| name | search-document |
| description | Search loaded document clauses using semantic and keyword search. Use when the user asks to "search", "find", or "look up" something in the documents. |
| argument-hint | <query> |
Search Document
Search loaded document clauses using semantic and keyword search across all documents.
Instructions
-
Run the search script:
uv run --with 'chromadb,sentence-transformers' \
"${CLAUDE_PLUGIN_ROOT}/skills/search-document/document-search.py" "$ARGUMENTS" [options]
-
Available options:
--section "4.*" — filter results by section number pattern--flag scope — filter by topic flag (scope, requirements, pricing, timeline, etc.)--source "filename" — filter by source document within the archive--full — show complete clause body instead of snippet--top N — number of results (default: 10)--json — output as JSON for further processing
-
Present results to the user with section numbers, source file, and relevance scores.
-
If the user asks a follow-up question about a specific clause, use
--section or --source to narrow results, or use --full to show the complete text.
Prerequisites
- Documents must be loaded first (use
/load-document if no data exists)
Constraints
- Always quote the query string to handle spaces
- Results span all loaded documents — note the source file for each result
- The search combines semantic (ChromaDB) and keyword (SQLite) results automatically
- Each search is logged to the audit trail
