Five tools with complete input schemas and descriptions. Naming follows verb_noun pattern (audit_site, audit_page, get_recommendations, propose_fixes, preview_fix). Descriptions are substantive (100-200 chars), explaining WHAT and WHEN. However, output schemas are not documented in the source, LLMs cannot see what fields to expect from responses. Parameters are well-described with types and constraints (e.g., max_pages capped at 500, include_speed_heuristics boolean with default). Error handling is minimal, no recovery guidance or actionable error messages visible. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only. Missing: per-tool output schema documentation, error classification, and recovery patterns.

• Document output schemas for all five tools. For audit_site and audit_page, specify the Issue array structure (code, severity, message, evidence fields). For propose_fixes, document FixProposal structure. For preview_fix, document before/after snippet fields and regression_gate status.
• Add tool annotations: set readOnlyHint=true and idempotentHint=true on all five tools. This signals to agents that these tools are safe to call repeatedly without side effects.
• Enhance error handling in audit_site and audit_page. Return structured errors with classification (e.g., {error: 'unreachable', code: 'NETWORK_TIMEOUT', recovery: 'Check the URL and try again'}) instead of raw exceptions.
• Document the structure of complex input parameters. For propose_fixes, add a description of the Issue object schema (code, severity, message, evidence). For preview_fix, clarify the FixProposal structure expected.
• Add pagination support to audit_site. Accept an optional offset parameter and return a total_pages count so agents can crawl large sitemaps incrementally without hitting memory limits.