KENBOARD

A kanban for BOTs.

For humans

Regenerate the screenshot after a UI change: pdm run screenshots

For BOTs

KENBOARD ships ken, a CLI built for Claude Code and other assistants: JSON output, native filters, clean exit codes.

Automatic onboarding

1. An admin clicks Copy onboard link on a project in the kenboard
2. The link is handed to the agent (Claude Code, GPT, etc.)
3. The agent opens the link and gets a complete .ken file with a pre-filled API token — zero human interaction for the API key
4. The agent runs pip install kenboard, creates the .ken, and starts working immediately

Daily workflow

ken list --status todo --who Claude --json   # assigned tasks
ken show <id> --json                         # task details
ken move <id> --to doing                     # mark in progress
ken add "MODULE / Title" --desc "..." --who Claude  # create
ken move <id> --to review                    # submit
ken wiki groom <id> <section>                # classify for the wiki

Full workflow: todo → doing → review → groom → done. The agent handles todo → doing → review then ken wiki groom. Only the user moves review → done. ken move --to review and ken update --status review print a one-line reminder about the grooming step so the agent doesn't forget.

Wiki — exporting the board as a structured doc tree (#376)

Inspired by Karpathy's LLM Wiki pattern: the kanban board is the source of truth, and an LLM agent classifies each task into a section of ARCHITECTURE.md (a YAML-frontmatter file at the project root that declares the wiki's section tree). The CLI then exports the result as a navigable MD/HTML tree.

ken wiki groom                       # list unclassified + sections
ken wiki groom <id> <section>        # assign a task to a section
ken wiki sync   --out wiki           # MD tree (committed in /wiki)
ken wiki build  --in wiki --out wiki-html   # standalone HTML
ken wiki lint   --strict             # orphans / unclassified (CI gate)

• The server stays unaware of ARCHITECTURE.md — it stores opaque (task_id, section_path) pairs, the CLI does the validation.
• Each section index splits "En cours" (todo/doing/review) and "Archivé" (done). Per-task pages mirror the board's full-screen task view (header + meta + description).
• ken wiki lint --strict exits 1 on orphans / unclassified tasks — wire it into your CI to keep the wiki in sync with the board.

A live snapshot of the export lives under /wiki in this repo.

Browser extension — quick-capture from any page (#480)

Sideload-only browser extension that pops a "create task" dialog from the current tab with Ctrl+Shift+K (Cmd+Shift+K macOS). Pre-fills the title from the page title, attaches a PNG screenshot of the visible viewport, posts to POST /api/v1/tasks with your existing api_key. Zero new server code — reuses the standard bearer auth.

Install + first-run guide: extension/README.md. Pre-zipped sideload package on every release: GitHub Releases.

References

• Full CLI: doc/ken-cli.md
• Agent guide: ken help
• REST API: doc/api.md, doc/openapi.yaml

Enterprise

KENBOARD is designed for self-hosted enterprise deployment:

• OIDC authentication — sign in through a corporate Identity Provider (Microsoft ADFS, Google Workspace, Authentik, Keycloak, etc.) alongside or instead of password login. See doc/oidc-adfs.md for the ADFS guide.
• Self-hosted — no cloud dependencies. MySQL + Flask + gunicorn on your own infrastructure, behind your reverse proxy / WAF.
• Per-project API keys — each agent or integration gets a scoped token (read/write) for a specific project. AI agents self-onboard through the runbook served by the server.
• Commercial support — setup assistance, IdP integration, and operational support available on request. Contact: 2113.ch

Installation

See INSTALL.md for the full setup (MySQL, users, migrations, reverse proxy, OIDC).

License

MIT — see LICENSE.