Andy acdd7d9b1e refactor(github-review): replace confidence scoring with evidence-based validation (#628)
* refactor(github-review): replace confidence scoring with evidence-based validation

Removes confidence scores (0-100) from PR review system in favor of
evidence-based validation. This addresses the root cause of false
positives: reviews reporting issues they couldn't prove with actual code.

Key changes:
- Remove confidence field from PRReviewFinding and pydantic models
- Add required 'evidence' field for code snippets proving issues exist
- Deprecate confidence.py module with migration guidance
- Update response_parsers.py to filter by evidence presence, not score
- Add "NEVER ASSUME - ALWAYS VERIFY" sections to all reviewer prompts
- Update finding-validator to use binary evidence verification
- Update tests for new evidence-based validation model

The new model is simple: either you can show the problematic code, or
you don't report the finding. No more "80% confident" hedging.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* feat: enhance worktree path resolution with legacy support

Updated the find_worktree function to first check the new worktree path at .auto-claude/worktrees/tasks/ for task directories. Added a fallback to check the legacy path at .worktrees/ for backwards compatibility, ensuring that existing workflows are not disrupted.

This change improves the robustness of worktree handling in the project.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

* fix: remove validation_confidence and confidence references causing runtime errors

The evidence-based validation migration missed updating:
- parallel_followup_reviewer.py:647 - accessed non-existent validation.confidence
- parallel_followup_reviewer.py:663 - passed validation_confidence to PRReviewFinding
- parallel_orchestrator_reviewer.py:589,972 - passed confidence to PRReviewFinding

PRReviewFinding no longer has confidence field (replaced with evidence).
FindingValidationResult no longer has confidence field (replaced with
evidence_verified_in_file).

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-03 20:45:42 +01:00
2025-12-22 14:17:08 +01:00
2026-01-02 11:56:36 +01:00
2026-01-02 11:56:36 +01:00
2026-01-02 11:56:36 +01:00
2026-01-02 11:56:36 +01:00
2026-01-02 11:56:36 +01:00
2025-12-10 09:10:55 +01:00
2026-01-02 11:56:36 +01:00
2026-01-02 14:12:36 +01:00
2026-01-02 11:56:36 +01:00
2026-01-02 11:56:36 +01:00
2025-12-22 14:17:08 +01:00
2026-01-02 11:56:36 +01:00
2026-01-02 11:56:36 +01:00
2026-01-02 11:56:36 +01:00
2025-12-15 21:10:27 +01:00

Auto Claude

Autonomous multi-agent coding framework that plans, builds, and validates software for you.

Auto Claude Kanban Board

Version

License Discord CI


Download

Stable Release

Stable

Platform Download
Windows Auto-Claude-2.7.2-win32-x64.exe
macOS (Apple Silicon) Auto-Claude-2.7.2-darwin-arm64.dmg
macOS (Intel) Auto-Claude-2.7.2-darwin-x64.dmg
Linux Auto-Claude-2.7.2-linux-x86_64.AppImage
Linux (Debian) Auto-Claude-2.7.2-linux-amd64.deb

Beta Release

⚠️ Beta releases may contain bugs and breaking changes. View all releases

Beta

Platform Download
Windows Auto-Claude-2.7.2-beta.10-win32-x64.exe
macOS (Apple Silicon) Auto-Claude-2.7.2-beta.10-darwin-arm64.dmg
macOS (Intel) Auto-Claude-2.7.2-beta.10-darwin-x64.dmg
Linux Auto-Claude-2.7.2-beta.10-linux-x86_64.AppImage
Linux (Debian) Auto-Claude-2.7.2-beta.10-linux-amd64.deb
Linux (Flatpak) Auto-Claude-2.7.2-beta.10-linux-x86_64.flatpak

All releases include SHA256 checksums and VirusTotal scan results for security verification.


Requirements

  • Claude Pro/Max subscription - Get one here
  • Claude Code CLI - npm install -g @anthropic-ai/claude-code
  • Git repository - Your project must be initialized as a git repo
  • Python 3.12+ - Required for the backend and Memory Layer

Quick Start

  1. Download and install the app for your platform
  2. Open your project - Select a git repository folder
  3. Connect Claude - The app will guide you through OAuth setup
  4. Create a task - Describe what you want to build
  5. Watch it work - Agents plan, code, and validate autonomously

Features

Feature Description
Autonomous Tasks Describe your goal; agents handle planning, implementation, and validation
Parallel Execution Run multiple builds simultaneously with up to 12 agent terminals
Isolated Workspaces All changes happen in git worktrees - your main branch stays safe
Self-Validating QA Built-in quality assurance loop catches issues before you review
AI-Powered Merge Automatic conflict resolution when integrating back to main
Memory Layer Agents retain insights across sessions for smarter builds
GitHub/GitLab Integration Import issues, investigate with AI, create merge requests
Linear Integration Sync tasks with Linear for team progress tracking
Cross-Platform Native desktop apps for Windows, macOS, and Linux
Auto-Updates App updates automatically when new versions are released

Interface

Kanban Board

Visual task management from planning through completion. Create tasks and monitor agent progress in real-time.

Agent Terminals

AI-powered terminals with one-click task context injection. Spawn multiple agents for parallel work.

Agent Terminals

Roadmap

AI-assisted feature planning with competitor analysis and audience targeting.

Roadmap

Additional Features

  • Insights - Chat interface for exploring your codebase
  • Ideation - Discover improvements, performance issues, and vulnerabilities
  • Changelog - Generate release notes from completed tasks

Project Structure

Auto-Claude/
├── apps/
│   ├── backend/     # Python agents, specs, QA pipeline
│   └── frontend/    # Electron desktop application
├── guides/          # Additional documentation
├── tests/           # Test suite
└── scripts/         # Build utilities

CLI Usage

For headless operation, CI/CD integration, or terminal-only workflows:

cd apps/backend

# Create a spec interactively
python spec_runner.py --interactive

# Run autonomous build
python run.py --spec 001

# Review and merge
python run.py --spec 001 --review
python run.py --spec 001 --merge

See guides/CLI-USAGE.md for complete CLI documentation.


Configuration

Create apps/backend/.env from the example:

cp apps/backend/.env.example apps/backend/.env
Variable Required Description
CLAUDE_CODE_OAUTH_TOKEN Yes OAuth token from claude setup-token
GRAPHITI_ENABLED No Enable Memory Layer for cross-session context
AUTO_BUILD_MODEL No Override the default Claude model
GITLAB_TOKEN No GitLab Personal Access Token for GitLab integration
GITLAB_INSTANCE_URL No GitLab instance URL (defaults to gitlab.com)
LINEAR_API_KEY No Linear API key for task sync

Building from Source

For contributors and development:

# Clone the repository
git clone https://github.com/AndyMik90/Auto-Claude.git
cd Auto-Claude

# Install all dependencies
npm run install:all

# Run in development mode
npm run dev

# Or build and run
npm start

System requirements for building:

  • Node.js 24+
  • Python 3.12+
  • npm 10+

Installing dependencies by platform:

Windows
winget install Python.Python.3.12
winget install OpenJS.NodeJS.LTS
macOS
brew install python@3.12 node@24
Linux (Ubuntu/Debian)
sudo apt install python3.12 python3.12-venv
curl -fsSL https://deb.nodesource.com/setup_24.x | sudo -E bash -
sudo apt install -y nodejs
Linux (Fedora)
sudo dnf install python3.12 nodejs npm

See CONTRIBUTING.md for detailed development setup.

Building Flatpak

To build the Flatpak package, you need additional dependencies:

# Fedora/RHEL
sudo dnf install flatpak-builder

# Ubuntu/Debian
sudo apt install flatpak-builder

# Install required Flatpak runtimes
flatpak install flathub org.freedesktop.Platform//25.08 org.freedesktop.Sdk//25.08
flatpak install flathub org.electronjs.Electron2.BaseApp//25.08

# Build the Flatpak
cd apps/frontend
npm run package:flatpak

The Flatpak will be created in apps/frontend/dist/.


Security

Auto Claude uses a three-layer security model:

  1. OS Sandbox - Bash commands run in isolation
  2. Filesystem Restrictions - Operations limited to project directory
  3. Dynamic Command Allowlist - Only approved commands based on detected project stack

All releases are:

  • Scanned with VirusTotal before publishing
  • Include SHA256 checksums for verification
  • Code-signed where applicable (macOS)

Available Scripts

Command Description
npm run install:all Install backend and frontend dependencies
npm start Build and run the desktop app
npm run dev Run in development mode with hot reload
npm run package Package for current platform
npm run package:mac Package for macOS
npm run package:win Package for Windows
npm run package:linux Package for Linux
npm run package:flatpak Package as Flatpak
npm run lint Run linter
npm test Run frontend tests
npm run test:backend Run backend tests

Contributing

We welcome contributions! Please read CONTRIBUTING.md for:

  • Development setup instructions
  • Code style guidelines
  • Testing requirements
  • Pull request process

Community


License

AGPL-3.0 - GNU Affero General Public License v3.0

Auto Claude is free to use. If you modify and distribute it, or run it as a service, your code must also be open source under AGPL-3.0.

Commercial licensing available for closed-source use cases.

S
Description
Autonomous multi-session AI coding
Readme AGPL-3.0 59 MiB
Languages
TypeScript 98.7%
JavaScript 0.6%
CSS 0.5%
Shell 0.2%