Signet has 4 tools with minimal schema documentation and weak descriptions. Tool names lack clear verb-noun structure ('do' is dangerously generic). Input schemas are present but sparse, only 'text' or 'query' or 'job_id' with basic string types, no enums, ranges, or validation hints. Descriptions are terse (10-60 chars) and lack WHEN/WHY context. No output schemas documented. Error handling is absent from tool definitions. The 'do' tool is particularly problematic: it executes 'multi-step requests' but provides no guidance on what requests are valid, what side effects occur, or how to recover from failures. Security-critical 'approve' tool lacks permission documentation.

• Rename 'do' to a specific verb: 'execute_request', 'run_agent_loop', or 'process_task'. Document what types of requests it accepts (e.g. 'multi-step planning requests only').
• Expand all descriptions to 100-200 chars. Include WHEN to use (e.g. 'Use capture to record voice input; use do for multi-step planning; use search to find past records'). Add prerequisites and expected output type.
• Document output schemas for each tool. E.g. 'capture returns {success: bool, capture_id: string, timestamp: ISO8601}'. Include all fields the agent needs for chaining.
• Add enum constraints to 'search' query parameter (e.g. search by date range, tag, speaker). Add range/length constraints to 'text' and 'query' (e.g. 'max 500 chars').
• For 'approve' tool, add description of required permissions (e.g. 'Requires approval_admin scope'), what job_id format is valid, and what happens on approval/rejection.

Inferred effective spec: 2025-06-18+.

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