Autopilot — Agente IA Autônomo
mahmoud20138/Autopilot
Agente autônomo que executa projetos inteiros sem paradas
Ver no GitHubOrquestrador de agente IA que executa tarefas de forma totalmente autônoma, do início ao fim, sem pedir confirmações. Descobre automaticamente skills, servidores MCP e ferramentas CLI, decomõe objetivos em fases, mapeia para os melhores recursos disponíveis e executa em paralelo com monitoramento contínuo.
Instalação
npx convez-ai mahmoud20138/Autopilot1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Executar projetos de desenvolvimento end-to-end sem intervenção
- Automatizar pipelines complexos com múltiplas fases e dependências
- Descobrir e orquestrar skills, MCP servers e ferramentas CLI para tarefas específicas
Veredito

Skill robusto e bem-pensado que resolve o problema real de confirmações excessivas em agentes IA. Requer integração com Claude Code, OpenClaude ou similares, e é indicado para times que executam workflows complexos e previsíveis. Implementação técnica sólida, com suporte cross-platform e self-healing, mas exige confiança alta na autonomia do agente.
README
A tradução para o português deste README ainda está na fila.
Autopilot Agent Skill
Fully autonomous AI agent orchestrator — Give it a goal, it runs everything end-to-end without confirmation.
Works with Claude Code, OpenClaude, GitHub Copilot CLI, Cursor, and Kilo.
Why Autopilot?
Most AI coding assistants stop and ask "should I continue?" between every step. Autopilot doesn't.
You: /autopilot "build a REST API with JWT auth and tests"
Agent: [discovered 12 skills, 3 MCP servers, 6 CLI tools]
[6 phases detected → mapped to 4 skills + codegraph + playwright]
[planning → implementing → testing → reviewing → shipping]
Done. All tests pass. Ready to merge.
Zero confirmations. Full autonomy. End-to-end execution.
Quick Start
# Install the skill
npm i -g autopilot-agent-skill
# Or use npx (no install)
npx autopilot-agent-skill
CLI Commands
npx autopilot-agent-skill # Install the skill
npx autopilot-agent-skill --update # Update to latest version
npx autopilot-agent-skill --remove # Uninstall the skill
npx autopilot-agent-skill --list # Show installed skills
npx autopilot-agent-skill --version # Show current version
npx autopilot-agent-skill --help # Show help
Supported CLIs
| CLI | Status | Skills Directory |
|---|---|---|
| Claude Code | ✓ Supported | ~/.claude/skills/ |
| OpenClaude | ✓ Supported | ~/.openclaude/skills/ |
| GitHub Copilot | ✓ Supported | ~/.config/github-copilot/skills/ |
| Cursor | ✓ Supported | ~/.cursor/skills/ |
| Kilo | ✓ Supported | ~/.config/kilo/skills/ |
| Custom | ✓ Supported | Any path you specify |
How It Works
1. Memory Check → Read MEMORY.md; apply past decisions
2. Discovery → Scan skills, MCP servers, CLI tools, project context
3. Analysis → Parse goal, identify type/scope/constraints
4. Phase Detection → Decompose into minimal viable phases
5. Skill Mapping → Map each phase to best skill + MCP + CLI
6. Session Plan → Write checkpoint file (.local/session_plan.md)
7. Execution → Run phases sequentially or in parallel
8. Monitoring → Typecheck + build + tests after each phase
9. Verification → Final build, test, lint, security checks
10. Completion → Clean up, update memory, report
Key Features
- Smart Discovery — Automatically finds your skills, MCP servers, and CLI tools across 5 CLI platforms
- Memory-Aware — Reads past decisions from MEMORY.md before starting
- Phase Decomposition — Breaks any goal into minimal viable phases with dependency tracking
- Dynamic Skill Mapping — Routes each phase to your best installed skill + MCP tools + CLI tools
- Prompt Generation — 6 phase templates (Plan, Implement, Test, Review, Debug, Ship) with customization
- Parallel Execution — Runs independent phases concurrently via background agents
- Self-Healing — Retries failed phases with adjusted prompts; Root Cause Analysis + Incremental Rollback
- Context-Aware — Uses Codegraph MCP for codebase understanding; respects 10-file read limit per turn
- Checkpoint/Resume — Session plan enables resume from interruption
- Final Verification — Builds, tests, lints, and security scans before reporting done
- Zero Config — Works out of the box
Examples
Simple Task
/autopilot "add a health check endpoint to my API"
Bug Fix
/autopilot "fix the login timeout error"
Full Project
/autopilot "build a user authentication system with JWT, tests, and docs"
See examples/ for detailed walkthroughs.
Contributing
See CONTRIBUTING.md for development setup and guidelines.
License
MIT © Mahmoud20