Four tools with minimal descriptions (10-20 chars each) and empty input schemas for 3 of 4 tools. No parameter descriptions, no output schemas documented, no error handling guidance. Naming is verb-based but descriptions lack context for LLM selection. avatar_speak has a single required parameter with description, but others expose no parameters at all. No indication of idempotency, side effects, or recovery paths. Fails baseline: 94% of A+ tools have docstrings; 100% have param descriptions.

• Expand all tool descriptions to 50 - 200 chars. Example: 'avatar_think' → 'Signal the avatar to enter thinking state. Use this when the LLM is reasoning or processing a complex request. The avatar will display a thinking animation and pause speech.'
• Add output schema documentation. Example: avatar_speak should return {type: 'object', properties: {status: {type: 'string', enum: ['speaking', 'queued', 'failed']}, duration_ms: {type: 'number'}}}.
• For avatar_speak, add optional parameters: volume (0 - 100, default 80), rate (0.5 - 2.0, default 1.0), language (enum: en, es, fr). Document each with type and constraints.
• Add error handling: 'If TTS engine unavailable, avatar_speak returns {status: "failed", reason: "tts_unavailable"}. Retry after checking TTS configuration or fall back to avatar_listen for user input.'
• Add tool annotations to schema: readOnlyHint: true for avatar_think/listen/idle; document idempotency for avatar_speak (calling twice with same text within 1 second speaks once).