Developers experimenting with AI coding assistants are increasingly building structured workflows around them—rather than relying on one-off prompts. A project highlighted on Hacker News, the “Cook” CLI, is designed to formalize those patterns.

Cook describes itself as “workflow loops for Claude Code, Codex, and OpenCode.” Instead of manually repeating prompts, it provides loop operators that can:

- Run multiple sequential passes (xN / repeat)

- Add a review → gate → iterate cycle (review)

- Coordinate task progression through a plan (ralph)

- Run parallel branches in isolated worktrees and resolve the outcome (vN / race, vs, pick/merge/compare)

Why this is useful for web development teams

- Consistency: standardizes iteration and review steps across contributors.

- Quality control: review gates can encode criteria (security, accessibility, performance) before accepting changes.

- Exploration: parallel “race” runs help compare approaches quickly, which is valuable for refactors and UX work.

Risks and best practices

- Keep strong tests and CI: workflow automation should not replace verification.

- Limit secrets exposure: ensure agent sandboxes and config avoid leaking tokens.

- Capture decisions: use the compare/merge outputs to document why a solution was selected.

Project links

- GitHub: https://github.com/rjcorwin/cook

- npm package: https://www.npmjs.com/package/@let-it-cook/cli

Draft angle for Kicukiro Tech: As agentic coding becomes mainstream, the differentiator shifts from “which model” to “which workflow.” Tools like Cook point toward a future where AI assistants are orchestrated like build pipelines.