// the find
tiangolo/library-skills
Library Agent Skills
A CLI (published for both Python via uvx and JS/TS via npx) that scans a project's dependencies, finds libraries that ship their own 'agent skill' docs, and symlinks those skills into .claude/skills or .agents/skills so coding agents follow current APIs instead of stale training-data patterns. It's for teams using AI coding assistants against libraries that have opted into shipping a SKILL.md, currently a short list led by FastAPI and Streamlit, both from the same author.
Symlink-based installation ties skill content directly to the installed library version, so `pip install --upgrade` or `npm update` updates the skill automatically without a separate registry to keep in sync. The tool tracks which symlinks it created and only touches those on repair/remove, leaving hand-authored skill directories alone. It's dogfooded: the library-skills package itself ships a SKILL.md installed through the same mechanism it distributes, and there are parallel Python/TS implementations with matching test suites rather than a single-language prototype.
The whole pitch depends on library authors adopting the convention and shipping a SKILL.md in their package — right now that's basically two projects from the same maintainer, so running this against a typical dependency tree will find nothing. Symlinks are a fragile default across platforms: on Windows they need Developer Mode or elevated permissions, and the documented fallback (`--copy`) quietly breaks the 'always up to date' promise since copies go stale the moment the library is upgraded. There's no published spec for what a skill must contain beyond following tiangolo's own SKILL.md shape, and no way to discover which packages have skills before you've already installed them — you find out by running the scanner.