invites experts to experience the thrill of multi-agent collaboration.

You can initiate a conversation in any main agent, and then other N sub-agents collaborate (supports claude code, codex, gemini, opencode, etc.).

For example, I often have such scenarios: using claude code for development, then codex for review (I feel codex has a stronger big-picture view, while claude code has strong coding capabilities). Then I send codex's review to claude code, and claude code mostly responds accordingly, then continue working, repeating manual operations each time.

After tinkering with it for a while, I kept thinking about implementing multi-agent collaboration. After finishing the requirements in claude code, codex should automatically review the code, then claude code evaluates, and then processes, automatically completing the entire process in a single session.

Multi-agent collaboration is now supported. We sincerely invite you to try it out. Additionally, you can use this feature in combination with some skills or tools to spark some unexpected creativity.

Open source address: https://github.com/xintaofei/codeg