Two tools with complete input schemas and detailed descriptions. memory_record has strong domain-specific guidance (authoritative policies, model auto-detection, JSON context). memory_search includes ranking logic and project_path rerank boost. However, output schemas are not documented in the source code, LLMs cannot see what fields to expect from responses. Parameter descriptions are good but lack explicit constraints (e.g., limit max=200 is mentioned in description but not in schema). No error handling guidance visible. No tool annotations (readOnlyHint/destructiveHint). Naming is clear (verb_noun pattern) but generic relative to domain.

• Document output schemas for both tools. E.g., memory_record returns {id, created_at, model, is_authoritative}; memory_search returns {results: [{id, title, type, tags, confidence, query_event_id}], total_count}.
• Add tool annotations: memory_record with destructiveHint=true (modifies state), memory_search with readOnlyHint=true (safe to retry).
• Formalize parameter constraints in JSON Schema: add enum for 'type' field, add maximum: 200 to 'limit', add pattern for 'tags' (comma-separated).
• Add error handling guidance: 'If no memories match, try broader tags or remove type filter. If dedup rejects a record, check for similar existing memories via memory_search.'
• Clarify pagination: 'Results are ranked by relevance. If limit is reached, use offset parameter to fetch next batch. Total count is returned to estimate remaining results.'