Docs Skills

The catalog lives at github.com/Docsbook-io/docs-skills.

What is a Skill?#

A Skill is a reusable workflow — a set of Guardrails, Steps, and Acceptance Criteria encoded in a SKILL.md file. Think of it as a QA Checklist: you apply it in any project, with any agent (Claude Code, Cursor, Copilot, Codex). It does not carry project-specific context.

A Subagent is an executor: it has a pinned model, specific tools, priorities, and an autonomous work plan. Subagents live in docs-subagents. They are built for a specific context and are not meant to be reused across projects.

Rule of thumb: "Would this workflow be useful in another project?" — yes → skill, no → subagent.

The four skills#

Documentation work is four jobs, and every request lands in one of them. The catalog is four orchestrator skills, one per job, and they hand off to each other rather than overlapping: docs-analyze finds a gap and hands it to docs-create; docs-create writes to docs-manage's rules; docs-manage executes what docs-analyze diagnosed; docs-automate makes any of it recur.

The catalog is open source and free to install. Running a skill with your own agent costs you nothing here — it is your agent, on your machine, spending your model budget.

Three consumption modes#

Local install#

Copy SKILL.md files into your agent's local skill directory:

npx skills add Docsbook-io/docs-skills --skill '*'          # the whole catalog
npx skills add Docsbook-io/docs-skills --skill docs-analyze  # one skill

The installer detects Claude Code, Cursor, and Cline layouts and writes to .claude/skills/, .cursor/rules/, or AGENTS.md as appropriate. Add -a claude-code -a cursor to target specific agents. Skills work offline once installed.

Runtime discovery via MCP#

If the agent is connected to the Docsbook MCP server, it can discover skills on demand:

find_skill({ query: "audit my docs for SEO" })

The tool returns matching SKILL.md entries with a raw_url field. The agent fetches the file directly from GitHub and follows its instructions. No local install required.

Have Docsbook run it for you#

The two modes above both end with your agent executing the skill, which needs it to be connected here, to have picked a workspace, and to be willing to spend twenty tool calls. The third mode hands the whole job to Docsbook instead: the skill runs on Docsbook's machines, against your workspace, and draws on that project's balance.

run_docs_analyze({ request: "why is our quickstart getting impressions but no clicks?" })
// → { run_id: "run_…", state: "queued" }
 
get_agent_run({ run_id: "run_…" })
// → state, live progress, and once it finishes the report and what changed

One tool per skill — run_docs_analyze, run_docs_create, run_docs_manage, run_docs_automate — each running against your workspace with the full administrative toolset the skill was written for. The job takes minutes, so the call returns a run identifier rather than a result; read it with get_agent_run. run_docs_analyze changes nothing and works with a read-only token; the other three commit pages or settings and need a read-write one. See MCP Server.

Skills vs Subagents#

Catalog page#

Browse the full catalog at docsbook.io/skills, with install snippets and prompts for each skill.

What does running a skill cost?#

Reading the catalog costs nothing. find_skill is a discovery tool and is never metered, and a SKILL.md file fetched from GitHub and run by your own agent never touches Docsbook at all.

Two things do draw on the balance of the project they run against:

• The MCP tools a skill calls. A skill that reads analytics, searches your docs or commits a page spends what those calls spend — see what a call draws on.
• run_docs_*, because the run happens on Docsbook's machines and its own model budget. One call is one whole agent run: minutes of work, a report, and a run record.

Current amounts are on the Docsbook pricing page.

Next steps#

Ask a connected agent for find_skill({ query: "audit my docs" }) and run what it returns — no install, and nothing to configure first.

Related#

• MCP Server — how find_skill is exposed to agents, and how run_docs_* runs a skill for you.
• llms.txt — The other discovery surface for AI clients.
• docs-subagents — subagents for autonomous doc automation.
• markdown-lsp — open-source LSP parser for Markdown that powers local doc-graph search.