The Crab Engram defines 10 tools with explicit schemas and descriptions, but quality is inconsistent. All tools have descriptions (10-100+ chars) and input schemas with typed parameters. However, descriptions are often generic and lack actionable context for LLM selection. Parameter descriptions are minimal (e.g., 'Session ID for proper scoping' for session_id). No output schemas are documented. Error handling guidance is absent. Tool naming follows verb_noun convention (mem_*) but lacks clarity on when to use mem_search vs mem_context vs mem_capsules. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk profiles (WRITE, DESTRUCTIVE, READ_ONLY). Composition is reasonable, tools are single-purpose, but chaining is unclear (e.g., does mem_save return an ID for later mem_delete?).

• Add output schemas for all tools. Document mem_search return as {results: [{id, title, content, type, scope, created_at}], total_count, next_cursor}. This enables LLMs to chain tools correctly.
• Add tool annotations: mark mem_save, mem_delete, mem_merge_projects, mem_consolidate with destructiveHint=true or idempotentHint=true. Mark mem_search, mem_context, mem_stats, mem_timeline with readOnlyHint=true.
• Expand descriptions to 50-150 chars with selection context. E.g., mem_search: 'Find observations by semantic similarity. Use this to discover related learnings; use mem_context to retrieve a specific session's observations.' mem_boundaries: 'Retrieve anti-pattern rules and constraints for this project. Call before mem_save to validate observation types.'
• Add parameter constraints and format hints. E.g., session_id: 'UUID or session identifier (required for scoping observations to a conversation)'. topic_key: 'Optional alphanumeric key (max 50 chars) for organizing related observations into knowledge capsules.'
• Document error cases and recovery paths. E.g., mem_delete: 'Returns 404 if observation not found. Try mem_search to verify the observation exists. Returns 403 if user lacks admin permission.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry