Capo — AI skill for Claude Code

CAPO

[](https://github.com/whitewolfx7/capo/actions/workflows/ci.yml) [](https://github.com/whitewolfx7/capo/actions/workflows/codeql.yml) [](LICENSE)

**You run out of Claude Pro usage at 2pm. Your Codex quota is untouched. CAPO moves the whole agent team over and keeps going.**

CAPO runs one agent team across Claude Code and Codex. When the platform you are on reports a usage limit, every session writes a Markdown checkpoint, and the entire team, root included, relaunches on the other platform from those checkpoints. Two subscriptions you already pay for, one continuous session of work.

Status: v0.1, working, not yet verified against a real usage limit. See [What is not done](#what-is-not-done).

How it works

1. A run has one active platform at a time. Every session runs there.
2. Each role has a model per platform. Launching on a platform uses that platform's column.
3. On a usage limit, every session checkpoints, all of them close, and the team relaunches on the other platform. The root moves too.
4. When both platforms are capped, the run waits for the earliest known reset rather than flapping between them.
5. capo resume continues from the latest checkpoint set on disk, so a crash or a closed laptop costs you nothing.

The checkpoint is the whole trick. Because the team moves all at once, nobody on the new platform has any memory. Whatever each agent writes before it pauses is everything the other side gets. So checkpoints are plain Markdown you can read, edit, and paste into a chat window by hand.

Quickstart

git clone https://github.com/whitewolfx7/capo.git && cd capo
npm install && npm run build
node packages/cli/dist/main.js doctor

`doctor` checks Node, git, and both platform C