Skip to content

Repository files navigation

codoop-flow

English · 简体中文

Turn ideas into actionable requirements, then let AI build, verify, and review the code.

codoop-flow is a development workflow plugin for Codex, Claude Code, and Cursor. Describe what you want in plain language; it helps clarify requirements, break down tasks, and implement them. Use it to plan a new product, add features, or fix bugs in an existing project.

Install

Requires Git and Python 3.11+. Use it inside a local Git project.

Codex (Desktop / CLI) — run in your terminal, then reopen Codex:

codex plugin marketplace add Codoop/codoop-flow
codex plugin add codoop-flow@codoop-flow

Claude Code — run in your session:

/plugin marketplace add Codoop/codoop-flow
/plugin install codoop-flow@codoop-flow

Using Cursor, another agent, or having trouble? See the full installation guide.

Get started

Open your project and send these three prompts to your coding agent, one step at a time.

1. Set up your project (once)

Use codoop-init to inspect this project and set up codoop-flow.

The agent checks your project structure, creates the configuration, and asks for your language and role preferences.

2. Describe a feature

Use codoop-ticket to design a ticket for letting users search and filter orders.

Replace the example with your requirement. The agent clarifies key questions and creates requirements, an implementation plan, and a task list, with a viewable UI preview when needed. After your confirmation, the ticket enters the queue. You don't need to write ticket files yourself.

3. Let the agent implement it

Use codoop-execute to run the next ticket in this project.

The agent implements the ticket on a separate branch, runs verification and reviews, then commits the code and archives the ticket. If verification or review fails, it attempts fixes within a retry budget; if it cannot finish, it leaves a report for you. You decide whether to merge and when to push.

For each subsequent feature, repeat steps 2 and 3.

Other common tasks

What you want What to tell the agent
Explore a product idea before building Use codoop-discover to plan an order management tool for small teams.
Fix an existing bug Use codoop-ticket to create a fix ticket for search filters being lost when changing pages.
Implement a ticket already in the queue Use codoop-execute to run the next ticket in this project.
Check the experience from a user's perspective Use codoop-ux-walkthrough to try order search as a first-time operations manager.

Use these capabilities independently; you don't need to start with product planning every time.

Documentation

MIT License

About

Turn "AI writes code" into a ticket pipeline with guardrails: product design → ticket design → agent-driven implementation with deterministic verification.

Topics

Resources

Stars

5 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages