Do you often find yourself in this loop?

You: Hey Agent, implement dark mode. Agent: Done! I added the thing.

You: Hey Agent, review your work. Agent: Found a few issues.

You: Hey Agent, fix your work. Agent: Fixed!

Issue the task once and let it cook:

cook "Implement dark mode"

cook runs Claude, Codex, or OpenCode in a work → review → gate loop, iterating automatically until the agent is satisfied or your max iterations are hit. Agents run natively by default, using their own OS-level sandboxes — no Docker required. Get even fancier by defining what to review and the criteria for done and upping the max iterations:

cook "Implement dark mode" \ "Review the implementation. Check for visual regressions and missing theme variables. Categorize findings by High/Medium/Low." \ "Reply DONE if no High findings remain; otherwise ITERATE." \ 5

Install:

npm install -g @let-it-cook/cli