lo-eyes has 5 tools with clear verb-noun naming (eyes_screenshot, eyes_scan, eyes_audit, eyes_compare, eyes_devices). All tools have descriptions (10-60 chars, mostly adequate). Input schemas are present with type definitions and parameter descriptions. However, descriptions are terse (avg 35 chars vs baseline 194 chars), lack WHEN/WHY context, and omit actionable error guidance. Output schemas are not documented. No error handling guidance visible. Tool composition is sound (each does one thing), but descriptions don't explain prerequisites, dependencies, or recovery paths. Baseline compliance: 100% have descriptions, but only ~40% meet depth standards (50-200 chars, LLM-optimized).

• Expand tool descriptions to 100-200 chars. Add WHEN/WHY context: 'Use eyes_screenshot to capture a single viewport at a specific device size. Use eyes_scan if you need the full page auto-chunked into readable sections. Returns base64-encoded PNG image.'
• Document output schemas for all tools. E.g., eyes_screenshot returns {image: string (base64), device: string, url: string, timestamp: string}. eyes_audit returns {issues: [{type, severity, detail}], summary: string}.
• Add error handling guidance to descriptions. E.g., 'If the URL is unreachable, check BASE_URL env var or pass a full http(s):// URL. If Playwright times out, the page may be slow or require authentication.'
• List enum values in parameter descriptions. Change 'device preset: iphone-se, iphone-14, ipad, laptop, desktop' to a formal enum constraint in the schema, and repeat the list in the description for LLM clarity.
• Add a note to eyes_compare: 'This tool modifies state when save_baseline=true. Use with caution, baseline images are stored in BASELINE_DIR and used for future regression tests.'