Ls Library

Library — Agentics Catalog & Discovery

One command to see everything. Catalogs all skills, commands, and agents across every location into a searchable YAML index.

Commands

Workflow by Command

list

1. Read ${CLAUDE_SKILL_DIR}/library.yaml 2. Group entries by source (location), then sub-group by category 3. Display using this format per group:

## <Source Name> (<count>)

### <Category>
- <name> (<type>) — <description>

First-use auto-populate: The catalog ships empty (template). If catalog: is empty or missing, automatically run the refresh workflow below to scan the user's sources and build the catalog for their machine, then continue with list. Don't make the user run /library refresh manually — just do it.

search

1. Read ${CLAUDE_SKILL_DIR}/library.yaml 2. Filter catalog entries where query appears in name, category, OR description (case-insensitive) 3. Display matches using same grouped format as list 4. Show match count: "Found N items matching '<query>'"

describe

1. Read ${CLAUDE_SKILL_DIR}/library.yaml 2. Find entry matching <name> (fuzzy — partial match OK) 3. If multiple matches, list them and ask user to clarify 4. Read the source file at the entry's path 5. Present: name, type, source, category, full path, then a concise summary (not the entire file)

refresh

This is the core workflow. Scan all registered sources and rebuild the catalog.

1. Read ${CLAUDE_SKILL_DIR}/library.yaml to get sources section 2. For each source, scan using its scan_pattern or scan paths 3. For each discovered file, extract metadata:

Extraction rules by file type:

Category assignment heuristics:

4. Write updated catalog back to library.yaml, preserving the sources section 5. Report: "Refreshed: N skills, M commands, K agents across L sources"

Important: Use Glob to discover files, Read to extract metadata. Do NOT use Bash for scanning.

register

1. Read ${CLAUDE_SKILL_DIR}/library.yaml 2. Add new source entry with the provided path:

   new-source:
     path: "<user-provided-path>"
     type: registered
     scan:
       - ".claude/skills/*/SKILL.md"
       - ".claude/commands/*.md"
       - ".claude/agents/*.md"
       - "ai/skills/**/*.md"

3. Derive source name from the directory name (e.g., /home/user/my-project becomes my-project) 4. Run refresh for just that source 5. Report what was found

Filtering Tips

When the user asks vague questions, map to search:

• "what marketing skills do I have?" → /library search marketing
• "do I have anything for testing?" → /library search testing
• "what can I use for planning?" → /library search planning
• "show me all polymathic agents" → /library search polymathic
• "what's in <project>?" → /library search <project> (matches source field)

Notes

• The YAML file is a read index — Claude reads skills from their source path, never copies them
• Refresh is idempotent — safe to run repeatedly
• Categories are best-effort heuristics — manually edit the YAML to fix misassignments
• When scanning a registered codebase, skip directories without .claude/ or ai/skills/
• Superpowers version directory is dynamic — scan for the latest version subfolder