index-library

--- name: index-library description: "Index a library's documentation into a queryable local knowledge base. Invoke with /index-library to fetch and index API docs, guides, and patterns. Prevents Claude from inventing APIs or missing features. After indexing, the library knowledge is available to /research and /solution-design skills. Check if already indexed before fetching (skip unless --force)." allowed-tools: - Bash - WebFetch - WebSearch - Read metadata: token-cost-tier: context-heavy token-cost-estimate: 10000 --- # /index-library — Library Documentation Indexer ## Usage ``` /index-library Index from official docs URL /index-library Search for official docs then index /index-library --force Re-index even if already indexed /index-library --list List all currently indexed libraries /index-library --search Search across all indexed libraries ``` --- ## Step 0 — Check Existing Index ```bash ls .claude/lib-index/ 2>/dev/null || echo "no index yet" ``` If `--list`: show all indexed libraries with their index dates and entry counts. Exit. If `--search `: ```bash grep -rl "" .claude/lib-index/ 2>/dev/null | head -10 grep -r "" .claude/lib-index/ --include="*.md" -l 2>/dev/null | head -10 ``` Show matching sections. Exit. **Dedup check:** Before indexing, derive a slug from the library name/URL: - `https://fastapi.tiangolo.com` → `fastapi` - `https://docs.sqlalchemy.org` → `sqlalchemy` - `redis` → `redis` Check if `.claude/lib-index//` exists: ```bash ls ".claude/lib-index//" 2>/dev/null && echo "EXISTS" ``` If exists and no `--force`: ``` Library '' is already indexed. Last indexed: Entries: To re-index: /index-library --force To search: /index-library --search ``` Exit. --- ## Step 1 — Resolve Entry Point If URL provided → use directly. If name provided (e.g., `fastapi`, `sqlalchemy`, `redis`): ``` Search: " official documentation site:docs..io OR .tiangolo.com OR .io/docs" ``` Pick the first official docs URL. Announce: ``` Indexing: Entry point: Target: .claude/lib-index// Estimated: ~ tokens ``` --- ## Step 2 — Fetch Entry Point ``` WebFetch: ``` Extract: 1. Page title → library name + version 2. Navigation links → identify key sections to fetch 3. Short description (first 2-3 sentences) **Prioritize sections for fetching (max 10 pages total):** 1. Getting Started / Quick Start / Installation 2. API Reference (if exists as a single page — fetch it; if it's paginated, skip and note it) 3. Key concepts / Core concepts / Overview 4. Configuration reference 5. Examples / Cookbook / Recipes Skip: changelog, release notes, contributing guides, translation pages. --- ## Step 3 — Fetch Key Pages For each priority page (max 10, depth capped): ``` WebFetch: ``` Extract per page: - Section name - Key classes/functions/methods with signatures - Configuration options - Code examples (keep, they're high-value) - Important caveats or version notes - Links to sub-pages (if relevant API reference) **Token budget per page:** Extract ~500 chars of key content. Don't dump full page. --- ## Step 4 — Build Index Create `.claude/lib-index//`: **File: `.claude/lib-index//index.md`** (main knowledge file, ~2000 chars) ```markdown # Index Version: Indexed: Source: Pages fetched: ## Quick Reference ### Installation `````` ### Core Concept <1-2 sentence explanation of what this library does> ### Key Classes / Functions | Name | Purpose | Signature | |------|---------|-----------| | `ClassName` | what it does | `ClassName(param1, param2)` | | `function_name()` | what it does | `function_name(arg, kwarg=default) -> ReturnType` | ### Configuration Options | Option | Type | Default | Description | |--------|------|---------|-------------| ### Common Patterns #### Pattern 1: `````` #### Pattern 2: ... ### Gotchas & Common Mistakes - - ### What's NOT in this index - - API reference: — too large to index, fetch specific methods on demand ``` **File: `.claude/lib-index//index-meta.json`** ```json { "library": "", "version": "", "slug": "", "entry_url": "", "indexed_at": "", "pages_fetched": , "entry_count": } ``` --- ## Step 5 — Report ``` Library indexed: v Location: .claude/lib-index// Pages fetched: Key entries: classes/functions, config options, examples Quick search: /index-library --search Use in research: /research "" (will auto-load this index) Use in code review: agent will check this index before suggesting APIs ``` --- ## Integration with Other Skills The `research` skill automatically checks `.claude/lib-index/` at Step 1. The `code-review` skill can reference indexed libraries to verify API usage. The `solution-design` skill can use indexed libraries to enumerate real options. **Claude should check the index before suggesting any API from an indexed library:** ```bash grep -r "" .claude/lib-index/ 2>/dev/null | head -5 ``` If not found → do not assume it exists. Fetch the specific docs page first. --- ## Limits - Max 10 pages per library (prevents runaway fetching) - Max entry point depth: 2 levels (don't follow links of links) - Index file max size: ~3000 chars (keep it queryable, not a raw dump) - If API reference is paginated (100+ pages): index only the overview, note `"full API reference not indexed — fetch on demand"`