RecallOrigin has 4 tools with basic verb-noun naming (memory_put, memory_context, memory_search, memory_forget). All tools have descriptions (10-60 chars), but most are terse and lack actionable context for LLM selection. Input schemas are present with typed parameters and descriptions, but lack enums for constrained values (e.g., 'mode' accepts 'remember', 'capture', 'feedback', 'govern' but is declared as free-form string). No output schemas documented. Error handling and recovery guidance absent. Idempotency support noted in memory_put but not enforced across other write operations.

• Convert 'mode' parameters to enum constraints: memory_put mode ∈ {remember, capture, feedback, govern}; memory_context mode ∈ {evidence, fast}. Remove from description, add to schema.
• Document output schemas for all tools. E.g., memory_search returns {results: [{claim_id: string, content: string, scope: string, timestamp: ISO8601}], total_count: int, next_cursor?: string}.
• Expand tool descriptions to 80-150 chars with WHEN/WHY guidance. E.g., memory_context: 'Retrieve memory claims and supporting evidence for a query. Use evidence mode for detailed reasoning chains; fast mode for quick lookups. Returns structured evidence pack with claim provenance.'
• Add error classification to tool descriptions. E.g., 'Returns 404 if scope does not exist (user-fixable: verify scope format); 500 if database unavailable (retryable).'
• Implement confirmation pattern for memory_forget: add optional 'dry_run' boolean parameter (default false) that returns what would be deleted without executing. Document in description.