From grimoire
Semantic search across indexed knowledge bases using Qdrant vector similarity. Returns ranked results with source metadata for citation. Used by librarian agent for research queries.
How this skill is triggered — by the user, by Claude, or both
Slash command
/grimoire:librarian-library-searchThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Perform semantic search across indexed library content via a local Qdrant
Perform semantic search across indexed library content via a local Qdrant vector database.
This skill is the librarian agent's primary method for consulting local knowledge bases. Use it before falling back to grep-based file search.
Semantic search finds relevant content even when exact keywords don't match — a query for "flash loan price manipulation" will surface content about oracle attacks and sandwich attacks that grep would miss.
The search script is at skills/librarian-library-search/scripts/search.py
inside the plugin directory.
uv run /path/to/grimoire/skills/librarian-library-search/scripts/search.py \
"your natural language query here"
| Flag | Default | Description |
|---|---|---|
--limit | 5 | Maximum number of results |
--library | (all) | Filter to a specific library name |
# Broad search across all libraries
uv run .../search.py "reentrancy vulnerability in pull-payment pattern"
# Scoped to a single library
uv run .../search.py "access control bypass" --library smart-contract-vulnerabilities
# More results
uv run .../search.py "ERC-4626 share inflation" --limit 10
The script prints a JSON array to stdout. Each element has:
{
"score": 0.82,
"content": "[library-name] path/to/file\n\n...chunk text...",
"metadata": {
"library": "smart-contract-vulnerabilities",
"file": "vulnerabilities/reentrancy.md",
"chunk_idx": 2,
"source_url": "[email protected]:kadenzipfel/smart-contract-vulnerabilities.git"
}
}
librarian-index and fall back to grep for now.librarian-index
to build or rebuild the index.sentence-transformers/all-MiniLM-L6-v2). Do not change --embedding_model
unless you also re-indexed with that model.npx claudepluginhub joranhonig/grimoireBuilds or rebuilds a semantic search index over registered libraries using local vector embeddings (FastEmbed) and Qdrant. Useful for making library content searchable by meaning, not just exact text matches.
Queries local document knowledge base using semantic vector search and hybrid retrieval for AI-powered answers with citations via Claude Haiku.
Provides patterns and Python templates for similarity search with vector databases, including metrics, indexes, and Pinecone implementation. Use for semantic search, RAG, recommendations, and scaling.