deepcheck exposes 4 tools with minimal documentation. Tool descriptions are present but generic (10-60 chars, below the 50-200 char LLM-optimized baseline). Input schemas are visible but sparse: 'scan' and 'analyze_image' accept only a single string parameter with no enum/format constraints; 'extract_c2pa' and 'validate_c2pa' accept bytes with minimal guidance. No output schemas are documented anywhere in the codebase. Parameter descriptions are terse (e.g., 'path to image file or directory' lacks format/constraint details). No error handling guidance, no recovery patterns, no idempotent/destructive hints. Tool names are action-oriented (scan, analyze, extract, validate) but descriptions fail to explain WHEN to use each tool vs. the others, LLMs cannot distinguish analyze_image from scan. No composition guidance for multi-step workflows.

• Expand tool descriptions to 50-200 chars. For scan: 'Scan a file or directory for synthetic-media indicators and C2PA provenance. Returns findings with severity scores.' For analyze_image: 'Analyze a single JPEG/PNG image for synthesis artifacts and C2PA manifest validity. Returns detailed verdict and signal breakdown.'
• Document output schemas for all tools. E.g., scan returns {findings: [{id, severity, path}], score: number}; analyze_image returns {verdict, synthetic_score, c2pa: {present, valid, assertions}, signals: []}.
• Add enum constraints to parameters. extract_c2pa: fmt should be enum ['jpeg', 'png']. validate_c2pa: blob should document expected JUMBF structure.
• Add error handling guidance. E.g., 'If file not found, check path spelling. If format unsupported, convert to JPEG/PNG first. If C2PA validation fails, the manifest may be corrupted.'
• Document tool relationships. Add a note: 'Use scan for bulk directory analysis. Use analyze_image for detailed single-image inspection. Use extract_c2pa + validate_c2pa for low-level C2PA debugging.'