FAQ

Is KanbAI tied to Claude Code?

No. The core is harness-agnostic — the board is just Markdown files driven by the kanbai CLI. Claude Code gets an auto-installed integration, but any assistant that can run shell commands works. See Other AI harnesses.

Do I commit the .kanbai/ folder?

Usually yes — the point is that the board lives in your repo and travels with it, with a clean git history of every card and move. If you'd rather keep it local, add .kanbai/ to .gitignore.

Can I edit card files directly?

Prefer the CLI. The kanbai edit / move commands keep the frontmatter valid and in sync with the folder a card lives in. Hand-editing risks an inconsistent status or a malformed header.

Why doesn't Claude pick up my backlog cards?

kanbai next reads only the sprint (todo), never the backlog — by design. Plan a card into the sprint with kanbai move <id> todo (or the UI's Plan sprint button) and it becomes actionable.

Why is a card being skipped even though it's next?

It's probably blocked by an unfinished dependency. kanbai next skips blocked cards, and the web UI shows a 🔒. Finish (approve to done) the cards in its deps and it becomes actionable.

Who moves a card to done?

You do. The agent finishes a card into review with kanbai review and stops; approving it into done with kanbai done is a human step.

Can I add or rename columns?

Yes — edit columns in config.toml. The role resolution falls back to position, so the workflow keeps working; keep the first column as the backlog and the last as done for the cleanest results.

The web UI won't update live in my container.

Live updates use OS file-change events. Where those aren't delivered, run kanbai ui --poll to fall back to polling. See Web UI › Live updates.

kanbai ui says it needs an extra.

Install the UI dependencies: pip install 'kanbai[ui]' (or uv add --dev 'kanbai[ui]'). See Installation.