cwang · cwang · Aug 3, 2025 · Aug 3, 2025
diff --git a/plans/outer-loop-ai-responder-20250802/01-implementation-plan.md b/plans/outer-loop-ai-responder-20250802/01-implementation-plan.md
@@ -2,35 +2,41 @@
 
 ## 🚀 Final Implementation Plan
 
-**Ready-to-ship auto-pilot feature for CCManager in 1-2 weeks**
+**Ready-to-ship auto-pilot feature for CCManager in 2-3 weeks**
 
 Clean, focused implementation that adds intelligent LLM-based guidance to Claude Code sessions with minimal code changes and maximum user value.
 
 ## 📋 Complete Implementation Roadmap
 
-### Week 1: Core Auto-pilot (5 days)
+### Week 1: Core Auto-pilot (8 days)
 
 **PR1: Basic Auto-pilot Core** (3 days)
 - ✈️ Auto-pilot toggle with `'p'` keystroke
 - Basic LLM watchdog monitoring Claude Code output  
 - Simple intervention delivery to PTY terminal
 - Status indicator: `✈️ Auto-pilot: ACTIVE/STANDBY`
 
-**PR2: Pattern-Based Guidance** (2 days)
+**PR2: User Guide Prompt & Self-Updating** (5 days)
+- 🎨 Configurable guide prompts for personalized guidance
+- 🧠 LLM-based pattern learning from user interactions
+- 🔄 Self-updating prompts that evolve with user behavior
+- 🔒 Privacy controls with user review and approval
+
+### Week 2: Advanced Intelligence (4 days)
+
+**PR3: Pattern-Based Guidance** (2 days)
 - ⚡ Lightning-fast pattern detection (< 10ms)
 - 🚨 Error detection with immediate guidance
 - 🤔 Overthinking detection and progress nudges
 - Smart throttling: max 3 guidances/hour, critical always allowed
 
-### Week 2: Intelligence & Polish (5 days)
-
-**PR3: Context-Aware Intelligence** (2 days)  
+**PR4: Context-Aware Intelligence** (2 days)  
 - 🔍 Project type detection (React, Node.js, TypeScript, etc.)
 - 🛠 Framework-specific guidance patterns
 - 📊 Git status awareness and file change context
 - 🎯 Targeted guidance based on project characteristics
 
-**Testing & Polish** (3 days)
+### Week 3: Testing & Polish (3 days)
 - Comprehensive testing across project types
 - Performance optimization and error handling
 - Documentation and user experience refinement

diff --git a/plans/outer-loop-ai-responder-20250802/03-pr2-guide-prompt.md b/plans/outer-loop-ai-responder-20250802/03-pr2-guide-prompt.md
@@ -0,0 +1,193 @@
+# PR2: User Guide Prompt & Self-Updating Intelligence
+
+## 🎯 Goal
+Add configurable guide prompts that learn from user interactions to provide personalized, adaptive guidance for each developer's workflow.
+
+## ✨ Features Added
+- 🎨 **Simple Guide Prompt**: Text field for custom guidance instructions
+- 🧠 **Pattern Learning**: LLM-based extraction of user instruction themes
+- 🔄 **Self-Updating Prompts**: Automatically evolve based on user behavior
+- 🔒 **Privacy Controls**: Opt-in learning with user review and approval
+- 📊 **Pattern Analysis**: Detect recurring themes in user interactions
+- ⚙️ **Full User Control**: Manual override, pattern review, learning toggle
+
+## 📁 Implementation Approach
+
+### Phase 1: Simple Guide Prompt (Day 1)
+- **AutopilotConfig Extension**: Add `guidePrompt?: string` field
+- **LLMClient Enhancement**: Inject user guidance into analysis prompt
+- **UI Addition**: Textarea in autopilot configuration form
+- **Simple Integration**: Append guide prompt to existing analysis prompt
+
+### Phase 2: Input Monitoring (Day 2)
+- **UserInputPattern Interface**: Track user interactions with Claude
+- **PatternTracker Service**: Record user inputs, context, and timing
+- **Input Classification**: Categorize instructions, corrections, questions
+- **Privacy-First Storage**: Configurable data retention and opt-in
+
+### Phase 3: Pattern Learning Engine (Days 3-4)
+- **PatternLearner Service**: LLM-based analysis of user interaction patterns
+- **Theme Extraction**: Identify recurring preferences and workflow patterns
+- **PromptEvolver Service**: Generate updated guide prompts from learned patterns
+- **Confidence Scoring**: Rate pattern reliability before suggesting updates
+
+### Phase 4: User Control & Integration (Day 5)
+- **Pattern Review UI**: Show learned patterns before applying
+- **Approval Workflow**: User can approve, reject, or modify suggestions
+- **Learning Controls**: Toggle learning on/off, clear learned patterns
+- **Transparency**: Show which guidance comes from manual vs learned prompts
+
+## 🧠 Learning Strategy
+
+### What Gets Monitored
+```typescript
+interface UserInputPattern {
+  sessionId: string;
+  timestamp: Date;
+  input: string;
+  context: string; // Recent Claude output that prompted this input
+  inputType: 'instruction' | 'correction' | 'question';
+  isGuidanceRelated: boolean; // LLM determines if relevant for learning
+}
+```
+
+### Pattern Categories
+- **Code Style**: "Use TypeScript strict mode", "Write tests first"
+- **Workflow**: "Check existing utilities", "Follow project patterns"
+- **Architecture**: "Keep components small", "Prefer composition"
+- **Communication**: "Be more concise", "Explain your reasoning"
+- **Testing**: "Test edge cases", "Mock external dependencies"
+
+### Learning Examples
+```typescript
+// User frequently says: "Write tests first"
+// System learns: { 
+//   category: 'testing',
+//   instruction: 'Emphasize test-driven development',
+//   confidence: 0.85
+// }
+
+// User often corrects: "Use existing utility functions"
+// System learns: {
+//   category: 'workflow', 
+//   instruction: 'Check for existing utilities before implementing',
+//   confidence: 0.92
+// }
+```
+
+## 🔒 Privacy & Control Design
+
+### User Controls
+- **Learning Toggle**: Disabled by default, clear explanation
+- **Pattern Review**: See detected patterns before they're applied
+- **Manual Override**: Always maintain ability to edit final prompt
+- **Data Management**: Clear patterns, export/import, retention controls
+- **Transparency**: Visual indicators of manual vs learned guidance sources
+
+### Learning Approval Flow
+1. System detects patterns from recent interactions
+2. User receives notification: "New guidance patterns detected"
+3. Review dialog shows proposed additions to guide prompt
+4. User can approve all, approve some, or reject
+5. Approved patterns integrated into guide prompt
+6. User can always manually edit the final result
+
+## ⚙️ Configuration Schema
+
+### AutopilotConfig Extension
+```typescript
+interface AutopilotConfig {
+  // ... existing fields
+  guidePrompt?: string; // Manual guidance instructions
+  learningConfig?: {
+    enabled: boolean; // Opt-in learning
+    approvalRequired: boolean; // Always true for now
+    retentionDays: number; // Default 30 days
+    minPatternConfidence: number; // Default 0.7
+  };
+}
+
+interface LearnedPattern {
+  id: string;
+  category: 'style' | 'workflow' | 'testing' | 'architecture' | 'communication';
+  instruction: string;
+  confidence: number;
+  frequency: number;
+  lastSeen: Date;
+  approved: boolean;
+}
+```
+
+### Prompt Generation
+```typescript
+// Final analysis prompt structure:
+const finalPrompt = `
+${baseAnalysisPrompt}
+${projectContext}
+
+USER'S GUIDANCE INSTRUCTIONS:
+${config.guidePrompt || 'No custom guidance provided'}
+
+LEARNED USER PREFERENCES:
+${approvedPatterns.map(p => `- ${p.instruction}`).join('\n')}
+
+Focus guidance on these user preferences while maintaining general helpfulness.
+`;
+```
+
+## 🧪 Testing Approach
+
+### Learning Accuracy
+- **Pattern Detection**: Validate correct theme extraction from sample inputs
+- **False Positive Prevention**: Ensure one-off comments don't become patterns
+- **Confidence Calibration**: Test that confidence scores correlate with pattern quality
+- **Noise Filtering**: Verify non-guidance inputs are correctly ignored
+
+### User Experience
+- **Approval Flow**: Test pattern review and approval workflow
+- **Privacy Controls**: Validate learning can be disabled/enabled cleanly
+- **Performance**: Ensure pattern analysis doesn't slow down autopilot
+- **Transparency**: Verify users understand source of guidance
+
+### Integration
+- **Prompt Quality**: Test that learned patterns improve guidance relevance
+- **Manual Override**: Ensure user can always edit final prompt
+- **Data Safety**: Test pattern storage and privacy controls
+
+## 📋 Acceptance Criteria
+
+### Phase 1: Simple Guide Prompt
+- [ ] Guide prompt field in autopilot configuration UI
+- [ ] User guidance properly injected into LLM analysis prompt
+- [ ] Guide prompt persists in configuration storage
+- [ ] Examples and helpful placeholder text provided
+
+### Phase 2-4: Self-Updating Intelligence
+- [ ] User inputs monitored with proper privacy controls
+- [ ] LLM correctly extracts recurring patterns from user interactions
+- [ ] Pattern review UI shows learned patterns before applying
+- [ ] Users can approve, reject, or modify suggested patterns
+- [ ] Learning can be toggled on/off with clear data implications
+- [ ] Learned patterns improve guidance relevance and personalization
+- [ ] Manual prompt editing always available as override
+- [ ] Pattern data respects retention policies and privacy controls
+
+## 🚀 Estimated Timeline: 5 days
+- **Day 1**: Simple guide prompt implementation and UI
+- **Day 2**: Input monitoring infrastructure and storage
+- **Days 3-4**: Pattern learning engine and prompt evolution
+- **Day 5**: User control interface and approval workflow
+
+## 🔮 Future Enhancements
+- **Cross-Session Learning**: Patterns learned across multiple projects
+- **Team Pattern Sharing**: Export/import patterns between team members
+- **Pattern Categories**: More granular categorization and control
+- **A/B Testing**: Compare guidance effectiveness with different prompts
+- **Adaptive Confidence**: Adjust learning sensitivity based on user feedback
+
+## 🎯 Success Metrics
+- **Adoption**: % of users who configure guide prompts
+- **Learning Engagement**: % of users who enable pattern learning
+- **Pattern Quality**: User approval rate for suggested patterns
+- **Guidance Relevance**: Improved user satisfaction with autopilot suggestions
+- **Personalization**: Measurable differences in guidance style per user
diff --git a/...onder-20250802/03-pr2-pattern-guidance.md → ...onder-20250802/04-pr3-pattern-guidance.md b/...onder-20250802/03-pr2-pattern-guidance.md → ...onder-20250802/04-pr3-pattern-guidance.md
@@ -1,4 +1,4 @@
-# PR2: Pattern-Based Guidance
+# PR3: Pattern-Based Guidance
 
 ## 🎯 Goal
 Add fast pattern recognition for common Claude Code issues and smart guidance delivery.

diff --git a/...nder-20250802/04-pr3-context-awareness.md → ...nder-20250802/05-pr4-context-awareness.md b/...nder-20250802/04-pr3-context-awareness.md → ...nder-20250802/05-pr4-context-awareness.md
@@ -1,4 +1,4 @@
-# PR3: Context-Aware Intelligence
+# PR4: Context-Aware Intelligence
 
 ## 🎯 Goal
 Make auto-pilot aware of project context to provide smarter, framework-specific guidance and enable future workflow automation features like context-aware PR creation.

diff --git a/plans/outer-loop-ai-responder-20250802/README.md b/plans/outer-loop-ai-responder-20250802/README.md
@@ -10,9 +10,10 @@ plans/outer-loop-ai-responder-20250802/
 ├── 00-autopilot-overview.md       # Feature overview and concept
 ├── 01-implementation-plan.md      # Complete implementation roadmap
 ├── 02-pr1-basic-autopilot.md     # PR1: Basic auto-pilot core (3 days)
-├── 03-pr2-pattern-guidance.md    # PR2: Pattern-based guidance (2 days)
-├── 04-pr3-context-awareness.md   # PR3: Context-aware intelligence (2 days)
+├── 03-pr2-guide-prompt.md        # PR2: User guide prompt & self-updating (5 days)
+├── 04-pr3-pattern-guidance.md    # PR3: Pattern-based guidance (2 days)
 ├── 05-context-aware-pr-feature.md # Context-aware PR creation design (4 days)
+├── 05-pr4-context-awareness.md   # PR4: Context-aware intelligence (2 days)
 ├── 06-modular-architecture.md    # Modular design for portability
 └── 07-extended-roadmap.md        # Advanced PRs for complete system
 ```
@@ -23,31 +24,27 @@ plans/outer-loop-ai-responder-20250802/
 **Ready-to-ship auto-pilot for CCManager**
 
 - **PR1**: Basic auto-pilot toggle and LLM monitoring
-- **PR2**: Fast pattern detection and smart guidance  
-- **PR3**: Context-aware intelligence and project understanding
-- **Result**: Working auto-pilot that helps Claude Code sessions stay productive
+- **PR2**: User guide prompts with self-updating intelligence
+- **PR3**: Fast pattern detection and smart guidance  
+- **PR4**: Context-aware intelligence and project understanding
+- **Result**: Working auto-pilot that learns user preferences and helps Claude Code sessions stay productive
 
-### **Phase 2: Modular System** (Weeks 3-4)
-**Portable, extensible intelligence platform**
+### **Phase 2: Workflow Automation** (Weeks 3-4)
+**Context-aware workflow automation and enhanced intelligence**
 
-- **PR4**: Plugin system and advanced framework patterns
-- **PR5**: Multi-session intelligence coordination
-- **PR6**: Learning and adaptation from user feedback  
-- **Result**: Sophisticated system that improves over time and works across projects
+- **PR5**: Context-aware PR creation design
+- **PR6**: Plugin system and advanced framework patterns
+- **PR7**: Multi-session intelligence coordination
+- **Result**: Automated workflow assistance with intelligent PR creation
 
 ### **Phase 3: Production Platform** (Weeks 5-6)
 **Enterprise-ready intelligent development assistant**
 
-- **PR7**: Analytics dashboard and performance insights
-- **PR8**: Advanced LLM providers and code generation capabilities
+- **PR8**: Advanced learning and adaptation from user feedback
+- **PR9**: Analytics dashboard and performance insights
+- **PR10**: Advanced LLM providers and code generation capabilities
 - **Result**: Comprehensive platform ready for team deployment
 
-### **Phase 4: Workflow Automation** (Week 7)
-**Context-aware workflow automation and PR creation**
-
-- **PR9**: Context-aware PR creation with session analysis and compliance validation
-- **Result**: Complete workflow automation that transforms session insights into actionable PRs
-
 ## ✈️ Auto-pilot Quick Start
 
 ### **What It Does**
@@ -104,24 +101,20 @@ autopilot.on('guidanceProvided', handleGuidance);
 ### **For MVP Implementation** (Weeks 1-2)
 1. **Start**: `00-autopilot-overview.md` - Understand the concept
 2. **Plan**: `01-implementation-plan.md` - Review complete roadmap  
-3. **Implement**: `02-pr1` → `03-pr2` → `04-pr3` - Follow PR sequence
+3. **Implement**: `02-pr1` → `03-pr2` → `04-pr3` → `05-pr4` - Follow PR sequence
 4. **Deploy**: Ready-to-ship auto-pilot feature!
 
-### **For Modular System** (Weeks 3-4)
-1. **Architecture**: `06-modular-architecture.md` - Understand modular design
-2. **Advanced Features**: `07-extended-roadmap.md` - Review PR4-PR6 plans
-3. **Implement**: Follow extended PR sequence for plugins and coordination
+### **For Workflow Automation** (Weeks 3-4)
+1. **Context-Aware PRs**: `05-context-aware-pr-feature.md` - Detailed design and implementation plan
+2. **Architecture**: `06-modular-architecture.md` - Understand modular design
+3. **Advanced Features**: `07-extended-roadmap.md` - Review PR6-PR7 plans
+4. **Implement**: Follow extended PR sequence for plugins and coordination
 
 ### **For Production Platform** (Weeks 5-6)  
-1. **Analytics**: `07-extended-roadmap.md` - PR7 analytics implementation
-2. **Advanced AI**: `07-extended-roadmap.md` - PR8 multi-provider support
+1. **Analytics**: `07-extended-roadmap.md` - PR9 analytics implementation
+2. **Advanced AI**: `07-extended-roadmap.md` - PR10 multi-provider support
 3. **Deploy**: Enterprise-ready intelligent development platform
 
-### **For Workflow Automation** (Week 7)
-1. **Context-Aware PRs**: `05-context-aware-pr-feature.md` - Detailed design and implementation plan
-2. **GitHub Integration**: `07-extended-roadmap.md` - PR9 implementation guide
-3. **Deploy**: Complete workflow automation with PR creation capabilities
-
 ## 🎯 Key Benefits
 
 ### **For Developers**