TinyClaw exposes 6 tools with basic schemas and minimal descriptions. All tools have empty input schemas (no parameters documented), violating the parameter description requirement. Descriptions are present but generic (10-50 chars), below the 50-200 char baseline for LLM-optimized tools. No output schemas documented. No error handling guidance. Tool names follow verb_noun convention (get_, set_, reset_, capture_) but lack context about when to use each. No parameter descriptions exist because schemas declare empty properties. Security-sensitive operations (reset_state, set_config) lack confirmation patterns or permission gates.

• Expand all tool descriptions to 80-150 characters. Include WHAT the tool does, WHEN to use it (vs similar tools), and what it returns. Example: 'get_occupancy_status: Returns current occupancy snapshot with zone counts and timestamps. Use this for real-time status checks; use capture_frame for fresh detection with latency.'
• Document output schemas for all tools. Define the JSON structure returned (e.g., occupancy_status returns {zones: [{id, count, timestamp}], total_occupants, last_update}). This enables LLMs to extract and chain results.
• Add parameter descriptions to set_config. Document cooldown_seconds (float, 0+, controls detection cooldown) and heartbeat_interval_seconds (int, 1+, controls health check frequency). Explain constraints and defaults.
• Implement confirmation pattern for reset_state. Add a 'confirm' boolean parameter (default false) or split into reset_state_preview (dry-run) and reset_state_execute (confirmed). Prevents accidental state loss.
• Add error handling guidance to tool descriptions. Document expected error cases (e.g., 'Returns error if device offline' for get_device_health) and recovery steps (e.g., 'Retry after 5 seconds or check device connectivity').

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

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