Two tools with basic schemas and descriptions, but significant gaps in parameter documentation and output structure. ask_httpx_assistant has a single string parameter with minimal context; get_eval_quality_summary has no parameters but lacks output schema documentation. Neither tool documents what fields the LLM should expect in responses, forcing the agent to parse unstructured text. Descriptions are present but generic (10 - 20 words), below the 50 - 200 char LLM-optimized baseline. No error handling guidance, no parameter constraints, no output schema declarations.

• Document output schemas for both tools. ask_httpx_assistant should return {answer: string, sources: [{source: string, relevance: number}]}. get_eval_quality_summary should return {run_id: string, timestamp: ISO8601, n_questions: int, metrics: {faithfulness: float, relevancy: float, groundedness: float, refusal_accuracy: float}}.
• Expand parameter descriptions. For ask_httpx_assistant, add: 'A natural-language question about httpx library features, API, or best practices. Questions should be specific (e.g., "How do I set a custom timeout?") rather than open-ended. Max 500 characters.'
• Add error handling with recovery guidance. Wrap httpx calls in try/except; return structured errors like {error: 'Service unavailable', retry_after: 30, suggestion: 'The RAG service is temporarily down. Try again in 30 seconds or check the service status.'}
• Enhance tool descriptions to 100 - 150 characters. ask_httpx_assistant: 'Query a RAG system trained on httpx source code and docs. Returns grounded answers with citations. Use when you need specific library details; for general Python HTTP advice, consider broader sources.' get_eval_quality_summary: 'Retrieve the latest automated quality metrics (faithfulness, relevancy, groundedness, refusal accuracy) from the httpx Q&A assistant evaluation harness. Use to assess assistant reliability.'
• Add optional parameters for filtering and pagination. ask_httpx_assistant: add optional max_sources (default 3) and answer_length (default 'medium', enum: ['brief', 'medium', 'detailed']). get_eval_quality_summary: add optional run_id (to fetch specific runs) and limit (default 1, max 10) to retrieve historical runs.