Let Cursor Cook
vKongv/let-cursor-cook
Orquestra Cursor CLI agents como equipe técnica subordinada
Ver no GitHubAgent portável que orquestra Cursor CLI como subordinado técnico para implementação e revisão de código. Permite que agentes host (Codex, Claude Code) assumam papel de gerente enquanto Cursor agents lidam com pesquisa de codebase, implementação, execução de comandos e debugging.
Instalação
npx convez-ai vKongv/let-cursor-cook1 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
- Delegação de implementação técnica para agents Cursor
- Revisão de código com agents Cursor especializados
- Executar passa fechadas em Cursor com goals explícitos
- Coordenar múltiplos agents em worktrees isolados
Veredito

Skill sólida para quem já trabalha com Cursor e quer integração programática com outros agentes. Requer entendimento de orquestração multi-agent e CLI—não é entrada para iniciantes, mas resuelve bem o padrão manager/executor.
README
A tradução para o português deste README ainda está na fila.
Let Cursor Cook
A portable AI skill for Codex, Claude Code, and other external coding harnesses that can shell out to Cursor's agent CLI.
Use it when you want the host agent to act like a manager/team lead while Cursor agents handle the technical cooking: codebase research, implementation, command execution, debugging, and code review.
Install
Install with skills.sh:
npx skills add vkongv/let-cursor-cook
Install for specific agents:
npx skills add vkongv/let-cursor-cook -a codex -a cursor -a claude-code
What It Teaches
- Keep the host harness responsible for requirements, planning, user communication, and final verification.
- Keep the host harness in a manager/orchestrator role, not an implementer role.
- Delegate technical work to Cursor CLI agents by default.
- Use explicit Cursor chat IDs for serious orchestration.
- Use isolated Cursor worktrees for risky or parallel implementation.
- Use separate Cursor reviewer agents for non-trivial code changes.
- Send valid review-fix code changes back to Cursor instead of patching them locally.
- Run closed-loop Cursor passes with explicit goals, checks, exit conditions, and iteration caps.
- Keep native subagents available for context hygiene and verification sidecars.
- Escalate to native subagents or local implementation if Cursor fails twice on the same task.
Requirements
- Cursor Agent CLI available as
agentorcursor agent - A host AI harness that can run shell commands
Check Cursor Agent:
agent --help
Recommended implementation command:
agent -p --trust --sandbox enabled --workspace /path/to/repo "Implement the scoped change and run relevant local tests..."
Use -f only when headless Cursor still blocks necessary local commands:
agent -p -f --trust --sandbox enabled --workspace /path/to/repo "Implement the scoped change. Allowed commands: npm test and npm run lint..."
Skill
The skill lives in SKILL.md.