Single tool with minimal definition quality. Tool name 'search_vais' is verb-noun compliant but lacks clarity (VAIS is domain-specific jargon). Description is 44 chars, below the 50-200 char LLM-optimized range. Input schema present but parameter description is generic ('The search query to execute'). No output schema documented. Error handling returns unstructured dict with 'error' and 'status_code' keys, no recovery guidance. No parameter constraints (min/max length, format). Tool does not follow pattern:tool-description or pattern:response-shaper standards.

• Expand tool description to 80-150 chars: 'Search private data indexed in Vertex AI Search. Returns up to 5 results with extractive segments. Use when user asks questions about internal documents or knowledge bases.'
• Document output schema in code comments or docstring: 'Returns {"response": [{"title": str, "snippet": str, "uri": str, "extractive_segments": [str]}]} on success, {"error": str} on failure.'
• Add parameter constraints: 'search_query (str, 1-500 chars): The natural language search query. E.g., "How do I reset my password?"'
• Enhance error messages with recovery guidance: Instead of returning {"error": str, "status_code": 500}, return {"error": str, "recovery": "Check VAIS_ENGINE_ID and VAIS_LOCATION are set correctly"} for configuration errors.
• Add input validation: Reject empty queries before calling VAIS API. Return {"error": "Search query cannot be empty", "recovery": "Provide a non-empty search query"} with HTTP 400.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry