Search Memory Command

Perform semantic search across indexed code using natural language queries.

Your Task

1. Extract the query from command arguments ($ARGUMENTS)

2. Call the query_memory MCP tool:

   • Pass the query string
   • Use default parameters unless specified:
     • max_results: 5 (can go up to 100)
     • similarity_threshold: 0.7 (range: 0.0-1.0)

3. Format results clearly and naturally:

   • Don't show raw JSON - present results in a readable format
   • For each result, show:
     • File path (make clickable if possible)
     • Similarity score (as percentage for clarity: 0.94 → "94% match")
     • Brief code snippet or description
   • Adapt verbosity based on query type:
     • Specific queries (class/function names): Show detailed snippets
     • Broad/conceptual queries: Show summaries to avoid verbosity
     • Use judgment based on result count
   • Show top 3-5 results with full details
   • Summarize remaining results if many matches

4. Offer to explain or dive deeper:

   • Ask if user wants more detail on any result
   • Offer to explain how the feature works
   • Suggest related searches if appropriate

Example Interaction

User: /search authentication middleware

Your Response:

Found 5 matches for "authentication middleware":

1. src/middleware/auth.py (94% match)
   JWT authentication middleware that validates tokens on protected routes.
   Extracts user from token and adds to request context.

   Key function: authenticate_request(request)

2. src/config/auth.py (87% match)
   Authentication configuration including JWT secret, token expiry,
   and middleware initialization.

3. tests/test_auth_middleware.py (78% match)
   Test suite for authentication middleware covering token validation,
   expired tokens, and invalid signatures.

Would you like me to explain how the authentication middleware works,
or dive deeper into any of these files?

Query Tips for Users

When suggesting queries to users, recommend:

• Be specific: "JWT authentication" > "auth"
• Use technical terms: "connection pooling" > "database stuff"
• Conceptual queries work well: "how is logging configured?"
• Pattern queries: "decorator usage", "context managers", "async functions"
• Cross-file queries: "where is UserModel used?", "all API endpoints"

Similarity Score Interpretation

• > 0.9 (90%): Highly relevant, likely exact match
• 0.8-0.9 (80-90%): Very relevant, strong semantic match
• 0.7-0.8 (70-80%): Relevant, conceptual match
• < 0.7 (<70%): Tangentially related (may not show by default)

Handling No Results

If total_results: 0:

• Inform user no matches found
• Suggest:
  • Try broader search terms
  • Lower similarity threshold (use --threshold flag)
  • Check if project is indexed (/status)
  • Try alternative phrasing

Advanced Usage

Users can customize searches:

• /search "error handling" --limit 10 (show up to 10 results)
• /search "UserModel" --threshold 0.8 (higher precision, fewer results)
• /search "async await" --limit 20 --threshold 0.6 (broad search)

Implementation Note

This command calls the existing query_memory MCP tool - no new code needed. The tool returns JSON with tiered results (full content, snippets, paths only).