Single tool 'agent_handoff' has a complete input schema with proper types and enums, but the description is verbose (250+ chars) and mixes implementation details with usage guidance. Parameter descriptions are present but generic. No output schema is documented. The tool performs multiple distinct actions (write/read/resolve) under one name, violating single-responsibility principle. Error handling and recovery guidance are absent.

• Split agent_handoff into three separate tools: create_handoff_note (action=write), list_handoff_notes (action=read), resolve_handoff_note (action=resolve). Each tool should have a single, clear responsibility.
• Document the output schema for each action. E.g., create_handoff_note returns {note_id, created_at, from_agent, to, message}; list_handoff_notes returns {notes: [{note_id, from_agent, message, created_at}], total_count}.
• Trim tool description to 50-150 chars. Remove implementation details ('Notes persist locally (JSON)') and focus on user intent: 'Create, retrieve, or close handoff notes between agents for async task delegation.'
• Add conditional parameter descriptions. For 'message': 'Note body (required when action=write, ignored for read/resolve).' For 'note_id': 'Note ID (required when action=resolve, ignored for write/read).'
• Add error handling guidance in the tool description or as a separate error_handling section. E.g., 'If note_id not found, returns error with available note IDs. If 'to' is invalid, suggests similar agent names.'

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

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