Cook CLI: Orchestrate Claude Code, Codex, and OpenAI Like a Pro

If you are using Claude Code, Codex, or OpenCode for coding tasks, you are probably running them one prompt at a time. That is like using a power tool without any attachments—functional, but missing half the potential.

Cook is a free, open-source CLI tool that wraps your AI coding agents with powerful workflow primitives. Run multiple agents in parallel, add review gates, iterate on results, and compose complex automation pipelines—all from a simple command line.

What Cook Actually Does

Cook parses three types of tokens: work (the prompt), loop operators (iteration), and composition (parallel execution). You combine them into commands that do surprisingly complex things:

The Primitives That Matter

1. Repeat (xN)

Run the same work N times sequentially. Each pass sees the previous output, so the agent can build on its own work. This is perfect for iterative improvements:

2. Review Gate

After the work is done, a reviewer checks quality and a gate decides DONE or ITERATE. On ITERATE, it runs an "iterate" step, then repeats. You can customize prompts:

This is huge. It means your AI agent can self-correct until it meets your standards—no manual hand-holding required.

3. Versions (vN)

Run N identical cooks in parallel using isolated git worktrees, then pick the best result. This is essentially automated A/B testing for code:

4. VS (Comparison)

Run two different approaches in parallel, then pick the winner or merge results:

Why This Matters for Indie Hackers

As an indie hacker, you probably do not have a team of developers reviewing your code. Cook gives you:

• Automated quality gates — no more shipping broken code
• Parallel experimentation — test multiple approaches in the time it takes one to run
• Task automation — let agents work through a plan.md file autonomously
• Agent flexibility — mix and match Claude, Codex, and OpenCode in the same workflow

Getting Started

Installation is a two-step process:

Then run cook init in your project to scaffold the config files:

• COOK.md — project instructions and agent prompt template
• .cook/config.json — agent/model/sandbox defaults
• .cook/Dockerfile — project dependencies for Docker sandbox mode
• .cook/logs/ — session logs (gitignored)

The Bottom Line

Cook transforms your AI coding agent from a fancy autocomplete into a workflow engine. You can now build systems that:

• Generate multiple solutions and pick the best one
• Self-review and iterate until code passes your standards
• Run complex multi-step projects with minimal supervision

The best part? It is free and open source. No API keys, no subscriptions, just a CLI tool that makes your existing AI agents dramatically more powerful.

---