3ilm provides three well-named, read-only tools with clear descriptions and proper Zod schemas. Tool names follow verb_noun convention (search_, get_, list_). All three tools have descriptions (194 - 250 chars, within baseline 34 - 392 range). Input schemas are present and typed. However, output schemas are not formally documented, responses are constructed as markdown text without explicit field definitions. Parameter descriptions are adequate but could be more prescriptive about constraints. Error handling is basic (returns text messages, not structured recovery guidance). No tool annotations (readOnlyHint, etc.) despite all being read-only. Overall solid for a small, focused domain tool, but lacks production-grade output documentation and error classification.

• Add explicit output schemas to all three tools. Define fields like 'pattern', 'acceptance_rate', 'total_findings', 'examples_accepted', 'examples_rejected' with types and descriptions. Return structured JSON objects, not markdown.
• Add readOnlyHint: true to all tool definitions via server.tool() options. This signals to clients that these tools are safe to cache and retry.
• Enhance error responses with recovery guidance. E.g., 'Pattern "xyz" not found. Available patterns: [list]. Call list_patterns() to see all options.'
• Document parameter constraints in descriptions. E.g., 'query: 3 - 100 character keyword string. Examples: "oracle", "reentrancy". Shorter queries may return more results.'
• Add a discovery hint to list_patterns: 'Call this first to see all 12 vulnerability categories and their acceptance rates, then use search_vulnerabilities or get_pattern_details for details.'