Vesper has 5 tools with minimal parameter documentation and weak descriptions. Four tools (play, pause, next, previous) have empty input schemas and trivial descriptions (10-15 chars), violating the 20-char minimum for description scoring. The 'ask' tool has a single parameter with a description, but lacks output schema documentation. No error handling guidance, no parameter constraints, and no indication of what these tools return. Tool names are verb-based (good), but descriptions are too terse to guide LLM selection. Structured output is enabled, but the actual response structure is not documented anywhere in the visible code.

• Expand descriptions for play, pause, next, previous to 50-150 chars. Example: 'Resume playback of the current track in Cider Apple Music. Use when the user wants to continue listening after pausing.'
• Document output schemas for all tools. Specify what fields are returned (e.g., {status: string, current_track: {title, artist}, timestamp: ISO8601}).
• Add parameter descriptions and constraints to 'ask'. Example: 'Natural language music request (e.g., "play jazz", "skip to next"). Max 500 chars. Avoid system commands.'
• Add error handling guidance. Example: 'If text is empty, returns validation error. Retry with non-empty request. If Cider is not running, returns connection error, ensure Cider is open.'
• Consider adding tool annotations (destructiveHint for pause/next/previous, idempotentHint for play) to clarify side effects for LLM planning.