# Kanban > Stand up a personal Kanban board and wire this agent to it, end to end. Serve the board, publish it to a public URL, verify the connection, and hand the user their live link. Each user gets their OWN board at their OWN URL — self-contained, no shared server, no collisions. Use when the user wants to set up / publish / deploy their Kanban board, or connect their agent to it. For day-to-day task operations (create/move/list tasks) use the nested companion skill `kanban-agent` (in kanban_agent_skill/) instead — it is auto-discovered when this skill is installed. - Skill: `starchild-ai-agent/kanban` (Agent Skill, multi-file: 6 files) - Install (CLI): `npx skillmds add starchild-ai-agent/kanban` - Raw SKILL.md: https://api.skillmd.com/api/skills/starchild-ai-agent/kanban/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - Author: starchild-ai-agent (https://skillmd.com/u/starchild-ai-agent) - Updated: 2026-09-09 - Page: https://skillmd.com/skills/starchild-ai-agent/kanban --- ## What this is > **The official agent connector is `kanban_agent_skill/`.** This is THE skill > to hand to any agent — your own, a teammate's, anyone's — to give it read/write > access to a Kanban board. Give that folder out, install it, and the agent is > wired to the board. Everything else in this directory (`server.js`, > `kanban.html`) is for *hosting* a board; `kanban_agent_skill/` is for *using* > one. A self-contained Kanban system that lives in this skill's own folder (`skills/kanban/`): - **`kanban.html`** — the board UI (React, single file). `API_BASE` is derived from `window.location.pathname`, so the same file works locally and behind any published path prefix. No edits needed on publish. - **`server.js`** — Node server on port **5555**, file-backed (`data.json`), exposing the `/ajax/tasks/*` API the UI and the agent both use. - **`kanban_agent_skill/`** — the **connector skill for agents** (registered as `kanban-agent`, auto-discovered when this skill is installed). This is the folder an agent — yours or a collaborator's — uses to read/write the board. It **auto-resolves** the board URL at runtime (see its SKILL.md), so once you publish, the agent is wired up with zero edits and the file never needs updating. **Deployment model — one board per user.** Each user serves their own `server.js` and publishes it to `https://community.iamstarchild.com/{USER_ID}-kanban-board`. Your agent connects to *your* board. There's no shared server and no shared `data.json`, so two users never collide. (To collaborate on one board, use `kb_use_board(url)` from the companion skill to point at someone else's URL.) --- ## Setup — publish your board (run these in order) ### 1. Serve the board locally The board files (`server.js`, `kanban.html`) live in **this skill's own directory** — normally `skills/kanban`. Serve from there; do NOT copy the files anywhere else (copies drift). Use the `preview` tool (NOT a raw `node` call — the preview registry is what `publish_preview` reads): ``` preview(action="serve", title="Kanban Board", dir="skills/kanban", command="node server.js", port=5555) ``` If a preview is already registered (check `/data/previews.json` for a `kanban`-suffixed id), reuse its id and make sure its `dir` points at `skills/kanban` — don't serve a duplicate. ### 2. Publish it to a public URL Load the **community-publish** skill (`read_file` its SKILL.md), then: ```python publish_preview(preview_id="", slug="kanban-board", title="Kanban Board") ``` This maps the running service to `https://community.iamstarchild.com/{USER_ID}-kanban-board`. The `USER_ID` prefix is added automatically — keep the slug exactly `kanban-board` so the companion skill's auto-detect finds it without extra config. > Requires running inside the Starchild container (needs `FLY_MACHINE_ID`). > Publishing only allocates the URL — it does NOT list the board on the public > gallery. That's a separate, deliberate `list_in_dashboard()` call if wanted. ### 3. Record the published URL in the connector's SKILL.md After publishing, take the **actual** public URL returned by `publish_preview` (or read it from `/data/previews.json`) and write it into the connector skill's doc so the live board link is captured in the skill itself, not just guessed. Edit `skills/kanban/kanban_agent_skill/SKILL.md` and replace the placeholder board URL with the real published URL. Specifically, in its **Overview** section change the auto-detect line: ``` 2. `https://community.iamstarchild.com/{USER_ID}-kanban-board` — your published board, if reachable. ``` to the concrete URL you just published, e.g.: ``` 2. `https://community.iamstarchild.com/1357-kanban-board` — your published board (live). ``` Use `edit_file` for this, then run `skill_refresh` so the change is reloaded. This keeps the connector's documentation pointing at the real live board rather than a placeholder. (The runtime resolver still works either way — this is so the URL is written down, per request.) ### 4. Verify the connection Load the companion skill (`read_file skills/kanban/kanban_agent_skill/SKILL.md`) and run: ```python kb_health() ``` Expect `resolved_via: "community (auto)"`, `ok: true`, and `resolved_url` matching your `{USER_ID}-kanban-board` URL. If it resolved to `localhost`, the publish didn't take — recheck step 2. ### 5. Hand the user their link Give them the live URL from step 2, e.g. `https://community.iamstarchild.com/{USER_ID}-kanban-board/`. On Telegram/WeChat, send the full public URL (no preview panel there). --- ## After setup — using the board Day-to-day task management is the companion skill's job. Its functions auto-resolve to the board you just published — no arguments needed: ```python from exports import kb_health, kb_summary, kb_create_task, kb_move_task kb_summary() # orient: columns + task counts ``` See `kanban_agent_skill/SKILL.md` for the full function reference. --- ## Collaborate — give a colleague's agent access to your board Your board is one shared URL. To let a teammate's agent read/write it, they need the **connector skill only** — not the whole board (they don't run a server). **Prepare a pre-wired copy** (recommended — no manual config for the recipient). From the connector skill, run: ```python kb_export_for_sharing() ``` This stamps a copy of `kanban_agent_skill/` with **your** board URL and zips it to `output/kanban-agent-share/`. Send that folder/zip. The recipient: 1. Drops `kanban_agent_skill/` into their own `skills/` directory. 2. Their agent runs **skill_refresh** (registers it as `kanban-agent`). 3. Runs `kb_health()` — it auto-connects to **your** board, zero config. Why not just send the raw folder? A raw copy auto-detects the *recipient's own* board, not yours — stamping is what wires it to your board. (Manual alternative: send the folder and have them set `KANBAN_URL=` in their `.env` or call `kb_use_board()`.) Remember the API is open: anyone with the URL has full read/write. --- ## Notes & gotchas - **Keep the slug `kanban-board`.** The companion skill auto-detects `{USER_ID}-kanban-board`. A different slug means you must set the `KANBAN_URL` env var so the agent can find it. - **The board must stay running.** `publish_preview` maps a *running* service; if the container restarts, the preview auto-starts from `/data/previews.json` (whose entry must point at `/data/workspace/skills/kanban`). - **No password / open API.** The UI loads directly and the API is unauthenticated. Anyone who knows the published URL can read/write the board. Don't publish sensitive data unless you're comfortable with that. To lock it down you'd add auth on the `/ajax/tasks/*` routes in `server.js`. - **Local dev without publishing:** run `node server.js` in `skills/kanban/` and open `http://localhost:5555` — the companion skill falls back to localhost automatically when no community board is reachable.