Two tools with clear, actionable descriptions (194 and 156 chars respectively, within baseline 34-392 range). Both have proper input schemas with typed parameters and descriptions. However, output schemas are completely undocumented, LLMs cannot plan downstream operations or extract structured data. No error handling guidance, no pagination support despite search tool returning potentially large result sets, and no parameter constraints (enums, ranges) to prevent invalid inputs. Tool naming is adequate but generic ('banking_regulation' lacks action verb; 'banking_regulations_search' is verbose). Missing idempotency hints, rate-limit guidance, and recovery instructions.

• Document output schemas for both tools. Specify that banking_regulation returns {citation: string, text: string, effective_date: string, ...} and banking_regulations_search returns {results: [{citation, title, summary}], total_count, next_cursor}.
• Add pagination to banking_regulations_search: include limit (default 20, max 100), offset/cursor, and total_count in response.
• Add error handling guidance to descriptions: 'Returns 404 if citation not found. Try search_banking_regulations() with keywords. Returns 503 if eCFR API is temporarily unavailable, retry in 30 seconds.'
• Rename tools to action-verb form: get_banking_regulation and search_banking_regulations. Update descriptions to clarify: 'Retrieves the full text of a single banking regulation by citation' and 'Searches across all US banking regulations (12 CFR) by keyword.'
• Add parameter constraints: citation should document valid formats (e.g. '1026.19', '12 CFR 1026.19', 'Regulation Z'); query should specify minimum length (e.g. 3 chars) and note that very short queries may return thousands of results.