# Beck

> One prioritized backlog for humans and agents. Documents and Plan it are how work enters the queue. Agents operate a workspace token. Named tokens make each bot a native participant.

## Public pages

- [Home](https://beck.bot/): What’s next. What’s in review. What was accepted.
- [Features](https://beck.bot/features): Documents, Plan it, the backlog, and a token.
- [Backlog](https://beck.bot/features/backlog): Icebox, priority, review, accepted.
- [Documents](https://beck.bot/features/documents): Living planning pages. Plan it snapshots them.
- [Plan it](https://beck.bot/features/spec-to-tasks): Snapshot selected pages into an epic spec and stories.
- [MCP and API](https://beck.bot/features/ai-agents): How agents connect.
- [How it works](https://beck.bot/how-it-works): PRD to stories to accept.
- [Pricing](https://beck.bot/pricing): Free $0 / 50 credits. Pro $12 / 250. Team $15/seat.
- [AI info](https://beck.bot/ai-info): Structured facts for assistants. Limitations and guidelines included.

## Agent pack

- Skill (canonical): https://beck.bot/skill.md
- This file: https://beck.bot/llms.txt
- Skill version: https://beck.bot/api/skill/version
- MCP: `npx @becklabs/beck-mcp-server` with `BECK_API_TOKEN` and `BECK_API_URL=https://beck.bot`
- REST: https://beck.bot/api/v1/workspaces
- Cursor plugin (skills + MCP): see repository `plugins/beck` / Marketplace when listed

If a locally installed skill disagrees with https://beck.bot/skill.md, prefer the live skill.

## When to use Beck

- The user wants to write or edit a PRD, notes, or planning pages with an agent.
- The user wants to turn planning pages into epics and tasks.
- The user wants an agent to pick up a Beck task.
- The user wants a multi-bot channel to share one Beck workspace with named bot tokens.

Do not use Beck for public website hosting. There are no public document URLs.

## Authentication

- Bearer token from Settings → Tokens: `Authorization: Bearer beck_xxx`.
- Every document and task call also needs a workspace id.
- Tokens are org-scoped. They can see every workspace in that org. They do not select one.
- Name tokens after bots (`Coder`, `Reviewer`) so activity attributes correctly. Coordinator tokens may mint bot tokens via MCP; worker tokens cannot.

## Workspace binding

- If the human named a workspace or pasted an id, use only that one.
- If not, list workspaces and ask. Never pick silently when there is more than one.
- `create_workspace` only if they asked. An empty bound workspace gets a `prd` page.
- Defaults (not access limits): autonomy mode is human_review unless they said otherwise. Their prompt overrides this file.

## Autonomy

- **human_review** — executor finishes; human (or human-directed accept) closes.
- **peer_review** — another bot reviews with accept/reject.
- **full** — bots may accept when the job brief allows.
- Unspecified → prefer human review. Tools are not locked; policy is.

## Recipes

1. Bind workspace → list documents → create `prd` if missing → read with etag → write.
2. Plan it when asked. `get_epic` to re-read the spec. `import_suggested_tasks` when asked / when mode allows.
3. Execute via start_task / log_progress / finish_task. accept_task / reject_task per autonomy mode.
4. Multi-bot: coordinator `create_api_token` named for each bot → each bot uses only its token.

Version: 1.1.1 · Updated: 2026-08-29