Five tools with clear verb-noun naming (hi_ask, hi_multiline, hi_choose, hi_confirm, hi_notify) and consistent descriptions (50-120 chars). All parameters have type definitions and descriptions. However, output schemas are completely undocumented, no response structure defined for any tool. Error handling guidance is absent. Tool descriptions lack WHEN/WHY context and don't explain return values. Parameters like 'agent_name' are optional but lack guidance on when to use them. No mention of idempotency, side effects, or recovery paths.

• Document output schema for each tool. E.g., hi_ask returns {response: string, cancelled: boolean, timestamp: string}. This enables LLMs to extract and chain results.
• Expand descriptions to 100-150 chars with WHEN/WHY context. E.g., 'Ask for single-line input (name, email, ID). Use instead of hi_multiline for brief responses. Returns user text or cancellation status.'
• Add error handling guidance: 'If user cancels, returns cancelled=true. If timeout (30s), returns error with retry guidance. LLM should ask user to try again or proceed without response.'
• Clarify 'agent_name' parameter: 'Optional. Identifies the calling agent in logs and UI. Helps humans track which agent asked which question. If omitted, defaults to 'unknown'.'
• For hi_choose, document choice format and limits: 'choices: array of 1-20 strings, each 1-100 chars. Returns selected choice(s) as string or array depending on multi_select flag.'

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

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