Critical gaps across all dimensions. Tool definitions are auto-generated (scripts/gen_tool_docs.py) with minimal descriptions (10-50 chars, well below 50-200 char baseline). All 34 tools lack parameter descriptions beyond 'User identifier (injected by server)' boilerplate. No output schemas documented. No error handling guidance. Naming is inconsistent: gmail_* vs inbox_* vs pdf_* vs webhook_* prefixes create cognitive load. Many tools combine multiple concerns (e.g., gmail_compose + gmail_send should be separate). No enums for constrained inputs (e.g., pdf_export format, webhook events). Security concern: user_id injected but no evidence of permission gates or audit logging. Composition is poor, tools like gmail_reply_to_thread require thread_id but no tool returns thread_id in a structured format for chaining.

• Expand all tool descriptions to 50-200 chars. Include WHAT (action), WHEN (use case), and WHAT IT RETURNS. Example: 'List inbox messages with pagination. Returns thread summaries (id, subject, sender, date, unread count). Use before gmail_get_thread to find specific conversations. Limit default 20, max 100.'
• Add descriptions to ALL parameters. For 'limit', specify range (1-100, default 20). For 'format' in pdf_export, use enum: ['pdf', 'png', 'jpg']. For 'events' in webhook_subscribe, list valid event types. For 'judgment' in inbox_save_curation, define allowed values (e.g., 'important', 'spam', 'archive').
• Document output schemas for all tools. Example for gmail_list_inbox: 'Returns {threads: [{id, subject, sender, date, unread_count, snippet}], total_count, next_cursor}'. This enables LLMs to chain tools correctly.
• Add confirmation/dry-run for irreversible operations. gmail_send should accept 'dry_run=true' to preview before sending. gmail_discard_draft should require explicit confirmation. Add error recovery guidance: 'If send fails, check gmail_status() and retry.'
• Consolidate tool namespaces. Rename inbox_* tools to gmail_inbox_* for consistency. Add batch variants: gmail_mark_threads_read (array of thread_ids) to reduce sequential calls. Ensure every response includes IDs needed for downstream tools (thread_id, draft_id, etc.).

Inferred effective spec: 2025-06-18+.

• Current-spec patterns present: structured tool output, Streamable HTTP