diff --git a/apps/frontend/src/shared/__tests__/ai-triage-constants.test.ts b/apps/frontend/src/shared/__tests__/ai-triage-constants.test.ts
new file mode 100644
index 00000000..b521b2ab
--- /dev/null
+++ b/apps/frontend/src/shared/__tests__/ai-triage-constants.test.ts
@@ -0,0 +1,161 @@
+/**
+ * Tests for AI triage constants and utility functions.
+ */
+import { describe, it, expect } from 'vitest';
+import {
+  CONFIDENCE_HIGH,
+  CONFIDENCE_MEDIUM,
+  DEFAULT_BATCH_SIZE,
+  DEFAULT_CONFIRM_ABOVE,
+  MAX_SPLIT_SUB_ISSUES,
+  APPLY_INTER_ITEM_DELAY,
+  THRESHOLD_MIN,
+  THRESHOLD_MAX,
+  THRESHOLD_STEP,
+  TRUST_LEVEL_LABELS,
+  ENRICHMENT_COMMENT_FOOTER,
+  getConfidenceLevel,
+  isValidThreshold,
+  clampThreshold,
+  estimateBatchCost,
+} from '../constants/ai-triage';
+import type { TrustLevel } from '../constants/ai-triage';
+
+describe('confidence constants', () => {
+  it('CONFIDENCE_HIGH is 0.8', () => {
+    expect(CONFIDENCE_HIGH).toBe(0.8);
+  });
+
+  it('CONFIDENCE_MEDIUM is 0.5', () => {
+    expect(CONFIDENCE_MEDIUM).toBe(0.5);
+  });
+});
+
+describe('batch constants', () => {
+  it('DEFAULT_BATCH_SIZE is 50', () => {
+    expect(DEFAULT_BATCH_SIZE).toBe(50);
+  });
+
+  it('DEFAULT_CONFIRM_ABOVE is 10', () => {
+    expect(DEFAULT_CONFIRM_ABOVE).toBe(10);
+  });
+
+  it('MAX_SPLIT_SUB_ISSUES is 5', () => {
+    expect(MAX_SPLIT_SUB_ISSUES).toBe(5);
+  });
+
+  it('APPLY_INTER_ITEM_DELAY is 100', () => {
+    expect(APPLY_INTER_ITEM_DELAY).toBe(100);
+  });
+});
+
+describe('threshold constants', () => {
+  it('THRESHOLD_MIN is 0.5', () => {
+    expect(THRESHOLD_MIN).toBe(0.5);
+  });
+
+  it('THRESHOLD_MAX is 1.0', () => {
+    expect(THRESHOLD_MAX).toBe(1.0);
+  });
+
+  it('THRESHOLD_STEP is 0.05', () => {
+    expect(THRESHOLD_STEP).toBe(0.05);
+  });
+});
+
+describe('TRUST_LEVEL_LABELS', () => {
+  it('has entry for every trust level', () => {
+    const allLevels: TrustLevel[] = ['crawl', 'walk', 'run'];
+    for (const level of allLevels) {
+      expect(TRUST_LEVEL_LABELS).toHaveProperty(level);
+      expect(typeof TRUST_LEVEL_LABELS[level]).toBe('string');
+    }
+  });
+});
+
+describe('ENRICHMENT_COMMENT_FOOTER', () => {
+  it('contains Auto-Claude marker', () => {
+    expect(ENRICHMENT_COMMENT_FOOTER).toContain('Auto-Claude');
+  });
+
+  it('starts with horizontal rule', () => {
+    expect(ENRICHMENT_COMMENT_FOOTER).toMatch(/^---/);
+  });
+});
+
+describe('getConfidenceLevel', () => {
+  it('returns high for >= 0.8', () => {
+    expect(getConfidenceLevel(0.8)).toBe('high');
+    expect(getConfidenceLevel(0.95)).toBe('high');
+    expect(getConfidenceLevel(1.0)).toBe('high');
+  });
+
+  it('returns medium for >= 0.5 and < 0.8', () => {
+    expect(getConfidenceLevel(0.5)).toBe('medium');
+    expect(getConfidenceLevel(0.79)).toBe('medium');
+  });
+
+  it('returns low for < 0.5', () => {
+    expect(getConfidenceLevel(0.49)).toBe('low');
+    expect(getConfidenceLevel(0.1)).toBe('low');
+    expect(getConfidenceLevel(0)).toBe('low');
+  });
+});
+
+describe('isValidThreshold', () => {
+  it('returns true for values in [0.5, 1.0]', () => {
+    expect(isValidThreshold(0.5)).toBe(true);
+    expect(isValidThreshold(0.75)).toBe(true);
+    expect(isValidThreshold(1.0)).toBe(true);
+  });
+
+  it('returns false for values below 0.5', () => {
+    expect(isValidThreshold(0.49)).toBe(false);
+    expect(isValidThreshold(0)).toBe(false);
+    expect(isValidThreshold(-1)).toBe(false);
+  });
+
+  it('returns false for values above 1.0', () => {
+    expect(isValidThreshold(1.01)).toBe(false);
+    expect(isValidThreshold(2)).toBe(false);
+  });
+
+  it('returns false for non-numbers', () => {
+    expect(isValidThreshold(Number.NaN)).toBe(false);
+  });
+});
+
+describe('clampThreshold', () => {
+  it('returns value when in range', () => {
+    expect(clampThreshold(0.75)).toBe(0.75);
+  });
+
+  it('clamps to min when below', () => {
+    expect(clampThreshold(0.1)).toBe(0.5);
+    expect(clampThreshold(-1)).toBe(0.5);
+  });
+
+  it('clamps to max when above', () => {
+    expect(clampThreshold(1.5)).toBe(1.0);
+    expect(clampThreshold(99)).toBe(1.0);
+  });
+});
+
+describe('estimateBatchCost', () => {
+  it('estimates haiku cost correctly', () => {
+    expect(estimateBatchCost(100, 'haiku')).toBe('~$0.08');
+  });
+
+  it('estimates non-haiku cost correctly', () => {
+    expect(estimateBatchCost(100, 'sonnet')).toBe('~$0.35');
+  });
+
+  it('handles small batches', () => {
+    expect(estimateBatchCost(1, 'haiku')).toBe('~$0.00');
+  });
+
+  it('handles large batches', () => {
+    const cost = estimateBatchCost(1000, 'haiku');
+    expect(cost).toBe('~$0.80');
+  });
+});
diff --git a/apps/frontend/src/shared/__tests__/ai-triage-types.test.ts b/apps/frontend/src/shared/__tests__/ai-triage-types.test.ts
new file mode 100644
index 00000000..b6c947a5
--- /dev/null
+++ b/apps/frontend/src/shared/__tests__/ai-triage-types.test.ts
@@ -0,0 +1,94 @@
+/**
+ * Tests for AI triage type factories and utilities.
+ */
+import { describe, it, expect } from 'vitest';
+import {
+  createDefaultProgressiveTrust,
+  mapTriageCategory,
+} from '../types/ai-triage';
+import type { ProgressiveTrustConfig } from '../types/ai-triage';
+
+describe('createDefaultProgressiveTrust', () => {
+  it('returns all categories disabled', () => {
+    const config = createDefaultProgressiveTrust();
+    expect(config.autoApply.type.enabled).toBe(false);
+    expect(config.autoApply.priority.enabled).toBe(false);
+    expect(config.autoApply.labels.enabled).toBe(false);
+    expect(config.autoApply.duplicate.enabled).toBe(false);
+  });
+
+  it('returns 0.9 threshold for all categories', () => {
+    const config = createDefaultProgressiveTrust();
+    expect(config.autoApply.type.threshold).toBe(0.9);
+    expect(config.autoApply.priority.threshold).toBe(0.9);
+    expect(config.autoApply.labels.threshold).toBe(0.9);
+    expect(config.autoApply.duplicate.threshold).toBe(0.9);
+  });
+
+  it('returns default batch size of 50', () => {
+    const config = createDefaultProgressiveTrust();
+    expect(config.batchSize).toBe(50);
+  });
+
+  it('returns default confirmAbove of 10', () => {
+    const config = createDefaultProgressiveTrust();
+    expect(config.confirmAbove).toBe(10);
+  });
+
+  it('returns a fresh object each time (no shared references)', () => {
+    const a = createDefaultProgressiveTrust();
+    const b = createDefaultProgressiveTrust();
+    expect(a).not.toBe(b);
+    expect(a.autoApply).not.toBe(b.autoApply);
+  });
+});
+
+describe('mapTriageCategory', () => {
+  it('maps bug to bug', () => {
+    expect(mapTriageCategory('bug')).toBe('bug');
+  });
+
+  it('maps feature to feature', () => {
+    expect(mapTriageCategory('feature')).toBe('feature');
+  });
+
+  it('maps documentation to documentation', () => {
+    expect(mapTriageCategory('documentation')).toBe('documentation');
+  });
+
+  it('maps question to question', () => {
+    expect(mapTriageCategory('question')).toBe('question');
+  });
+
+  it('maps duplicate to bug (flagged separately via isDuplicate)', () => {
+    expect(mapTriageCategory('duplicate')).toBe('bug');
+  });
+
+  it('maps spam to chore', () => {
+    expect(mapTriageCategory('spam')).toBe('chore');
+  });
+
+  it('maps feature_creep to enhancement', () => {
+    expect(mapTriageCategory('feature_creep')).toBe('enhancement');
+  });
+
+  it('maps unknown categories to chore', () => {
+    expect(mapTriageCategory('unknown_xyz')).toBe('chore');
+  });
+
+  it('maps enhancement to enhancement', () => {
+    expect(mapTriageCategory('enhancement')).toBe('enhancement');
+  });
+
+  it('maps chore to chore', () => {
+    expect(mapTriageCategory('chore')).toBe('chore');
+  });
+
+  it('maps security to security', () => {
+    expect(mapTriageCategory('security')).toBe('security');
+  });
+
+  it('maps performance to performance', () => {
+    expect(mapTriageCategory('performance')).toBe('performance');
+  });
+});
diff --git a/apps/frontend/src/shared/constants/ai-triage.ts b/apps/frontend/src/shared/constants/ai-triage.ts
new file mode 100644
index 00000000..7ae605d6
--- /dev/null
+++ b/apps/frontend/src/shared/constants/ai-triage.ts
@@ -0,0 +1,73 @@
+/**
+ * AI triage constants and utility functions for Phase 3 (AI Power).
+ */
+
+// ============================================
+// Confidence Thresholds
+// ============================================
+
+export const CONFIDENCE_HIGH = 0.8;
+export const CONFIDENCE_MEDIUM = 0.5;
+
+// ============================================
+// Batch Limits
+// ============================================
+
+export const DEFAULT_BATCH_SIZE = 50;
+export const DEFAULT_CONFIRM_ABOVE = 10;
+export const MAX_SPLIT_SUB_ISSUES = 5;
+export const APPLY_INTER_ITEM_DELAY = 100;
+
+// ============================================
+// Threshold Bounds
+// ============================================
+
+export const THRESHOLD_MIN = 0.5;
+export const THRESHOLD_MAX = 1.0;
+export const THRESHOLD_STEP = 0.05;
+
+// ============================================
+// Trust Levels
+// ============================================
+
+export type TrustLevel = 'crawl' | 'walk' | 'run';
+
+export const TRUST_LEVEL_LABELS: Record<TrustLevel, string> = {
+  crawl: 'Suggestions Only',
+  walk: 'Auto-Apply Above Threshold',
+  run: 'Fully Automated',
+};
+
+// ============================================
+// Enrichment Comment
+// ============================================
+
+export const ENRICHMENT_COMMENT_FOOTER = '---\n*Generated by Auto-Claude AI Triage*';
+
+// ============================================
+// Utility Functions
+// ============================================
+
+export function getConfidenceLevel(confidence: number): 'high' | 'medium' | 'low' {
+  if (confidence >= CONFIDENCE_HIGH) return 'high';
+  if (confidence >= CONFIDENCE_MEDIUM) return 'medium';
+  return 'low';
+}
+
+export function isValidThreshold(value: number): boolean {
+  if (Number.isNaN(value)) return false;
+  return value >= THRESHOLD_MIN && value <= THRESHOLD_MAX;
+}
+
+export function clampThreshold(value: number): number {
+  return Math.max(THRESHOLD_MIN, Math.min(THRESHOLD_MAX, value));
+}
+
+/**
+ * Estimates the cost of a batch triage operation.
+ * Uses rough per-issue token cost estimates.
+ */
+export function estimateBatchCost(issueCount: number, model: string): string {
+  const costPerIssue = model.includes('haiku') ? 0.0008 : 0.0035;
+  return `~$${(issueCount * costPerIssue).toFixed(2)}`;
+}
diff --git a/apps/frontend/src/shared/types/ai-triage.ts b/apps/frontend/src/shared/types/ai-triage.ts
new file mode 100644
index 00000000..a0f8862e
--- /dev/null
+++ b/apps/frontend/src/shared/types/ai-triage.ts
@@ -0,0 +1,160 @@
+/**
+ * AI triage types for Phase 3 (AI Power).
+ * Progressive trust, enrichment, splitting, and review queue types.
+ */
+import type { TriageCategory as EnrichmentTriageCategory } from './enrichment';
+
+// Re-export TriageResult from triage-handlers for convenience
+// (cannot import directly due to main/renderer boundary — consumers
+// in renderer use IPC; only main-process code imports triage-handlers)
+
+// ============================================
+// Progressive Trust
+// ============================================
+
+export interface ProgressiveTrustConfig {
+  autoApply: {
+    type: { enabled: boolean; threshold: number };
+    priority: { enabled: boolean; threshold: number };
+    labels: { enabled: boolean; threshold: number };
+    duplicate: { enabled: boolean; threshold: number };
+  };
+  batchSize: number;
+  confirmAbove: number;
+}
+
+// ============================================
+// AI Enrichment
+// ============================================
+
+export interface AIEnrichmentResult {
+  issueNumber: number;
+  problem: string;
+  goal: string;
+  scopeIn: string[];
+  scopeOut: string[];
+  acceptanceCriteria: string[];
+  technicalContext: string;
+  risksEdgeCases: string[];
+  confidence: number;
+}
+
+// ============================================
+// Issue Splitting
+// ============================================
+
+export interface SplitSuggestion {
+  issueNumber: number;
+  subIssues: Array<{
+    title: string;
+    body: string;
+    labels: string[];
+  }>;
+  rationale: string;
+  confidence: number;
+}
+
+export interface CreateIssueParams {
+  title: string;
+  body: string;
+  labels?: string[];
+  assignees?: string[];
+}
+
+export interface CreateIssueResult {
+  number: number;
+  url: string;
+}
+
+// ============================================
+// Triage Review Queue
+// ============================================
+
+export type TriageReviewStatus = 'pending' | 'accepted' | 'rejected' | 'auto-applied';
+
+export interface TriageReviewItem {
+  issueNumber: number;
+  issueTitle: string;
+  result: {
+    category: string;
+    confidence: number;
+    labelsToAdd: string[];
+    labelsToRemove: string[];
+    isDuplicate: boolean;
+    duplicateOf?: number;
+    isSpam: boolean;
+    isFeatureCreep: boolean;
+    suggestedBreakdown: string[];
+    priority: 'high' | 'medium' | 'low';
+    comment?: string;
+    triagedAt: string;
+  };
+  status: TriageReviewStatus;
+}
+
+// ============================================
+// Progress Types
+// ============================================
+
+export interface EnrichmentProgress {
+  phase: 'analyzing' | 'generating' | 'complete';
+  progress: number;
+  message: string;
+}
+
+export interface SplitProgress {
+  phase: 'analyzing' | 'suggesting' | 'creating' | 'closing' | 'complete';
+  progress: number;
+  message: string;
+  createdCount?: number;
+  totalCount?: number;
+}
+
+export interface ApplyResultsProgress {
+  totalItems: number;
+  processedItems: number;
+  currentIssueNumber?: number;
+}
+
+// ============================================
+// Factory Functions
+// ============================================
+
+export function createDefaultProgressiveTrust(): ProgressiveTrustConfig {
+  return {
+    autoApply: {
+      type: { enabled: false, threshold: 0.9 },
+      priority: { enabled: false, threshold: 0.9 },
+      labels: { enabled: false, threshold: 0.9 },
+      duplicate: { enabled: false, threshold: 0.9 },
+    },
+    batchSize: 50,
+    confirmAbove: 10,
+  };
+}
+
+// ============================================
+// Category Mapping
+// ============================================
+
+/**
+ * Maps Python triage runner categories to enrichment TriageCategory.
+ * The runner may return 'duplicate', 'spam', or 'feature_creep' which
+ * don't exist in the enrichment type system — this function bridges them.
+ */
+export function mapTriageCategory(category: string): EnrichmentTriageCategory {
+  const mapping: Record<string, EnrichmentTriageCategory> = {
+    bug: 'bug',
+    feature: 'feature',
+    documentation: 'documentation',
+    question: 'question',
+    enhancement: 'enhancement',
+    chore: 'chore',
+    security: 'security',
+    performance: 'performance',
+    duplicate: 'bug',
+    spam: 'chore',
+    feature_creep: 'enhancement',
+  };
+  return mapping[category] ?? 'chore';
+}