agentsclimarketplace

Plugin docs

Skill element22llc/e22-plugins/.claude/skills/plugin-docs

Reconcile the Zensical documentation site under docs/ with the plugin's source of truth — refresh the generated reference pages (skills, hooks, rules) from plugins/steer/, flag stale concept/workflow prose, run the docs validator, and optionally serve the site. Delegates deep accuracy review to the documentation-reviewer subagent. Repo-local dev helper for e22-plugins; does not ship.From its SKILL.md

Install
npx -y skills add element22llc/e22-plugins --skill plugin-docs

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.5 KB, as published. Nobody here has run it

/plugin-docs — keep the docs site in sync with the plugin

The documentation site (docs/, served via mise run docs:serve) must track the plugin's source of truth: the skills in plugins/steer/skills/, hooks in plugins/steer/hooks/hooks.json, and rules in plugins/steer/rules/. This skill reconciles the generated reference pages, flags stale prose, and runs the validator. It edits only docs/ (and never bumps plugin.json — docs ship nothing).

Steps

  1. Survey the source of truth.

    • Skills: ls plugins/steer/skills/ and read each SKILL.md frontmatter (name, description, when_to_use, argument-hint, user-invocable, disallowed-tools).
    • Hooks: read plugins/steer/hooks/hooks.json (events, matchers, scripts).
    • Rules: the numeric-prefixed files in plugins/steer/rules/ (first heading).
  2. Reconcile the generated reference pages so they match disk exactly:

    • docs/reference/skills.md — every skill present as /steer:<skill>; internal skills (user-invocable: false) listed under the internal section; read-only skills noted as such.
    • docs/reference/hooks.md — one row per hook, correct event + matcher.
    • docs/reference/configuration.md — the rule table matches the files on disk. Make the minimal edits needed; preserve hand-written prose and diagrams.
  3. Flag stale concept/workflow prose. For docs/concepts/* and docs/workflows/*, check claims against the current skills/rules. Propose edits for anything that contradicts the source of truth (e.g. a renamed mode, a changed argument-hint, a dropped lifecycle state). Do not invent content.

  4. (Optional) Deep review. For a thorough accuracy pass, dispatch the documentation-reviewer subagent (read-only) and fold its findings into the proposed edits.

  5. Validate. Run mise run docs:check (or uv run python scripts/validate_docs.py) and resolve every finding. If asked to preview, run mise run docs:serve; for a strict link/render check run mise run docs:build.

  6. Report what changed and what (if anything) still needs a human decision (e.g. a new skill that needs its own workflow page written from docs-templates/workflow.md).

Boundaries

  • Docs only. Edit under docs/ (and, when adding a page, mkdocs.yml nav). Never edit plugins/steer/** from this skill — if the docs and the plugin disagree, the plugin is the source of truth.
  • No changelog, no version bump. docs/ and .claude/ ship nothing.
  • Grounded. Every documented claim must trace to a SKILL.md, hooks.json, rule, or template. When unsure, leave a TODO and surface it rather than guess.
  • New pages start from docs-templates/ (outside docs/, since Zensical builds every file under docs_dir) and must be added to the mkdocs.yml nav (the validator fails on orphans).

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.