kankan coordinates AI coding agents (Claude Code and Codex CLI) working in parallel on one Git repository, taking tasks from a Trello board: one agent implements a card, an agent of a different type reviews it, integrates it into main and closes it. Ownership lives in Git refs changed only by compare-and-swap; Trello expresses human intent.
pipx install . # Python ≥ 3.10, standard library only
kankan skill install
cd ~/coding/myproject
kankan init https://trello.com/b/<shortLink>/<name>
kankan up claude=2 codex=1
kankan doctor # when something does not startThe agents integrate into the branch kankan-agents; main stays yours: git merge kankan-agents takes their work, kankan merge-request <branch> gives them yours.
- User manual: docs/user-manual.md
- Developer manual: docs/developer-manual.md
- Specification: the initial functional and technical specification, plus one pair per feature: board settings (draft)
- Kanban backend requirements: aispecs/kanban_requirements.md
- Adapter observations: docs/adapter-notes.md
- Implementation plan: aiplans/2026-11-08-kankan.md
Development: python -m venv .venv && .venv/bin/pip install -e '.[dev]' && .venv/bin/python -m pytest -q (tests marked tmux need tmux; tests marked trello_live run against a real test board when one is configured and skip otherwise, see the developer manual).
MIT, see LICENSE.