All 5 tools have basic descriptions and input schemas, but quality is significantly below production standards. Tool names use Chinese characters without English verb prefixes, violating naming conventions. Descriptions are minimal (10-20 chars) and lack context on WHEN to use each tool or what to do with results. Parameters have type definitions but lack detailed descriptions explaining constraints, formats, or dependencies. No output schemas documented. Error handling returns None silently instead of actionable recovery guidance. No pagination support despite tools returning potentially large result sets (rankings, search results). Tools lack idempotency guarantees and composition support.

• Rename all tools to English verb_noun format: 'search_bilibili_videos', 'get_bilibili_ranking', 'get_bilibili_trending', 'get_bilibili_trending_keywords', 'get_video_details'. LLMs parse intent from names before reading descriptions.
• Expand descriptions to 50-200 chars with WHAT/WHEN/HOW structure. Example: 'Search Bilibili videos by keyword. Returns up to 20 results with title, BVID, author, and view count. Use this to find specific content; use get_bilibili_trending for popular videos without a query.'
• Document output schemas explicitly. Add a 'Returns' section to each tool docstring listing key fields (bvid, title, author, views, duration, etc.) and their types. Specify if results are paginated and how to fetch next page.
• Replace silent None returns with structured error objects. Return {'error': 'search_failed', 'reason': 'keyword too short', 'hint': 'Use 2+ characters'} instead of None. Let LLM decide whether to retry or ask user.
• Add pagination parameters (limit: 1-100 default 20, offset: 0+) to search and ranking tools. Cap results at 20-50 items per call. Document in description: 'Returns up to 20 results per page. Use offset to fetch next page.'

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

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