incorrectly detect coding phase)
+      return null;
    }

    // Run.py phase detection
@@ -6,6 +6,8 @@ import { AgentEvents } from './agent-events';
 import { AgentProcessManager } from './agent-process';
 import { AgentQueueManager } from './agent-queue';
 import { getClaudeProfileManager, initializeClaudeProfileManager } from '../claude-profile-manager';
+import type { ClaudeProfileManager } from '../claude-profile-manager';
+import { getOperationRegistry } from '../claude-profile/operation-registry';
 import {
  SpecCreationMetadata,
  TaskExecutionOptions,
@@ -32,6 +34,9 @@ export class AgentManager extends EventEmitter {
    metadata?: SpecCreationMetadata;
    baseBranch?: string;
    swapCount: number;
+    projectId?: string;
+    /** Generation counter to prevent stale cleanup after restart */
+    generation: number;
  }> = new Map();

  constructor() {
@@ -51,26 +56,40 @@ export class AgentManager extends EventEmitter {
    });

    // Listen for task completion to clean up context (prevent memory leak)
-    this.on('exit', (taskId: string, code: number | null) => {
+    this.on('exit', (taskId: string, code: number | null, _processType?: string, _projectId?: string) => {
      // Clean up context when:
      // 1. Task completed successfully (code === 0), or
      // 2. Task failed and won't be restarted (handled by auto-swap logic)

+      // Capture generation at exit time to prevent race conditions with restarts
+      const contextAtExit = this.taskExecutionContext.get(taskId);
+      const generationAtExit = contextAtExit?.generation;
+
      // Note: Auto-swap restart happens BEFORE this exit event is processed,
      // so we need a small delay to allow restart to preserve context
      setTimeout(() => {
        const context = this.taskExecutionContext.get(taskId);
        if (!context) return; // Already cleaned up or restarted

+        // Check if the context's generation matches - if not, a restart incremented it
+        // and this cleanup is for a stale exit event that shouldn't affect the new task
+        if (generationAtExit !== undefined && context.generation !== generationAtExit) {
+          return; // Stale exit event - task was restarted, don't clean up new context
+        }
+
        // If task completed successfully, always clean up
        if (code === 0) {
          this.taskExecutionContext.delete(taskId);
+          // Unregister from OperationRegistry
+          getOperationRegistry().unregisterOperation(taskId);
          return;
        }

        // If task failed and hit max retries, clean up
        if (context.swapCount >= 2) {
          this.taskExecutionContext.delete(taskId);
+          // Unregister from OperationRegistry
+          getOperationRegistry().unregisterOperation(taskId);
        }
        // Otherwise keep context for potential restart
      }, 1000); // Delay to allow restart logic to run first
@@ -84,6 +103,46 @@ export class AgentManager extends EventEmitter {
    this.processManager.configure(pythonPath, autoBuildSourcePath);
  }

+  /**
+   * Register a task with the unified OperationRegistry for proactive swap support.
+   * Extracted helper to avoid code duplication between spec creation and task execution.
+   * @private
+   */
+  private registerTaskWithOperationRegistry(
+    taskId: string,
+    operationType: 'spec-creation' | 'task-execution',
+    metadata: Record<string, unknown>
+  ): void {
+    const profileManager = getClaudeProfileManager();
+    const activeProfile = profileManager.getActiveProfile();
+    if (!activeProfile) {
+      return;
+    }
+
+    // Keep internal state tracking for backward compatibility
+    this.assignProfileToTask(taskId, activeProfile.id, activeProfile.name, 'proactive');
+
+    // Register with unified registry for proactive swap
+    // Note: We don't provide a stopFn because restartTask() already handles stopping
+    // the task internally via killTask() before restarting. Providing a separate
+    // stopFn would cause a redundant double-kill during profile swaps.
+    const operationRegistry = getOperationRegistry();
+    operationRegistry.registerOperation(
+      taskId,
+      operationType,
+      activeProfile.id,
+      activeProfile.name,
+      (newProfileId: string) => this.restartTask(taskId, newProfileId),
+      { metadata }
+    );
+    console.log('[AgentManager] Task registered with OperationRegistry:', {
+      taskId,
+      profileId: activeProfile.id,
+      profileName: activeProfile.name,
+      type: operationType
+    });
+  }
+
  /**
   * Start spec creation process
   */
@@ -93,11 +152,12 @@ export class AgentManager extends EventEmitter {
    taskDescription: string,
    specDir?: string,
    metadata?: SpecCreationMetadata,
-    baseBranch?: string
+    baseBranch?: string,
+    projectId?: string
  ): Promise<void> {
    // Pre-flight auth check: Verify active profile has valid authentication
    // Ensure profile manager is initialized to prevent race condition
-    let profileManager;
+    let profileManager: ClaudeProfileManager;
    try {
      profileManager = await initializeClaudeProfileManager();
    } catch (error) {
@@ -173,10 +233,13 @@ export class AgentManager extends EventEmitter {
    }

    // Store context for potential restart
-    this.storeTaskContext(taskId, projectPath, '', {}, true, taskDescription, specDir, metadata, baseBranch);
+    this.storeTaskContext(taskId, projectPath, '', {}, true, taskDescription, specDir, metadata, baseBranch, projectId);
+
+    // Register with unified OperationRegistry for proactive swap support
+    this.registerTaskWithOperationRegistry(taskId, 'spec-creation', { projectPath, taskDescription, specDir });

    // Note: This is spec-creation but it chains to task-execution via run.py
-    await this.processManager.spawnProcess(taskId, autoBuildSource, args, combinedEnv, 'task-execution');
+    await this.processManager.spawnProcess(taskId, autoBuildSource, args, combinedEnv, 'task-execution', projectId);
  }

  /**
@@ -186,11 +249,12 @@ export class AgentManager extends EventEmitter {
    taskId: string,
    projectPath: string,
    specId: string,
-    options: TaskExecutionOptions = {}
+    options: TaskExecutionOptions = {},
+    projectId?: string
  ): Promise<void> {
    // Pre-flight auth check: Verify active profile has valid authentication
    // Ensure profile manager is initialized to prevent race condition
-    let profileManager;
+    let profileManager: ClaudeProfileManager;
    try {
      profileManager = await initializeClaudeProfileManager();
    } catch (error) {
@@ -251,9 +315,12 @@ export class AgentManager extends EventEmitter {
    // which allows per-phase configuration for planner, coder, and QA phases

    // Store context for potential restart
-    this.storeTaskContext(taskId, projectPath, specId, options, false);
+    this.storeTaskContext(taskId, projectPath, specId, options, false, undefined, undefined, undefined, undefined, projectId);

-    await this.processManager.spawnProcess(taskId, autoBuildSource, args, combinedEnv, 'task-execution');
+    // Register with unified OperationRegistry for proactive swap support
+    this.registerTaskWithOperationRegistry(taskId, 'task-execution', { projectPath, specId, options });
+
+    await this.processManager.spawnProcess(taskId, autoBuildSource, args, combinedEnv, 'task-execution', projectId);
  }

  /**
@@ -262,7 +329,8 @@ export class AgentManager extends EventEmitter {
  async startQAProcess(
    taskId: string,
    projectPath: string,
-    specId: string
+    specId: string,
+    projectId?: string
  ): Promise<void> {
    // Ensure Python environment is ready before spawning process (prevents exit code 127 race condition)
    const pythonStatus = await this.processManager.ensurePythonEnvReady('AgentManager');
@@ -290,7 +358,7 @@ export class AgentManager extends EventEmitter {

    const args = [runPath, '--spec', specId, '--project-dir', projectPath, '--qa'];

-    await this.processManager.spawnProcess(taskId, autoBuildSource, args, combinedEnv, 'qa-process');
+    await this.processManager.spawnProcess(taskId, autoBuildSource, args, combinedEnv, 'qa-process', projectId);
  }

  /**
@@ -387,11 +455,14 @@ export class AgentManager extends EventEmitter {
    taskDescription?: string,
    specDir?: string,
    metadata?: SpecCreationMetadata,
-    baseBranch?: string
+    baseBranch?: string,
+    projectId?: string
  ): void {
    // Preserve swapCount if context already exists (for restarts)
    const existingContext = this.taskExecutionContext.get(taskId);
    const swapCount = existingContext?.swapCount ?? 0;
+    // Increment generation on each store (restarts) to invalidate pending cleanup callbacks
+    const generation = (existingContext?.generation ?? 0) + 1;

    this.taskExecutionContext.set(taskId, {
      projectPath,
@@ -402,7 +473,9 @@ export class AgentManager extends EventEmitter {
      specDir,
      metadata,
      baseBranch,
-      swapCount // Preserve existing count instead of resetting
+      swapCount, // Preserve existing count instead of resetting
+      projectId,
+      generation, // Incremented to prevent stale exit cleanup
    });
  }

@@ -458,13 +531,18 @@ export class AgentManager extends EventEmitter {
      console.log('[AgentManager] Restarting task now:', taskId);
      if (context.isSpecCreation) {
        console.log('[AgentManager] Restarting as spec creation');
+        if (!context.taskDescription) {
+          console.error('[AgentManager] Cannot restart spec creation: taskDescription is missing');
+          return;
+        }
        this.startSpecCreation(
          taskId,
          context.projectPath,
-          context.taskDescription!,
+          context.taskDescription,
          context.specDir,
          context.metadata,
-          context.baseBranch
+          context.baseBranch,
+          context.projectId
        );
      } else {
        console.log('[AgentManager] Restarting as task execution');
@@ -472,7 +550,8 @@ export class AgentManager extends EventEmitter {
          taskId,
          context.projectPath,
          context.specId,
-          context.options
+          context.options,
+          context.projectId
        );
      }
    }, 500);
@@ -13,7 +13,7 @@ function createMockProcess() {
  return {
    stdout: { on: vi.fn() },
    stderr: { on: vi.fn() },
-    on: vi.fn((event: string, callback: any) => {
+    on: vi.fn((event: string, callback: (code: number) => void) => {
      if (event === 'exit') {
        // Simulate immediate exit with code 0
        setTimeout(() => callback(0), 10);
@@ -12,6 +12,7 @@ import { AgentState } from './agent-state';
 import { AgentEvents } from './agent-events';
 import { ProcessType, ExecutionProgressData } from './types';
 import type { CompletablePhase } from '../../shared/constants/phase-protocol';
+import { parseTaskEvent } from './task-event-parser';
 import { detectRateLimit, createSDKRateLimitInfo, getBestAvailableProfileEnv, detectAuthFailure } from '../rate-limit-detector';
 import { getAPIProfileEnv } from '../services/profile';
 import { projectStore } from '../project-store';
@@ -29,7 +30,7 @@ import { killProcessGracefully, isWindows } from '../platform';
 /**
 * Type for supported CLI tools
 */
-type CliTool = 'claude' | 'gh';
+type CliTool = 'claude' | 'gh' | 'glab';

 /**
 * Mapping of CLI tools to their environment variable names
@@ -37,7 +38,8 @@ type CliTool = 'claude' | 'gh';
 */
 const CLI_TOOL_ENV_MAP: Readonly<Record<CliTool, string>> = {
  claude: 'CLAUDE_CLI_PATH',
-  gh: 'GITHUB_CLI_PATH'
+  gh: 'GITHUB_CLI_PATH',
+  glab: 'GITLAB_CLI_PATH'
 } as const;


@@ -201,12 +203,14 @@ export class AgentProcessManager {
    // Detect and pass CLI tool paths to Python backend
    const claudeCliEnv = this.detectAndSetCliPath('claude');
    const ghCliEnv = this.detectAndSetCliPath('gh');
+    const glabCliEnv = this.detectAndSetCliPath('glab');

    return {
      ...augmentedEnv,
      ...gitBashEnv,
      ...claudeCliEnv,
      ...ghCliEnv,
+      ...glabCliEnv,
      ...extraEnv,
      ...profileEnv,
      PYTHONUNBUFFERED: '1',
@@ -277,7 +281,11 @@ export class AgentProcessManager {
    } : 'NONE');

    if (!bestProfile) {
-      console.log('[AgentProcess] No alternative profile available - falling back to manual modal');
+      // Single account case: let backend handle with intelligent pause
+      // Don't show manual modal - backend will pause intelligently and resume when ready
+      console.log('[AgentProcess] No alternative profile - backend will handle with intelligent pause');
+      // Return false to let handleProcessFailure emit sdk-rate-limit event
+      // The frontend can then show appropriate UI (e.g., "Paused until X time")
      return false;
    }

@@ -302,19 +310,84 @@ export class AgentProcessManager {
    console.log('[AgentProcess] No rate limit detected - checking for auth failure');
    const authFailureDetection = detectAuthFailure(allOutput);

-    if (authFailureDetection.isAuthFailure) {
-      console.log('[AgentProcess] Auth failure detected:', authFailureDetection);
+    if (!authFailureDetection.isAuthFailure) {
+      console.log('[AgentProcess] Process failed but no rate limit or auth failure detected');
+      return false;
+    }
+
+    console.log('[AgentProcess] Auth failure detected:', authFailureDetection);
+
+    // Try auto-swap if enabled
+    const wasHandled = this.handleAuthFailureWithAutoSwap(taskId, authFailureDetection);
+
+    if (!wasHandled) {
+      // Fall back to UI notification
      this.emitter.emit('auth-failure', taskId, {
        profileId: authFailureDetection.profileId,
        failureType: authFailureDetection.failureType,
        message: authFailureDetection.message,
        originalError: authFailureDetection.originalError
      });
-      return true;
    }

-    console.log('[AgentProcess] Process failed but no rate limit or auth failure detected');
-    return false;
+    return true;
+  }
+
+  /**
+   * Attempt to auto-swap to another profile on authentication failure.
+   * Only works when autoSwitchOnAuthFailure is enabled and an alternative
+   * authenticated profile is available.
+   */
+  private handleAuthFailureWithAutoSwap(
+    taskId: string,
+    authFailureDetection: ReturnType<typeof detectAuthFailure>
+  ): boolean {
+    const profileManager = getClaudeProfileManager();
+    const autoSwitchSettings = profileManager.getAutoSwitchSettings();
+
+    console.log('[AgentProcess] Auth failure auto-switch settings:', {
+      enabled: autoSwitchSettings.enabled,
+      autoSwitchOnAuthFailure: autoSwitchSettings.autoSwitchOnAuthFailure
+    });
+
+    // Check if auto-switch on auth failure is enabled
+    if (!autoSwitchSettings.enabled || !autoSwitchSettings.autoSwitchOnAuthFailure) {
+      console.log('[AgentProcess] Auth failure auto-switch disabled - falling back to UI');
+      return false;
+    }
+
+    const currentProfileId = authFailureDetection.profileId;
+    const bestProfile = profileManager.getBestAvailableProfile(currentProfileId);
+
+    console.log('[AgentProcess] Best available profile for auth failure swap:', bestProfile ? {
+      id: bestProfile.id,
+      name: bestProfile.name,
+      isAuthenticated: bestProfile.isAuthenticated
+    } : 'NONE');
+
+    // Verify the best profile is actually authenticated
+    if (!bestProfile || !bestProfile.isAuthenticated) {
+      console.log('[AgentProcess] No authenticated alternative profile - falling back to UI');
+      return false;
+    }
+
+    console.log('[AgentProcess] AUTH-FAILURE AUTO-SWAP:', currentProfileId, '->', bestProfile.id);
+    profileManager.setActiveProfile(bestProfile.id);
+
+    // Emit auth-failure event with swap metadata for UI notification
+    this.emitter.emit('auth-failure', taskId, {
+      profileId: authFailureDetection.profileId,
+      failureType: authFailureDetection.failureType,
+      message: authFailureDetection.message,
+      originalError: authFailureDetection.originalError,
+      wasAutoSwapped: true,
+      swappedToProfile: { id: bestProfile.id, name: bestProfile.name }
+    });
+
+    // Reuse existing restart event
+    console.log('[AgentProcess] Emitting auto-swap-restart-task event for auth failure:', taskId);
+    this.emitter.emit('auto-swap-restart-task', taskId, bestProfile.id);
+    return true;
  }

  /**
@@ -506,7 +579,8 @@ export class AgentProcessManager {
    cwd: string,
    args: string[],
    extraEnv: Record<string, string> = {},
-    processType: ProcessType = 'task-execution'
+    processType: ProcessType = 'task-execution',
+    projectId?: string
  ): Promise<void> {
    const isSpecRunner = processType === 'spec-creation';
    this.killProcess(taskId);
@@ -558,7 +632,7 @@ export class AgentProcessManager {
      // spawn() failed synchronously (e.g., command not found, permission denied)
      // Clean up tracking entry and propagate error
      this.state.deleteProcess(taskId);
-      this.emitter.emit('error', taskId, err instanceof Error ? err.message : String(err));
+      this.emitter.emit('error', taskId, err instanceof Error ? err.message : String(err), projectId);
      throw err;
    }

@@ -605,7 +679,7 @@ export class AgentProcessManager {
      message: isSpecRunner ? 'Starting spec creation...' : 'Starting build process...',
      sequenceNumber: ++sequenceNumber,
      completedPhases: [...completedPhases]
-    });
+    }, projectId);

    const isDebug = ['true', '1', 'yes', 'on'].includes(process.env.DEBUG?.toLowerCase() ?? '');

@@ -617,6 +691,17 @@ export class AgentProcessManager {
        console.log(`[PhaseDebug:${taskId}] Found marker in line: "${line.substring(0, 200)}"`);
      }

+      // Log all task event markers for debugging
+      if (line.includes('__TASK_EVENT__')) {
+        console.log(`[AgentProcess:${taskId}] Found __TASK_EVENT__ marker in line:`, line.substring(0, 300));
+      }
+
+      const taskEvent = parseTaskEvent(line);
+      if (taskEvent) {
+        console.log(`[AgentProcess:${taskId}] Parsed task event:`, taskEvent.type, taskEvent);
+        this.emitter.emit('task-event', taskId, taskEvent, projectId);
+      }
+
      const phaseUpdate = this.events.parseExecutionPhase(line, currentPhase, isSpecRunner);

      if (isDebug && hasMarker) {
@@ -674,7 +759,7 @@ export class AgentProcessManager {
          message: lastMessage,
          sequenceNumber: ++sequenceNumber,
          completedPhases: [...completedPhases]
-        });
+        }, projectId);
      }
    };

@@ -694,7 +779,7 @@ export class AgentProcessManager {

      for (const line of lines) {
        if (line.trim()) {
-          this.emitter.emit('log', taskId, line + '\n');
+          this.emitter.emit('log', taskId, line + '\n', projectId);
          processLog(line);
          if (isDebug) {
            console.log(`[Agent:${taskId}] ${line}`);
@@ -715,11 +800,11 @@ export class AgentProcessManager {

    childProcess.on('exit', (code: number | null) => {
      if (stdoutBuffer.trim()) {
-        this.emitter.emit('log', taskId, stdoutBuffer + '\n');
+        this.emitter.emit('log', taskId, stdoutBuffer + '\n', projectId);
        processLog(stdoutBuffer);
      }
      if (stderrBuffer.trim()) {
-        this.emitter.emit('log', taskId, stderrBuffer + '\n');
+        this.emitter.emit('log', taskId, stderrBuffer + '\n', projectId);
        processLog(stderrBuffer);
      }

@@ -734,7 +819,7 @@ export class AgentProcessManager {
        console.log('[AgentProcess] Process failed with code:', code, 'for task:', taskId);
        const wasHandled = this.handleProcessFailure(taskId, allOutput, processType);
        if (wasHandled) {
-          this.emitter.emit('exit', taskId, code, processType);
+          this.emitter.emit('exit', taskId, code, processType, projectId);
          return;
        }
      }
@@ -747,10 +832,10 @@ export class AgentProcessManager {
          message: `Process exited with code ${code}`,
          sequenceNumber: ++sequenceNumber,
          completedPhases: [...completedPhases]
-        });
+        }, projectId);
      }

-      this.emitter.emit('exit', taskId, code, processType);
+      this.emitter.emit('exit', taskId, code, processType, projectId);
    });

    // Handle process error
@@ -765,9 +850,9 @@ export class AgentProcessManager {
        message: `Error: ${err.message}`,
        sequenceNumber: ++sequenceNumber,
        completedPhases: [...completedPhases]
-      });
+      }, projectId);

-      this.emitter.emit('error', taskId, err.message);
+      this.emitter.emit('error', taskId, err.message, projectId);
    });
  }

@@ -417,7 +417,12 @@ export class AgentQueueManager {

    // Track completed types for progress calculation
    const completedTypes = new Set<string>();
-    const totalTypes = 7; // Default all types
+    // Derive totalTypes from --types argument instead of hardcoding
+    const typesArgIndex = args.findIndex((arg) => arg === '--types');
+    const totalTypes =
+      typesArgIndex > -1 && args[typesArgIndex + 1]
+        ? args[typesArgIndex + 1].split(',').length
+        : 6; // Default to 6 if not specified

    // Handle stdout - explicitly decode as UTF-8 for cross-platform Unicode support
    childProcess.stdout?.on('data', (data: Buffer) => {
@@ -24,9 +24,24 @@ describe('AgentState - Queue Routing', () => {

    it('should group tasks by profile', () => {
      // Add mock processes
-      state.addProcess('task-1', { pid: 1001 } as any);
-      state.addProcess('task-2', { pid: 1002 } as any);
-      state.addProcess('task-3', { pid: 1003 } as any);
+      state.addProcess('task-1', {
+        taskId: 'task-1',
+        process: { pid: 1001 } as unknown as import('child_process').ChildProcess,
+        startedAt: new Date(),
+        spawnId: 1
+      });
+      state.addProcess('task-2', {
+        taskId: 'task-2',
+        process: { pid: 1002 } as unknown as import('child_process').ChildProcess,
+        startedAt: new Date(),
+        spawnId: 2
+      });
+      state.addProcess('task-3', {
+        taskId: 'task-3',
+        process: { pid: 1003 } as unknown as import('child_process').ChildProcess,
+        startedAt: new Date(),
+        spawnId: 3
+      });

      // Assign profiles
      state.assignProfileToTask('task-1', 'profile-1', 'Profile 1', 'proactive');
@@ -42,7 +57,12 @@ describe('AgentState - Queue Routing', () => {

    it('should use default profile for unassigned tasks', () => {
      // Add process without profile assignment
-      state.addProcess('task-1', { pid: 1001 } as any);
+      state.addProcess('task-1', {
+        taskId: 'task-1',
+        process: { pid: 1001 } as unknown as import('child_process').ChildProcess,
+        startedAt: new Date(),
+        spawnId: 1
+      });

      const result = state.getRunningTasksByProfile();

@@ -14,6 +14,9 @@ export interface PhaseParseResult<TPhase extends string = string> {
  message?: string;
  currentSubtask?: string;
  progress?: number;
+  // Pause phase metadata
+  resetTimestamp?: number;  // Unix timestamp for rate limit reset
+  profileId?: string;  // Profile that hit the limit
 }

 /**
@@ -9,6 +9,7 @@ import { BasePhaseParser, type PhaseParseResult, type PhaseParserContext } from
 import {
  EXECUTION_PHASES,
  TERMINAL_PHASES,
+  isPausePhase,
  type ExecutionPhase
 } from '../../../shared/constants/phase-protocol';
 import { parsePhaseEvent } from '../phase-event-parser';
@@ -40,11 +41,21 @@ export class ExecutionPhaseParser extends BasePhaseParser<ExecutionPhase> {
    // 1. Try structured event first (authoritative source)
    const structuredEvent = parsePhaseEvent(log);
    if (structuredEvent) {
-      return {
+      const result: PhaseParseResult<ExecutionPhase> = {
        phase: structuredEvent.phase as ExecutionPhase,
        message: structuredEvent.message,
        currentSubtask: structuredEvent.subtask
      };
+
+      // Include pause phase metadata if present
+      if (structuredEvent.reset_timestamp !== undefined) {
+        result.resetTimestamp = structuredEvent.reset_timestamp;
+      }
+      if (structuredEvent.profile_id !== undefined) {
+        result.profileId = structuredEvent.profile_id;
+      }
+
+      return result;
    }

    // 2. Terminal states can't be changed by fallback matching
@@ -52,7 +63,13 @@ export class ExecutionPhaseParser extends BasePhaseParser<ExecutionPhase> {
      return null;
    }

-    // 3. Fall back to text pattern matching
+    // 3. Pause phases should only be changed by structured events
+    // Don't allow fallback text matching to transition out of pause phases
+    if (isPausePhase(context.currentPhase)) {
+      return null;
+    }
+
+    // 4. Fall back to text pattern matching
    return this.parseFallbackPatterns(log, context);
  }

@@ -7,7 +7,10 @@ export const PhaseEventSchema = z.object({
  phase: BackendPhaseSchema,
  message: z.string().default(''),
  progress: z.number().int().min(0).max(100).optional(),
-  subtask: z.string().optional()
+  subtask: z.string().optional(),
+  // Pause phase metadata
+  reset_timestamp: z.number().int().optional(),  // Unix timestamp for rate limit reset
+  profile_id: z.string().optional()  // Profile that hit the limit
 });

 export type PhaseEventPayload = z.infer<typeof PhaseEventSchema>;
@@ -0,0 +1,120 @@
+/**
+ * Structured task event parser for Python -> TypeScript protocol.
+ * Protocol: __TASK_EVENT__:{...}
+ */
+
+import { validateTaskEvent, type TaskEventPayload } from './task-event-schema';
+
+export const TASK_EVENT_PREFIX = '__TASK_EVENT__:';
+
+const DEBUG = process.env.DEBUG?.toLowerCase() === 'true' || process.env.DEBUG === '1';
+
+export type TaskEvent = TaskEventPayload;
+
+export function parseTaskEvent(line: string): TaskEventPayload | null {
+  const markerIndex = line.indexOf(TASK_EVENT_PREFIX);
+  if (markerIndex === -1) {
+    return null;
+  }
+
+  if (DEBUG) {
+    console.log('[task-event-parser] Found marker at index', markerIndex, 'in line:', line.substring(0, 200));
+  }
+
+  const rawJsonStr = line.slice(markerIndex + TASK_EVENT_PREFIX.length).trim();
+  if (!rawJsonStr) {
+    if (DEBUG) {
+      console.log('[task-event-parser] Empty JSON string after marker');
+    }
+    return null;
+  }
+
+  const jsonStr = extractJsonObject(rawJsonStr);
+  if (!jsonStr) {
+    if (DEBUG) {
+      console.log('[task-event-parser] Could not extract JSON object from:', rawJsonStr.substring(0, 200));
+    }
+    return null;
+  }
+
+  if (DEBUG) {
+    console.log('[task-event-parser] Attempting to parse JSON:', jsonStr.substring(0, 200));
+  }
+
+  try {
+    const rawPayload = JSON.parse(jsonStr) as unknown;
+    const result = validateTaskEvent(rawPayload);
+
+    if (!result.success) {
+      if (DEBUG) {
+        console.log('[task-event-parser] Validation failed:', result.error.format());
+      }
+      return null;
+    }
+
+    if (DEBUG) {
+      console.log('[task-event-parser] Successfully parsed event:', result.data);
+    }
+
+    return result.data;
+  } catch (e) {
+    if (DEBUG) {
+      console.log('[task-event-parser] JSON parse FAILED for:', jsonStr);
+      console.log('[task-event-parser] Error:', e);
+    }
+    return null;
+  }
+}
+
+export function hasTaskMarker(line: string): boolean {
+  return line.includes(TASK_EVENT_PREFIX);
+}
+
+/**
+ * Extract a JSON object from a string that may have trailing garbage.
+ * Finds the matching closing brace for the first opening brace.
+ */
+function extractJsonObject(str: string): string | null {
+  const firstBrace = str.indexOf('{');
+  if (firstBrace === -1) {
+    return null;
+  }
+
+  let depth = 0;
+  let inString = false;
+  let isEscaped = false;
+
+  for (let i = firstBrace; i < str.length; i++) {
+    const char = str[i];
+
+    if (isEscaped) {
+      isEscaped = false;
+      continue;
+    }
+
+    if (char === '\\' && inString) {
+      isEscaped = true;
+      continue;
+    }
+
+    if (char === '"') {
+      inString = !inString;
+      continue;
+    }
+
+    if (inString) {
+      continue;
+    }
+
+    if (char === '{') {
+      depth++;
+    } else if (char === '}') {
+      depth--;
+      if (depth === 0) {
+        return str.slice(firstBrace, i + 1);
+      }
+    }
+  }
+
+  return null;
+}
@@ -0,0 +1,33 @@
+import { z } from 'zod';
+
+export const TaskEventSchema = z.object({
+  type: z.string(),
+  taskId: z.string(),
+  specId: z.string(),
+  projectId: z.string(),
+  timestamp: z.string(),
+  eventId: z.string(),
+  sequence: z.number().int().min(0)
+}).passthrough();
+
+export type TaskEventPayload = z.infer<typeof TaskEventSchema>;
+
+export interface ValidationResult {
+  success: true;
+  data: TaskEventPayload;
+}
+
+export interface ValidationError {
+  success: false;
+  error: z.ZodError;
+}
+
+export type ParseResult = ValidationResult | ValidationError;
+
+export function validateTaskEvent(data: unknown): ParseResult {
+  const result = TaskEventSchema.safeParse(data);
+  if (result.success) {
+    return { success: true, data: result.data as TaskEventPayload };
+  }
+  return { success: false, error: result.error };
+}
@@ -1,6 +1,6 @@
 import { ChildProcess } from 'child_process';
-import type { IdeationConfig } from '../../shared/types';
-import type { CompletablePhase } from '../../shared/constants/phase-protocol';
+import type { CompletablePhase, ExecutionPhase } from '../../shared/constants/phase-protocol';
+import type { TaskEventPayload } from './task-event-schema';

 /**
 * Agent-specific types for process and state management
@@ -18,7 +18,7 @@ export interface AgentProcess {
 }

 export interface ExecutionProgressData {
-  phase: 'idle' | 'planning' | 'coding' | 'qa_review' | 'qa_fixing' | 'complete' | 'failed';
+  phase: ExecutionPhase;
  phaseProgress: number;
  overallProgress: number;
  currentSubtask?: string;
@@ -30,10 +30,11 @@ export interface ExecutionProgressData {
 export type ProcessType = 'spec-creation' | 'task-execution' | 'qa-process';

 export interface AgentManagerEvents {
-  log: (taskId: string, log: string) => void;
-  error: (taskId: string, error: string) => void;
-  exit: (taskId: string, code: number | null, processType: ProcessType) => void;
-  'execution-progress': (taskId: string, progress: ExecutionProgressData) => void;
+  log: (taskId: string, log: string, projectId?: string) => void;
+  error: (taskId: string, error: string, projectId?: string) => void;
+  exit: (taskId: string, code: number | null, processType: ProcessType, projectId?: string) => void;
+  'execution-progress': (taskId: string, progress: ExecutionProgressData, projectId?: string) => void;
+  'task-event': (taskId: string, event: TaskEventPayload, projectId?: string) => void;
 }

 // IdeationConfig now imported from shared types to maintain consistency
@@ -48,6 +49,7 @@ export interface TaskExecutionOptions {
  workers?: number;
  baseBranch?: string;
  useWorktree?: boolean; // If false, use --direct mode (no worktree isolation)
+  useLocalBranch?: boolean; // If true, use local branch directly instead of preferring origin/branch
 }

 export interface SpecCreationMetadata {
@@ -71,6 +73,7 @@ export interface SpecCreationMetadata {
  thinkingLevel?: 'none' | 'low' | 'medium' | 'high' | 'ultrathink';
  // Workspace mode - whether to use worktree isolation
  useWorktree?: boolean; // If false, use --direct mode (no worktree isolation)
+  useLocalBranch?: boolean; // If true, use local branch directly instead of preferring origin/branch
 }

 export interface IdeationProgressData {
@@ -0,0 +1,45 @@
+/**
+ * App language tracking module for main process.
+ *
+ * Tracks the user's in-app language setting (not OS locale) for use in
+ * main process code that needs localized strings (e.g., context menus).
+ *
+ * Updated via IPC when user changes language in settings.
+ */
+
+import { app } from 'electron';
+
+// Current app language, defaults to 'en'
+// Updated via setAppLanguage() when renderer notifies of language change
+let currentAppLanguage = 'en';
+
+/**
+ * Get the current app language.
+ * Falls back to 'en' if not set.
+ */
+export function getAppLanguage(): string {
+  return currentAppLanguage;
+}
+
+/**
+ * Set the current app language.
+ * Called by IPC handler when renderer changes language.
+ */
+export function setAppLanguage(language: string): void {
+  currentAppLanguage = language;
+}
+
+/**
+ * Initialize app language from OS locale as a starting point.
+ * The renderer will update this once i18n initializes.
+ */
+export function initAppLanguage(): void {
+  try {
+    // app.getLocale() may not be available in test environments
+    const osLocale = app?.getLocale?.() || 'en';
+    // Extract base language (e.g., 'en-US' -> 'en')
+    currentAppLanguage = osLocale.split('-')[0] || 'en';
+  } catch {
+    currentAppLanguage = 'en';
+  }
+}
@@ -17,6 +17,8 @@
 * - APP_UPDATE_ERROR: Error during update process
 */

+import { accessSync, constants as fsConstants } from 'fs';
+import path from 'path';
 import { autoUpdater } from 'electron-updater';
 import type { UpdateInfo } from 'electron-updater';
 import { app, net } from 'electron';
@@ -24,6 +26,7 @@ import type { BrowserWindow } from 'electron';
 import { IPC_CHANNELS } from '../shared/constants';
 import type { AppUpdateInfo } from '../shared/types';
 import { compareVersions } from './updater/version-manager';
+import { isMacOS } from './platform';

 // GitHub repo info for API calls
 const GITHUB_OWNER = 'AndyMik90';
@@ -91,10 +94,13 @@ function formatReleaseNotes(releaseNotes: UpdateInfo['releaseNotes']): string |
 */
 export function setUpdateChannel(channel: UpdateChannel): void {
  autoUpdater.channel = channel;
+  // Enable pre-release scanning when beta channel is selected
+  // This allows electron-updater to find beta releases on GitHub
+  autoUpdater.allowPrerelease = channel === 'beta';
  // Clear any downloaded update info when channel changes to prevent showing
  // an Install button for an update from a different channel
  downloadedUpdateInfo = null;
-  console.warn(`[app-updater] Update channel set to: ${channel}`);
+  console.warn(`[app-updater] Update channel set to: ${channel}, allowPrerelease: ${autoUpdater.allowPrerelease}`);
 }

 // Enable more verbose logging in debug mode
@@ -187,8 +193,7 @@ export function initializeAppUpdater(window: BrowserWindow, betaUpdates = false)
    console.error('[app-updater] Update error:', error);
    if (mainWindow) {
      mainWindow.webContents.send(IPC_CHANNELS.APP_UPDATE_ERROR, {
-        message: error.message,
-        stack: error.stack
+        message: error.message
      });
    }
  });
@@ -293,13 +298,57 @@ export async function downloadUpdate(): Promise<void> {
  }
 }

+/**
+ * Check if the app is running from a read-only volume (e.g., DMG on macOS)
+ * Returns true if the app cannot be updated in place
+ */
+function isRunningFromReadOnlyVolume(): boolean {
+  if (!isMacOS()) {
+    return false;
+  }
+
+  const appPath = app.getAppPath();
+
+  // Check if the filesystem is read-only by testing write access.
+  // We don't use a /Volumes/ prefix check because writable external drives
+  // (USB, external SSDs) are also mounted under /Volumes/ on macOS.
+  try {
+    // Navigate from app.asar to the Contents/ directory (app.asar -> Resources -> Contents)
+    const contentsPath = path.resolve(appPath, '..', '..');
+
+    // Try to check if we can write to the app bundle's parent directory
+    accessSync(path.dirname(contentsPath), fsConstants.W_OK);
+    return false;
+  } catch (error: unknown) {
+    // Only treat as read-only if the filesystem itself is read-only (EROFS).
+    // Permission errors (EACCES) in managed/enterprise environments should not
+    // block updates — the updater may still have elevated access.
+    const code = error instanceof Error ? (error as NodeJS.ErrnoException).code : undefined;
+    return code === 'EROFS';
+  }
+}
+
 /**
 * Quit and install update
 * Called from IPC handler when user confirms installation
+ * Returns false if running from a read-only volume (update cannot proceed)
 */
-export function quitAndInstall(): void {
+export function quitAndInstall(): boolean {
+  // Check if running from read-only volume before attempting install
+  if (isRunningFromReadOnlyVolume()) {
+    console.warn('[app-updater] Cannot install: running from read-only volume');
+
+    if (mainWindow) {
+      mainWindow.webContents.send(IPC_CHANNELS.APP_UPDATE_READONLY_VOLUME, {
+        appPath: app.getAppPath()
+      });
+    }
+    return false;
+  }
+
  console.warn('[app-updater] Quitting and installing update');
  autoUpdater.quitAndInstall(false, true);
+  return true;
 }

 /**
@@ -402,7 +451,7 @@ async function fetchLatestStableRelease(): Promise<AppUpdateInfo | null> {

          const version = latestStable.tag_name.replace(/^v/, '');
          // Sanitize version string for logging (remove control characters and limit length)
-          // eslint-disable-next-line no-control-regex
+          // biome-ignore lint/suspicious/noControlCharactersInRegex: Intentionally matching control chars for sanitization
          const safeVersion = String(version).replace(/[\x00-\x1f\x7f]/g, '').slice(0, 50);
          console.warn('[app-updater] Found latest stable release:', safeVersion);

@@ -483,11 +532,8 @@ export async function setUpdateChannelWithDowngradeCheck(
  channel: UpdateChannel,
  triggerDowngradeCheck = false
 ): Promise<AppUpdateInfo | null> {
-  autoUpdater.channel = channel;
-  // Clear any downloaded update info when channel changes to prevent showing
-  // an Install button for an update from a different channel
-  downloadedUpdateInfo = null;
-  console.warn(`[app-updater] Update channel set to: ${channel}`);
+  // Use the shared channel-setting function to avoid code duplication
+  setUpdateChannel(channel);

  // If switching to stable and downgrade check requested, look for stable version
  if (channel === 'latest' && triggerDowngradeCheck) {
@@ -509,8 +555,8 @@ export async function setUpdateChannelWithDowngradeCheck(
 * Uses electron-updater with allowDowngrade enabled to download older stable versions
 */
 export async function downloadStableVersion(): Promise<void> {
-  // Switch to stable channel
-  autoUpdater.channel = 'latest';
+  // Switch to stable channel (resets allowPrerelease and clears downloadedUpdateInfo)
+  setUpdateChannel('latest');
  // Enable downgrade to allow downloading older versions (e.g., stable when on beta)
  autoUpdater.allowDowngrade = true;
  console.warn('[app-updater] Downloading stable version (allowDowngrade=true)...');
@@ -8,7 +8,7 @@ import { app } from 'electron';
 const __filename = fileURLToPath(import.meta.url);
 const __dirname = path.dirname(__filename);
 import { AUTO_BUILD_PATHS, DEFAULT_CHANGELOG_PATH } from '../../shared/constants';
-import { getToolPath } from '../cli-tool-manager';
+import { getToolPath, getToolInfo } from '../cli-tool-manager';
 import type {
  ChangelogTask,
  TaskSpecContent,
@@ -44,7 +44,6 @@ export class ChangelogService extends EventEmitter {
  private _pythonPath: string | null = null;
  private claudePath: string;
  private autoBuildSourcePath: string = '';
-  private cachedEnv: Record<string, string> | null = null;
  private debugEnabled: boolean | null = null;
  private generator: ChangelogGenerator | null = null;
  private versionSuggester: VersionSuggester | null = null;
@@ -175,26 +174,40 @@ export class ChangelogService extends EventEmitter {
    }
  }

+  /**
+   * Ensure prerequisites are met for changelog generation
+   * Validates auto-build source path and Claude CLI availability
+   * Returns the resolved Claude CLI path to ensure we use the freshly validated path
+   */
+  private ensurePrerequisites(): { autoBuildSource: string; claudePath: string } {
+    const autoBuildSource = this.getAutoBuildSourcePath();
+    if (!autoBuildSource) {
+      throw new Error('Auto-build source path not found');
+    }
+
+    const claudeInfo = getToolInfo('claude');
+    if (!claudeInfo.found || !claudeInfo.path) {
+      // Use claudeInfo.message directly to avoid redundant text
+      throw new Error(claudeInfo.message || 'Claude CLI not found. Install from https://claude.ai/download');
+    }
+
+    // Update cached path with freshly resolved value
+    this.claudePath = claudeInfo.path;
+    return { autoBuildSource, claudePath: claudeInfo.path };
+  }
+
  /**
   * Get or create the generator instance
   */
  private getGenerator(): ChangelogGenerator {
    if (!this.generator) {
-      const autoBuildSource = this.getAutoBuildSourcePath();
-      if (!autoBuildSource) {
-        throw new Error('Auto-build source path not found');
-      }
-
-      // Verify claude CLI is available
-      if (this.claudePath !== 'claude' && !existsSync(this.claudePath)) {
-        throw new Error(`Claude CLI not found. Please ensure Claude Code is installed. Looked for: ${this.claudePath}`);
-      }
+      const { autoBuildSource, claudePath } = this.ensurePrerequisites();

      const autoBuildEnv = this.loadAutoBuildEnv();

      this.generator = new ChangelogGenerator(
        this.pythonPath,
-        this.claudePath,
+        claudePath,
        autoBuildSource,
        autoBuildEnv,
        this.isDebugEnabled()
@@ -226,19 +239,11 @@ export class ChangelogService extends EventEmitter {
   */
  private getVersionSuggester(): VersionSuggester {
    if (!this.versionSuggester) {
-      const autoBuildSource = this.getAutoBuildSourcePath();
-      if (!autoBuildSource) {
-        throw new Error('Auto-build source path not found');
-      }
-
-      // Verify claude CLI is available
-      if (this.claudePath !== 'claude' && !existsSync(this.claudePath)) {
-        throw new Error(`Claude CLI not found. Please ensure Claude Code is installed. Looked for: ${this.claudePath}`);
-      }
+      const { autoBuildSource, claudePath } = this.ensurePrerequisites();

      this.versionSuggester = new VersionSuggester(
        this.pythonPath,
-        this.claudePath,
+        claudePath,
        autoBuildSource,
        this.isDebugEnabled()
      );
@@ -448,7 +453,7 @@ export class ChangelogService extends EventEmitter {
    }

    const parts = currentVersion.split('.').map(Number);
-    if (parts.length !== 3 || parts.some(isNaN)) {
+    if (parts.length !== 3 || parts.some(Number.isNaN)) {
      return '1.0.0';
    }

@@ -485,7 +490,7 @@ export class ChangelogService extends EventEmitter {
   * Suggest version using AI analysis of git commits
   */
  async suggestVersionFromCommits(
-    projectPath: string,
+    _projectPath: string,
    commits: import('../../shared/types').GitCommit[],
    currentVersion?: string
  ): Promise<{ version: string; reason: string }> {
@@ -496,7 +501,7 @@ export class ChangelogService extends EventEmitter {
      }

      const parts = currentVersion.split('.').map(Number);
-      if (parts.length !== 3 || parts.some(isNaN)) {
+      if (parts.length !== 3 || parts.some(Number.isNaN)) {
        return { version: '1.0.0', reason: 'Invalid current version, resetting to 1.0.0' };
      }

@@ -198,7 +198,6 @@ except Exception as e:
      case 'minor':
        newVersion = `${major}.${minor + 1}.0`;
        break;
-      case 'patch':
      default:
        newVersion = `${major}.${minor}.${patch + 1}`;
        break;
--- a/Show More
+++ b/Show More