Three tools with reasonable naming (verb_noun pattern) but critical gaps in descriptions and schemas. Tool descriptions are present but lack depth on WHEN to use them and prerequisites. Input schemas visible in code but lack parameter-level descriptions and type constraints. No output schemas documented. Error handling absent. Session-based state management suggests stateful design anti-pattern.

• Document output schemas for all three tools. Specify what fields are returned (e.g., submit_reasoning returns {ranked_personas: [{name, score, rationale}], session_id}). Include field types and descriptions.
• Expand parameter descriptions with constraints and discovery hints. For persona_pair, add: 'Array of exactly 2 persona names from the ranked options returned by submit_reasoning. E.g., ["Developer", "Security Expert"]. Invalid personas will be rejected.'
• Add WHEN/WHY context to tool descriptions. E.g., get_persona_guidance: 'Call this after submit_reasoning to receive structured guidance on how each selected persona should critique the reasoning. Returns critique prompts and focus areas for each persona.'
• Implement error handling with actionable recovery messages. E.g., 'Invalid session_id: session expired or not found. Call submit_reasoning() to start a new analysis session.' Include per-parameter validation errors.
• Refactor to stateless design: embed session context in request parameters rather than server-side storage. Or use immutable session tokens that encode state, eliminating server-side session tables.