Server has 4 tools with explicit schemas and descriptions, but quality is uneven. All tools have descriptions (10-100 chars) and input schemas with typed properties. However, descriptions are terse (avg 95 chars vs baseline 194), lack WHEN/WHY context, and omit error recovery guidance. Parameter descriptions are minimal. Output schemas are undocumented, LLMs cannot predict response structure. No tool annotations (readOnlyHint/destructiveHint). Error handling returns generic text without actionable recovery steps. Naming is clear (verb_noun pattern) but tools lack composition guidance.

• Document output schema for each tool. E.g., start_review returns {session_id: string, title: string, questions: {bucket: string, items: string[]}[]}. Enables LLM to chain calls.
• Expand descriptions to 100-200 chars with WHEN/WHY. E.g., 'Add a fact to a session bucket (reality, people, strategy, operations, risk, metric). Call this after start_review to populate data before propose_plan generates a plan.'
• Add tool annotations: propose_plan gets readOnlyHint=true; add_fact and commit_next_steps get destructiveHint=true. Guides LLM retry logic.
• Enhance error messages with recovery steps. E.g., 'Session {session_id} not found. Call start_review() to create a new session, or check the session ID spelling.'
• Document parameter formats: session_id (UUID format, returned by start_review), bucket (enum: reality|people|strategy|operations|risk|metric), plan_json (JSON object with keys: goal, steps, checkpoint).

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

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