Two tools with complete input schemas and descriptions, but significant gaps in LLM-optimization and output documentation. Tool names lack action verbs (llm_to_json_cleaner, json_to_zod are descriptive but not verb-first). Descriptions are adequate (60-80 chars) but lack WHEN/WHY guidance. Parameters have types and descriptions, but lack constraints (enums, ranges, patterns). No documented output schemas. No error handling guidance. No security considerations despite handling untrusted LLM output.

• Rename tools to verb-first: 'repair_json_from_llm_output' and 'generate_zod_schema'. This helps LLMs parse intent from the name alone.
• Document output schemas explicitly. E.g. llm_to_json_cleaner returns {json: object, repaired: boolean, issues: string[]}. json_to_zod returns {schema: string, schemaName: string, inferredTypes: object}.
• Add enum constraints to strategy parameters. repairStrategy: ['conservative', 'aggressive', 'lenient']; dateStrategy: ['coerce', 'parse', 'string']. Prevents hallucinated values.
• Expand descriptions to 80-150 chars with WHEN guidance. E.g. 'Repair malformed JSON from LLM outputs (trailing commas, unescaped quotes, markdown fences). Use strict=true for production, false for lenient recovery. Returns cleaned JSON and repair log.'
• Add error recovery guidance in tool description. E.g. 'If repair fails, the input may be too corrupted. Try providing a simpler structure or use the raw parameter to see what failed.'