Two tools with basic verb-noun naming (search_papers, extract_info). Both have descriptions (60-80 chars, below the 194-char baseline), but lack depth on WHEN to use them and what they return. Input schemas present with types and defaults, but parameter descriptions are minimal (10-20 chars vs 72-char baseline). No output schema documented. No error handling guidance. search_papers has a WRITE risk but description does not state it modifies state. extract_info returns unstructured JSON string instead of structured object. No pagination, no result limits stated, no chaining IDs for downstream calls.

• Document output schemas explicitly. search_papers should return {paper_ids: [string], count: int, saved_to: string}. extract_info should return {paper_id: string, title: string, authors: [string], summary: string, pdf_url: string, published: string} or error with recovery hint.
• Expand tool descriptions to 100-200 chars. Add WHEN/WHY context: 'Search arXiv for papers on a topic and cache them locally. Call this first to discover papers, then extract_info to retrieve cached details.' State that search_papers modifies local filesystem.
• Add parameter constraints and ranges. 'max_results: integer, 1-100 (default 5). Limit to prevent excessive API calls.' 'paper_id: string, arXiv ID format (e.g., 2301.12345 or 2301.12345v2).'
• Implement error recovery guidance. Return structured error objects: {error: string, recovery: string, alternatives: [string]}. E.g., 'Paper not found. Try search_papers(topic=...) to discover and cache papers first.'
• Add pagination to extract_info. Accept optional limit and offset parameters. Return {papers: [...], total: int, limit: int, offset: int} to support large collections without context explosion.