Claude controls your walking pad. When Claude works, you walk.
Four tools with clear verb-noun naming (start_walking, stop_walking, set_pace, walking_status). All have descriptions and input schemas. However, descriptions are terse (avg 80 chars vs baseline 194), parameter descriptions lack detail on constraints/ranges, and output schemas are undocumented. Error handling returns plain strings without recovery guidance. No tool annotations (readOnlyHint/destructiveHint). Composition is sound, each tool has one responsibility, but descriptions don't explain when to use each tool or dependencies between them.
Change the walking speed without stopping. Use this to speed up for big tasks or slow down as you wrap up.
Start the walking pad. CALL THIS whenever you begin working on a task. The user walks while you work. Always start at 1.0 km/h (safe default). Never set speed above 1.5 without asking the user first.
Stop the walking pad. Call this when you finish a task, the user asks to stop, or you're done working and want to give them a break.
Check the current walking pad status. Returns connection state, whether the belt is running, and current speed.
Output schemas undocumented. All four tools return plain strings; no structured response schema defined. LLMs cannot parse walking_status output (e.g., 'running at 1.5 km/h') into structured fields for downstream reasoning.
Parameter descriptions lack constraint details. 'speed' param in start_walking says '0.5-6.0' but doesn't explain why 1.0-1.5 is recommended or what happens if user requests 2.0+. Descriptions are 10-20 chars; baseline is 72 chars.
Error responses lack recovery guidance. Errors like 'Could not connect to walking pad: {e}. Is it powered on?' are user-facing, not LLM-actionable. No guidance on retryability, next steps, or alternative tools.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | <=2025-11-25 | v2 |
No tool annotations. Tools lack readOnlyHint (walking_status is read-only) and destructiveHint (stop_walking modifies state). Agents cannot infer safety/idempotency without explicit hints.
Tool descriptions don't explain composition. No guidance on when to call start_walking vs set_pace, or whether stop_walking is idempotent. Agents must infer from names alone.