MatrixClaw has 9 tools with mostly complete schemas and descriptions, but several critical gaps limit production readiness. All tools have input schemas with proper JSON Schema structure and type definitions. Descriptions are present but often lack actionable context (e.g., 'Create a one-time reminder' doesn't explain when to use it vs scheduled tasks). Parameter descriptions are sparse, many lack guidance on format, constraints, or dependencies. Error handling and recovery guidance are absent. Tool naming follows verb_noun convention well (create_*, plan_*, session_search). No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels. Output schemas are undocumented, agents cannot predict response structure. The memory tool combines 4 actions (list/add/replace/remove) in one tool, violating single-responsibility. Missing pagination on session_search despite accepting a limit parameter.

• Document output schemas for all 9 tools. Example: create_reminder should return {reminder_id: string, scheduled_at: string, text: string}. This enables agents to chain tools and extract IDs.
• Split memory tool into 4 separate tools: list_memory, add_memory, update_memory, delete_memory. Each should have a single, clear responsibility.
• Add tool annotations: mark session_search and plan_get with readOnlyHint=true; mark plan_clear with destructiveHint=true; mark plan_add_item with idempotentHint=true if applicable.
• Enhance descriptions with context: 'Create a one-time reminder that fires at a specific time. Use this for one-off alerts. For recurring tasks, use create_scheduled_ai_task instead.' Add 50-100 more characters of actionable guidance.
• Add parameter descriptions: cron_expr should say 'Standard 5-field cron expression (minute hour day month weekday)'; scope should explain 'global=shared across all users, user=private to caller, project=scoped to working_dir'.