What This Skill Does
Scans project documentation directories, extracts metadata (titles, descriptions), categorizes files by type, and generates standards-compliant llms.txt (index with links) and optionally llms-full.txt (index with inline content) files. Uses a Python script for deterministic work (scanning, extracting, categorizing) with Claude reviewing and improving the output.
When to Use
- Project needs an LLM-friendly documentation index
- User asks for "llms.txt", "LLM documentation", "AI-friendly docs"
- Publishing a docs site and want AI discoverability
- After major documentation updates (MeowKit Phase 6 — Reflect)
- Explicit invocation:
/mk:llms [path] [--full]
Core Capabilities
| Capability | Detail |
|---|---|
| File scanning | Recursively finds .md files in a directory |
| Metadata extraction | Extracts H1 title from each doc (falls back to filename) |
| Auto-categorization | Groups files: Getting Started, API Reference, Guides, Architecture, Configuration, Optional |
| Spec compliance | Strict llmstxt.org spec: H1, blockquote, H2 sections, one-sentence descriptions |
--full mode | Generates llms-full.txt with full document content inlined |
| Script-first | Python script handles all deterministic work; Claude reviews and improves |
Arguments
| Argument | Effect |
|---|---|
| (no args) | Scan ./docs directory, output to project root |
path | Scan specific directory |
--full | Also generate llms-full.txt with inline document content |
--output path | Custom output location (default: project root) |
--url base | Base URL prefix for links (e.g., https://example.com/docs) |
Workflow
5-Step Process
Example Prompt
/mk:llms ./docs --full --url https://myproject.dev/docs I need my project documentation to be discoverable by AI assistants. Generate both llms.txt and llms-full.txt for my docs directory with proper categorization and one-sentence descriptions per file.