Doc Fetcher

The nuxt-docs and nuxt-ui-docs skills are the same script twice, with a different repository and topic map. This bundle is that script as a template, plus the procedure for pointing it at any library.

What's inside

Path	Purpose
`skills/doc-fetcher/SKILL.md`	The scaffolding procedure: find the docs in the repo, copy the template, fill `CONFIG`, write the generated skill's `SKILL.md`, verify.
`skills/doc-fetcher/fetch.template.py`	The fetcher: topic map, 24-hour cache, installed-version detection, `--list`, `--status`, `--force`, `--update-all`, `--version`. Standard library only.
`CLAUDE.md`	A pointer block to paste into your project's `CLAUDE.md`.

What a generated skill does

python3 .claude/skills/<lib>-docs/fetch.py <topic>          # fetch, or serve from cache
python3 .claude/skills/<lib>-docs/fetch.py <topic> --force  # bypass the cache
python3 .claude/skills/<lib>-docs/fetch.py --list           # topics
python3 .claude/skills/<lib>-docs/fetch.py --status         # what's cached, how stale
python3 .claude/skills/<lib>-docs/fetch.py --update-all     # refresh everything cached

Docs come straight from raw.githubusercontent.com at the branch you configure, and land under the skill's own cache/ for 24 hours. Set PACKAGE_NAME to the package.json dependency and every fetch is stamped with the version actually installed in the project, and the cache busts when that version changes, because a fresh doc read against a stale install is worse than no doc at all.

Install

unzip doc-fetcher.zip
mkdir -p .claude/skills
cp -R doc-fetcher/skills/. .claude/skills/
cat doc-fetcher/CLAUDE.md >> CLAUDE.md

Then ask Claude to generate a docs skill for the library you care about; the procedure is the skill.

The cache

CACHE_DIR is SCRIPT_DIR / "cache", so every generated skill caches inside its own directory and nowhere else. Delete the skill and the cache goes with it; delete the cache and the next fetch rebuilds it. Nothing here is worth committing.

This bundle declares its own cache path in the gitignore frontmatter key, so installing through pnpx @patrity/skills puts it in the project's managed .gitignore block. A skill you generate lives under a new directory, so add its path yourself:

.claude/skills/<lib>-docs/cache/

Configuration

The template needs none: it reads a public GitHub repo over curl and takes everything else from CONFIG. If a generated skill grows to need a value, say a token for a private docs repo, read it from .claude/.env rather than the repo root .env, so the project can give Claude a different credential than the app uses. Publishing that skill as a bundle, declare the variable with the env frontmatter key and let the tool write .claude/.env.example; never ship an example file of your own.

Requirements

python3 on PATH (standard library only, no pip installs) and curl.
Network access to raw.githubusercontent.com the first time each topic is fetched.

Companion bundles

nuxt and nuxt-ui: ready-made docs skills built this way. Install those directly rather than regenerating them.