CKS MCP has 8 tools with schemas and descriptions present, but quality is inconsistent. Tool names lack clear action verbs (e.g., 'agent_status' vs 'get_agent_status', 'arbitrate_inference_conflict' is vague). Descriptions are present but often domain-specific and lack LLM-friendly guidance on WHEN to use each tool or WHAT happens (especially for WRITE operations). Parameter descriptions exist but are minimal (avg ~30 chars, well below 72-char baseline). No output schemas documented. Error handling is absent, no recovery guidance, no categorization of retryable vs fatal errors. Security concerns: ai_chat accepts 'messages' array and 'model' override without validation; approve_resolution accepts arbitrary 'resolution' objects. No tool annotations (readOnlyHint/destructiveHint) despite 4 WRITE tools. Composition is weak: ai_chat and arbitrate_inference_conflict appear to do overlapping reasoning work; no clear separation of concerns.

• Rename tools to verb_noun pattern: agent_status → get_agent_status, arbitrate_inference_conflict → resolve_inference_conflict, approve_resolution → apply_resolution. This enables LLMs to parse intent from names alone.
• Document output schemas for all tools. E.g., ai_chat should return {session_id: string, messages: array, final_conclusion: object, iteration_count: int}. Include all fields downstream tools need.
• Add error handling guidance to descriptions. E.g., ai_chat: 'If session_id is invalid, returns error with available session IDs. Call create_branch to start a new session.' This enables agent self-correction.
• Add tool annotations: mark ai_chat, approve_resolution, arbitrate_inference_conflict, create_branch, close_session with destructiveHint=true. Mark idempotent operations (e.g., get_agent_status) with idempotentHint=true. This guides agent retry logic.
• Expand parameter descriptions to 50-100 chars with context. E.g., session_id: 'Unique session identifier returned by create_branch or ai_chat. Pins reasoning state across multiple calls. Invalid IDs return 400 with available sessions.' This prevents LLM guessing.