Skip to main content
Scribe ships a user-installable Agent Skill that teaches coding agents how to drive Scribe over its REST API — creating documents, merging edits safely into live collaborative sessions, searching, and working with comment threads.

Install the skill

The skills CLI installs the scribe skill into your agent of choice (Claude Code, Cursor, and others) from the public skeptrunedev/scribe-skill repository — no repository access or GitHub credentials required.
The app serves the same skill file publicly. For Claude Code:
For other agents, download /skill.md and place it wherever your agent loads skills from.

Give the agent a key

The skill authenticates every request with the SCRIBE_API_KEY environment variable.
1

Mint a key

In the app, open Settings → API keys and create a key. Choose Can edit if the agent should write documents, Can comment if it should only read plus leave comments and suggestions, or Can read for read/search only. The scrb_agt_... token is shown exactly once.
2

Export it for the agent

Put it in a gitignored .env or your agent’s secret store — not in a shell profile you keep under version control (a committed ~/.bashrc or dotfiles repo would leak the key), and never in a document or repository.
Each key acts as its own agent identity in your organization: documents it creates are owned by the agent, and its edits, comments, and suggestions show up in activity feeds with an agent badge — so automation stays auditable.

What the skill does

Once installed, asking your agent to work with Scribe (for example, “create a launch-plan doc in Scribe and fill in the outline”) triggers the skill, which instructs the agent to:
  1. Fetch the OpenAPI spec from /openapi.json and consult it for exact request shapes.
  2. Use version-aware writes — GET /content for the X-Doc-Version token, then PATCH /content with X-Base-Version — so its edits merge with anything humans are typing at the same moment, instead of overwriting it.
  3. Handle Scribe’s error envelope and retry loops (409 version_gone) correctly.
The underlying mechanics are documented in the API overview; the full endpoint catalog is in the API reference tab.