Files
2025-11-29 18:16:25 +08:00

11 KiB

Meta-Automation Architect

A meta-skill that analyzes your project and generates a comprehensive automation system with custom subagents, skills, commands, and hooks.

What It Creates

The meta-skill generates:

  1. Custom Subagents - Specialized analysis and implementation agents that run in parallel
  2. Skills - Auto-invoked capabilities for common patterns in your project
  3. Commands - Slash commands for frequent workflows
  4. Hooks - Event-driven automation at lifecycle points
  5. MCP Configurations - External service integrations
  6. Complete Documentation - Usage guides and quick references

How to Use

Basic Invocation

Simply describe what you want:

"Set up automation for my project"

Or be more specific:

"Create comprehensive automation for my Next.js e-commerce project"
"Generate a custom automation system for my Python data science workflow"

What Happens

  1. Interactive Discovery - You'll be asked questions about:

    • Project type (with smart detection and recommendations)
    • Tech stack and frameworks
    • Team size and workflow
    • Pain points and priorities
    • Desired automation scope
  2. Smart Recommendations - Every question includes:

    • Data-driven analysis of your project
    • Confidence scores and reasoning
    • Recommended options based on evidence
    • Clear trade-offs and explanations
  3. Multi-Agent Generation - The system creates:

    • A coordinator agent that orchestrates everything
    • Specialized analysis agents (security, performance, quality, etc.)
    • Implementation agents (skill/command/hook generators)
    • Validation agents (testing and documentation)
  4. Parallel Execution - Agents run concurrently and communicate via the Agent Communication Protocol (ACP)

  5. Complete Delivery - You receive:

    • All automation artifacts
    • Comprehensive documentation
    • Usage examples
    • Customization guides

Example Sessions

Web Application Project

User: "Set up automation for my React TypeScript project"

Meta-Skill:
1. Detects: Web application (95% confidence)
   - Found package.json with React dependencies
   - Found src/App.tsx and TypeScript config
   - Detected testing with Jest and React Testing Library

2. Asks: "What are your main pain points?"
   - Recommends: Testing automation (detected low test coverage)
   - Recommends: Code quality checks (found 47 bug-fix commits recently)

3. Recommends: 6 agents for comprehensive coverage
   - Analysis: Security, Performance, Code Quality, Dependencies
   - Implementation: Skill Generator, Command Generator
   - Validation: Integration Tester

4. Generates automation system with:
   - /test-fix command for TDD workflow
   - PostToolUse hook for auto-formatting
   - GitHub MCP integration for PR automation
   - Custom skills for common React patterns

Python Data Science Project

User: "Create automation for my machine learning project"

Meta-Skill:
1. Detects: Data Science (88% confidence)
   - Found notebooks/ directory with 15 .ipynb files
   - Found requirements.txt with pandas, scikit-learn, tensorflow
   - Found data/ and models/ directories

2. Asks: "What would you like to automate first?"
   - Recommends: Experiment tracking (detected many model versions)
   - Recommends: Documentation generation (missing architecture docs)
   - Recommends: Data validation (found data pipeline code)

3. Generates automation system with:
   - /run-experiment command for standardized ML runs
   - Custom skill for model comparison and analysis
   - Hooks for auto-documenting experiments
   - MCP integration for MLflow or Weights & Biases

Agent Communication Protocol (ACP)

The generated subagents communicate via a file-based protocol:

Directory Structure

.claude/agents/context/{session-id}/
  ├── coordination.json       # Tracks agent status and dependencies
  ├── messages.jsonl          # Append-only event log
  ├── reports/               # Standardized agent outputs
  │   ├── security-analyzer.json
  │   ├── performance-analyzer.json
  │   └── ...
  └── data/                  # Shared data artifacts
      ├── vulnerabilities.json
      ├── performance-metrics.json
      └── ...

How Agents Communicate

  1. Check Dependencies - Read coordination.json to see which agents have completed
  2. Read Context - Review reports from other agents
  3. Log Progress - Write events to messages.jsonl
  4. Share Findings - Create standardized report in reports/
  5. Share Data - Store detailed artifacts in data/
  6. Update Status - Mark completion in coordination.json

Report Format

Every agent writes a standardized JSON report:

{
  "agent_name": "security-analyzer",
  "timestamp": "2025-01-23T10:00:00Z",
  "status": "completed",
  "summary": "Found 5 security vulnerabilities requiring immediate attention",
  "findings": [
    {
      "type": "issue",
      "severity": "high",
      "title": "SQL Injection Risk",
      "description": "User input not sanitized in query builder",
      "location": "src/db/queries.ts:42",
      "recommendation": "Use parameterized queries",
      "example": "db.query('SELECT * FROM users WHERE id = ?', [userId])"
    }
  ],
  "metrics": {
    "items_analyzed": 150,
    "issues_found": 5,
    "time_taken": "2m 34s"
  },
  "recommendations_for_automation": [
    "Skill: SQL injection checker",
    "Hook: Validate queries on PreToolUse",
    "Command: /security-scan for quick checks"
  ]
}

What Gets Generated

1. Custom Subagents

Specialized agents tailored to your project:

  • Analysis Agents - Security, performance, code quality, dependencies, documentation
  • Implementation Agents - Generate skills, commands, hooks, MCP configs
  • Validation Agents - Test integration, validate documentation

Each agent:

  • Has communication protocol built-in
  • Knows how to coordinate with others
  • Writes standardized reports
  • Suggests automation opportunities

2. Skills

Auto-invoked capabilities for your specific patterns:

.claude/skills/
  ├── api-doc-generator/        # Generate API docs from code
  ├── tdd-enforcer/             # Test-driven development workflow
  ├── security-checker/         # Quick security validation
  └── ...

3. Commands

Slash commands for frequent tasks:

.claude/commands/
  ├── test-fix.md              # Run tests and fix failures
  ├── deploy-check.md          # Pre-deployment validation
  ├── security-scan.md         # Quick security audit
  └── ...

4. Hooks

Event-driven automation:

.claude/hooks/
  ├── format_on_save.py        # PostToolUse: Auto-format code
  ├── security_check.py        # PreToolUse: Validate operations
  └── run_tests.py             # Stop: Execute test suite

5. Documentation

Complete usage guides:

  • .claude/AUTOMATION_README.md - Main system documentation
  • .claude/QUICK_REFERENCE.md - Cheat sheet for all features
  • .claude/agents/context/{session-id}/ - Generation session details

Monitoring the Generation Process

While agents work, you can monitor progress:

# Watch agent status
watch -n 2 'cat .claude/agents/context/*/coordination.json | jq ".agents"'

# Follow live events
tail -f .claude/agents/context/*/messages.jsonl | jq

# Check completion
cat .claude/agents/context/*/coordination.json | \
  jq '.agents | to_entries | map(select(.value.status == "completed")) | map(.key)'

Customizing Generated Automation

All generated artifacts can be customized:

Modify Agents

# Edit agent behavior
vim .claude/agents/security-analyzer.md

# Adjust analysis focus, tools, or process

Customize Skills

# Update skill behavior
vim .claude/skills/api-doc-generator/SKILL.md

# Modify when skill triggers or what it does

Update Commands

# Change command behavior
vim .claude/commands/test-fix.md

# Adjust workflow or add arguments

Adjust Hooks

# Modify hook logic
vim .claude/hooks/format_on_save.py

# Change trigger conditions or actions

Troubleshooting

Agent Failed

# Check status
jq '.agents | to_entries | map(select(.value.status == "failed"))' \
  .claude/agents/context/{session-id}/coordination.json

# Find error
jq 'select(.from == "failed-agent") | select(.type == "error")' \
  .claude/agents/context/{session-id}/messages.jsonl | tail -1

# Options:
# 1. Retry the agent
# 2. Continue without it
# 3. Manual intervention

Missing Reports

# List generated reports
ls .claude/agents/context/{session-id}/reports/

# Check if agent completed
jq '.agents["agent-name"]' \
  .claude/agents/context/{session-id}/coordination.json

Review What Happened

# Full event log
cat .claude/agents/context/{session-id}/messages.jsonl | jq

# Agent-specific events
jq 'select(.from == "agent-name")' \
  .claude/agents/context/{session-id}/messages.jsonl

# Events by type
jq -s 'group_by(.type) | map({type: .[0].type, count: length})' \
  .claude/agents/context/{session-id}/messages.jsonl

Advanced Usage

Specify Agent Count

"Create automation with 8 parallel agents for comprehensive coverage"

Target Specific Areas

"Focus automation on security and testing"

Prioritize Implementation

"Generate skills and commands first, hooks later"

Re-run Analysis

# Generate new session with different configuration
# Previous sessions remain in .claude/agents/context/

Architecture

The meta-skill uses a multi-phase architecture:

  1. Discovery Phase - Interactive questioning with recommendations
  2. Setup Phase - Initialize communication infrastructure
  3. Analysis Phase - Parallel agent execution for deep analysis
  4. Synthesis Phase - Coordinator reads all reports and makes decisions
  5. Implementation Phase - Parallel generation of automation artifacts
  6. Validation Phase - Sequential testing and documentation checks
  7. Delivery Phase - Complete documentation and user report

Benefits

  • Parallel Execution - Multiple agents work concurrently
  • Isolated Contexts - Each agent has focused responsibility
  • Communication Protocol - Agents share findings reliably
  • Data-Driven - Recommendations based on actual project analysis
  • Comprehensive - Covers security, performance, quality, testing, docs
  • Customizable - All generated artifacts can be modified
  • Transparent - Full event log shows what happened
  • Reusable - Generated automation works immediately

Support

For issues or questions:

  1. Review agent reports in reports/
  2. Check message log in messages.jsonl
  3. Consult individual documentation
  4. Review session details in context directory

Generated automation is project-specific but follows Claude Code best practices for skills, commands, hooks, and MCP integration.