Single-tool server with well-structured schema and comprehensive parameter documentation. The turn.wait tool has a properly defined JSON Schema with all parameters typed and described (in both English and Chinese). Descriptions are clear and actionable. However, no output schema is documented, and error handling guidance is minimal. The tool does one thing well (human-in-the-loop checkpoint) but lacks recovery patterns and error classification.

• Document the complete output schema: {user_response: string, timed_out: boolean, cancelled: boolean, timestamp: string}. Include examples for each response type.
• Add error response guidance: 'On timeout: retry with increased timeoutSeconds or ask the user for async input. On user cancellation: gracefully degrade or ask if the operation should proceed without human approval.'
• Implement toolAnnotations in the MCP schema: mark turn.wait with readOnlyHint=false (it reads state) and idempotentHint=false (repeated calls may receive different responses, and waiting is a side effect).
• Add a parameter description note for 'options': 'If provided, the user will see these as clickable buttons. Recommended for binary or bounded decisions (e.g., ["approve", "reject"]). If omitted, user may type free-form text.'
• Document the agentName parameter's purpose more explicitly: 'Used for audit logging and webhook events. Identifies which agent/system initiated the wait checkpoint. Recommended for multi-agent orchestration scenarios.'

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

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained