MCP Prompts Server

# Learning Loop Enhancements for dev-intelligence-orchestrator Skill ## Overview All skill scripts have been enhanced with **mcp-prompts integration** to enable a self-improving learning system. The scripts now follow a consistent learning pattern: 1. **Before execution**: Query mcp-prompts for learned configurations 2. **During execution**: Use learned config if available, fallback to defaults 3. **After execution**: Capture successful configurations for future use ## Enhanced Scripts ### 1. `mcp_query.sh` (NEW) **Purpose**: HTTP API wrapper for querying mcp-prompts server **Usage**: ```bash # List prompts ./scripts/mcp_query.sh list [category] [limit] # Get specific prompt ./scripts/mcp_query.sh get <prompt_id> [version] # Search prompts ./scripts/mcp_query.sh search <query> [category] # Create prompt ./scripts/mcp_query.sh create <json_data> # Update prompt ./scripts/mcp_query.sh update <prompt_id> <json_data> ``` **Features**: - Graceful degradation if server unavailable - Health check before operations - Supports all mcp-prompts HTTP API endpoints ### 2. `analyze_cpp.sh` (ENHANCED) **Learning Behavior**: - Queries for `cppcheck` configurations matching project type and focus - Uses learned flags if available - Captures successful configurations with metrics - Updates existing prompts with validation data **Example Output**: ``` 🔍 Checking for accumulated knowledge... ✓ Found 1 relevant knowledge item(s) ✓ Using learned configuration from: cppcheck-config-embedded-esp32-memory-optimized-2025-12-30 🔧 Running cppcheck on src/bpm_detector.cpp with focus: memory... 💡 Capturing successful configuration... ✓ Configuration captured for future use ``` ### 3. `analyze_python.sh` (ENHANCED) **Learning Behavior**: - Queries for `pylint` configurations - Applies learned options - Captures successful configurations ### 4. `run_tests.sh` (ENHANCED) **Learning Behavior**: - Queries for test framework configurations (pytest, platformio, etc.) - Uses learned test options - Captures successful test execution patterns ### 5. `parse_build_errors.py` (ENHANCED) **Learning Behavior**: - Queries for similar error patterns based on error signature - Uses learned diagnosis if available - Captures novel error patterns for future reference **Error Signature**: Creates a signature from top error categories (e.g., "flatbuffers_version_mismatch:3 missing_dependency:2") ## Seed Prompts Initial seed prompts have been created in `data/prompts/tool-config/`: 1. `cppcheck-config-embedded-esp32-memory-default.json` 2. `cppcheck-config-embedded-esp32-security-default.json` 3. `cppcheck-config-desktop-general-default.json` 4. `pylint-config-python-general-default.json` 5. `pylint-config-python-security-default.json` 6. `pytest-config-python-default.json` These provide starting points that the learning system will validate and improve. ## Learning Flow ### First Execution (No Knowledge) ``` 1. Query mcp-prompts → No results found 2. Use default configuration 3. Execute tool 4. If useful (found issues), capture configuration 5. Create new prompt with success_count=1, confidence=low ``` ### Subsequent Executions (With Knowledge) ``` 1. Query mcp-prompts → Found matching configuration 2. Use learned configuration 3. Execute tool 4. If useful, update existing prompt: - Increment success_count - Update last_used timestamp - Increase confidence (low → medium → high) ``` ### Confidence Levels - **low**: 1 successful use - **medium**: 2-3 successful uses - **high**: 4+ successful uses ## Configuration Structure Prompts store tool configurations in this format: ```json { "project_type": "embedded-esp32", "focus": "memory", "cppcheck_flags": ["--enable=warning,performance", "--std=c++11"], "success_count": 3, "confidence": "medium", "last_used": "2025-12-31T18:00:00Z", "last_findings": 12, "last_high_severity": 3 } ``` ## Graceful Degradation All scripts handle mcp-prompts unavailability gracefully: - If `mcp_query.sh` not found → Use defaults, warn user - If server not running → Use defaults, warn user - If query fails → Use defaults, continue execution - Learning is **optional**, not required for tool execution ## Testing the Learning Loop ### Test 1: First Run (No Knowledge) ```bash ./scripts/analyze_cpp.sh src/main.cpp memory . # Should: Use defaults, capture configuration ``` ### Test 2: Second Run (With Knowledge) ```bash ./scripts/analyze_cpp.sh src/main.cpp memory . # Should: Use learned configuration, update success_count ``` ### Test 3: Verify Learning ```bash ./scripts/mcp_query.sh search "cppcheck memory embedded-esp32" # Should: Show captured configuration with success_count > 0 ``` ## Integration with mcp-prompts Server The scripts require the mcp-prompts HTTP server to be running: ```bash # Start server (if not already running) MODE=http STORAGE_TYPE=file pnpm start:http # Or use existing server # Default: http://localhost:3000 # Override: MCP_PROMPTS_URL=http://other-host:3000 ``` ## Next Steps 1. **Use the enhanced scripts** in your development workflow 2. **Monitor learning** - Check captured prompts periodically 3. **Validate improvements** - Run same analysis twice, verify second uses learned config 4. **Expand patterns** - Add more seed prompts for other tools/configurations 5. **Cross-project learning** - Share learned configurations across projects ## Files Modified/Created - ✅ `scripts/mcp_query.sh` (NEW) - ✅ `scripts/analyze_cpp.sh` (ENHANCED) - ✅ `scripts/analyze_python.sh` (ENHANCED) - ✅ `scripts/run_tests.sh` (ENHANCED) - ✅ `scripts/parse_build_errors.py` (ENHANCED) - ✅ `scripts/detect_project_type.sh` (COPIED) - ✅ `scripts/seed-tool-config-prompts.js` (NEW) - ✅ `data/prompts/tool-config/*.json` (6 seed prompts created) ## Success Criteria ✅ **Learning is visible**: Scripts report when knowledge is found/used/captured ✅ **Graceful degradation**: Works without mcp-prompts server ✅ **Knowledge accumulation**: Second run uses learned configuration ✅ **Confidence increases**: Success count and confidence tracked ✅ **Cross-project sharing**: Learned configs available to all projects