sparesparrow/AGENTS/skills/dev-intelligence-orchestrator

dev-intelligence-orchestrator Skill

Overview

The dev-intelligence-orchestrator skill provides intelligent development tool orchestration with self-improving learning capabilities through mcp-prompts integration.

Features

Core Capabilities

• Project Type Detection: Automatically identifies languages, frameworks, and project nature
• Intelligent Build Error Analysis: Parses and diagnoses compilation errors with pattern recognition
• Static Code Analysis: C++ (cppcheck) and Python (pylint) analysis with learned configurations
• Test Execution: Framework-aware test running with pytest, PlatformIO, and more
• Self-Improving Learning: Captures successful configurations and reuses them automatically

Learning Loop

Every tool execution follows this pattern:

1. BEFORE: Query mcp-prompts for learned configurations
2. DURING: Use learned config if available, fallback to defaults
3. AFTER: Capture successful configurations for future use
4. NEXT: Automatically use learned configurations

Scripts

Core Scripts

detect_project_type.sh

Detects project characteristics:

• Languages (C++, Python, Kotlin, Java)
• Frameworks (PlatformIO, CMake, Conan, Gradle)
• Test frameworks (pytest, gtest, JUnit)
• Project nature (embedded, Android, desktop)
• FlatBuffers usage

Usage:

./detect_project_type.sh [directory]

parse_build_errors.py

Intelligent build error analysis with learning:

• Parses compilation, linking, dependency, and schema errors
• Generates diagnosis and recommendations
• Learns from similar error patterns
• Captures novel error patterns for future reference

Usage:

python3 parse_build_errors.py <log_file> <project_type> [build_system]

analyze_cpp.sh

C++ static analysis with cppcheck + learning:

• Queries for learned cppcheck configurations
• Uses learned flags when available
• Captures successful configurations
• Updates confidence based on success rate

Usage:

./analyze_cpp.sh <target> <focus> <project_root>
# focus: security|performance|memory|general

analyze_python.sh

Python static analysis with pylint + learning:

• Queries for learned pylint configurations
• Applies learned options
• Captures successful configurations
• Tracks success metrics

Usage:

./analyze_python.sh <target> <focus> <project_root>
# focus: security|performance|style|general

run_tests.sh

Test execution with framework detection + learning:

• Auto-detects test framework (pytest, PlatformIO, gtest, Gradle)
• Queries for learned test configurations
• Captures successful test patterns
• Supports coverage reporting

Usage:

./run_tests.sh <project_root> <test_path> <coverage>

Supporting Scripts

mcp_query.sh

HTTP API wrapper for mcp-prompts:

• Health checks
• List/search prompts
• Get specific prompts
• Create/update prompts
• Graceful degradation when server unavailable

Usage:

./mcp_query.sh <operation> [args...]
# operations: health|list|get|search|create|update|apply

seed-tool-config-prompts.js

Creates initial seed prompts for tool configurations:

• cppcheck configurations (embedded, desktop)
• pylint configurations (general, security)
• pytest configurations
• Ready for learning system validation

Usage:

node seed-tool-config-prompts.js

Learning Behavior

First Execution (No Knowledge)

🔍 Checking for accumulated knowledge...
ℹ No accumulated knowledge yet, using defaults (will capture learnings)
🔧 Running tool...
💡 Capturing successful configuration...
✓ Configuration captured for future use

Subsequent Executions (With Knowledge)

🔍 Checking for accumulated knowledge...
✓ Found 1 relevant knowledge item(s)
✓ Using learned configuration from: <prompt_id>
🔧 Running tool...
✓ Validating learned configuration...
✓ Configuration validated (success_count: 2, confidence: medium)

Confidence Levels

• low: 1 successful use
• medium: 2-3 successful uses
• high: 4+ successful uses

Configuration

Prerequisites

• mcp-prompts server running (optional, graceful degradation if unavailable)
• Analysis tools installed: pylint, cppcheck, pytest (as needed)
• jq for JSON parsing

Environment Variables

• MCP_PROMPTS_URL: mcp-prompts server URL (default: http://localhost:3000)
• PROJECT_ROOT: Project root directory (default: current directory)

Server Setup

# Start mcp-prompts server with file storage
MODE=http STORAGE_TYPE=file PROMPTS_DIR=./data pnpm start:http

Integration with mcp-prompts

Prompt Structure

Tool configurations are stored as prompts with this structure:

{
  "name": "cppcheck-config-embedded-esp32-memory-20251231",
  "description": "Successful cppcheck configuration for embedded-esp32 memory analysis",
  "template": {
    "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"
  },
  "category": "tool-config",
  "tags": ["cpp", "cppcheck", "memory", "embedded-esp32", "validated"]
}

Learning Domains

• Tool Configurations: cppcheck, pylint, pytest settings
• Error Patterns: Build error diagnosis and fixes
• Project Patterns: Project-specific optimizations
• Workflow Patterns: Successful development workflows

Usage Examples

Analyze C++ Code

# First run - captures learning
./analyze_cpp.sh src/main.cpp memory .

# Second run - uses learned config
./analyze_cpp.sh src/main.cpp memory .

Analyze Python Code

# Security-focused analysis
./analyze_python.sh src/auth.py security .

# General analysis
./analyze_python.sh src/utils.py general .

Parse Build Errors

# Analyze build log
python3 parse_build_errors.py build.log esp32 platformio

# Will learn from similar error patterns

Run Tests

# Run with coverage
./run_tests.sh . tests/ true

# Run specific test
./run_tests.sh . tests/test_auth.py false

Graceful Degradation

All scripts handle mcp-prompts unavailability gracefully:

• If server not running: Uses defaults, warns user
• If query fails: Uses defaults, continues execution
• Learning is optional, not required for tool execution

Success Criteria

The skill is successful when:

1. ✅ Claude reports learning status on every execution
2. ✅ Second analysis is faster/better than first due to learned configuration
3. ✅ User sees knowledge accumulating through visible capture messages
4. ✅ Confidence increases as more patterns are validated
5. ✅ Cross-project knowledge sharing works

Files Included

• detect_project_type.sh - Project detection
• parse_build_errors.py - Build error analysis with learning
• analyze_cpp.sh - C++ analysis with learning
• analyze_python.sh - Python analysis with learning
• run_tests.sh - Test execution with learning
• mcp_query.sh - mcp-prompts API wrapper
• seed-tool-config-prompts.js - Seed prompt generator
• SKILL.md - This documentation

Version

Version: 2.0.0 (with Learning Loop)
Last Updated: 2025-12-31
Status: Production Ready