Email Agent exhibits significant definition quality gaps. Tool names are verb-based (draft_email, send_email, etc.), which is good, but descriptions are inconsistent, some are adequate (draft_email: 127 chars), others are vague (update_status: 56 chars, below the 10-char minimum for meaningful context). Critical issue: input schemas are NOT visible in the provided source code. The backend/app.py excerpt shows Pydantic request models (ContactPayload, SendBody, etc.) but does NOT show explicit tool registration with JSON Schema input definitions. Parameters like 'config' and 'db' are passed as objects but lack type definitions and field descriptions. Error handling is absent, no recovery guidance, no categorization of retryable vs. fatal errors. Irreversible operations (send_email, process_scheduled_sends) lack confirmation or dry-run patterns. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) detected.

• Explicitly register all tools with MCP Tool objects including full JSON Schema inputSchema definitions. Map Pydantic models to JSON Schema with type, description, and required fields for every parameter.
• Add tool annotations: mark send_email and process_scheduled_sends with destructiveHint=true; mark draft_email with readOnlyHint=true; mark schedule_drafts with idempotentHint=true where applicable.
• Implement confirmation pattern for irreversible operations. Add a 'confirm_send' parameter or a separate 'confirm_email_send' tool that requires explicit approval before sending.
• Enhance error responses with recovery guidance. E.g., 'Failed to send email: Gmail API rate limit exceeded. Retry after 60 seconds or check Gmail MCP server status.'
• Document output schemas for all tools. Specify what fields are returned (e.g., draft_email returns {draft_id, subject, body, created_at}) so agents can chain calls correctly.