name: cache-docs description: When fetching library documentation, API references, or technical docs for project dependencies via WebFetch, automatically save a copy to the docs/ folder. Also triggers when user pastes documentation or discusses library APIs that should be cached for reference. Every non-stdlib import in the project should have corresponding reference docs cached.

Documentation Caching Skill

When you fetch or receive documentation for a library dependency, cache it locally.

When to Activate

• After any WebFetch of documentation sites (readthedocs, official docs, GitHub READMEs)
• When user pastes library documentation into the conversation
• When discussing a library's API and realizing docs should be cached
• When importing a new dependency that lacks cached documentation

Caching Process

1. Determine the library name from the URL or content context
2. Create the directory if it doesn't exist: docs/<library-name>/
3. Generate a slugified filename from the page title or topic
4. Write the file with YAML frontmatter:

---
source: <original URL>
fetched: <ISO 8601 date>
library: <library name>
summary: <one-line summary of what this doc covers>
---

# <Page Title>

<Content here>

1. Update the index at docs/_index.md with the new entry

File Organization

docs/
├── _index.md           # Auto-maintained index of all cached docs
├── pycrdt/
│   ├── quickstart.md
│   └── api-reference.md
├── stytch/
│   ├── magic-links.md
│   └── rbac.md
├── nicegui/
│   └── getting-started.md
└── sqlmodel/
    └── tutorial.md

Index Format

The _index.md should list all cached documentation:

# Cached Documentation Index

## pycrdt
- [Quickstart](pycrdt/quickstart.md) - Getting started with pycrdt CRDT library
- [API Reference](pycrdt/api-reference.md) - Full API documentation

## stytch
- [Magic Links](stytch/magic-links.md) - Email magic link authentication
- [RBAC](stytch/rbac.md) - Role-based access control

...

Important Notes

• Always preserve code examples from documentation
• Include version information when available
• If documentation is very long, summarize and link to source
• Cache critical information even if full doc doesn't fit