I am moving the article on the board.
https://github.com/cookyman74/my_harness
When working with AI, I realized that the results are much better when you have an "AI team that collaborates by dividing roles" rather than an "AI that does everything alone". So I made open source software to automate it.
To put it simply, if you say "build a harness for this project" to Claude Code, it's a tool that automatically creates an AI agent team tailored to your project. (In English, "build a harness for this project")
Example prompt
/my-harness I want to make a TCG game.
/my-harness I want to make a stock investment program.
/my-harness @{plan} I want to make a react-based app based on the above plan document.
What's good about it?
Usually, when you ask AI to do something complex, the quality drops because one AI tries to do everything from beginning to end. The harness breaks down the task and
Who will be in charge (roles such as analysts, writers, reviewers)
How to do it (work procedures and tools)
It automatically designs. Think of it as a kind of "factory that creates an AI team".
In addition, the work is cross-checked by the AI themselves and further cross-verified by external AI (Codex/Gemini), so the reliability of the results is high.
I'm using it myself, and it's pretty good.
You can use this for (There are many example prompts in the README that you can copy and paste)
Deep research (investigation from multiple angles → cross-verification → report)
Full-stack web development (design → front-end → back-end → QA)
Webtoon/YouTube content planning
Code review & refactoring
Automatic generation of technical documents, etc.
Good plugins to use together
context-mode
caveman
graphify
Reference
The license is Apache 2.0 (free to use)
Claude Code is the main one, and Codex runtime is also supported.
ps:
Please press a star because I'm a fanboy ㅋ
webcopilot I also made a Chrome extension app called and released it.
