HermitAgent exposes 4 tools with basic schemas and descriptions. All tools have names starting with action verbs (run_, reply_, check_, cancel_) and descriptions present. However, descriptions are minimal (10-40 chars), parameter descriptions lack detail, and output schemas are not documented. No error recovery guidance, no parameter constraints (enums), and no indication of idempotency or side effects. Tool composition is reasonable (single responsibility), but descriptions do not meet the 50-200 char LLM-optimized baseline. Schemas are present but sparse, parameters lack type constraints and validation rules.

• Expand tool descriptions to 50-200 chars. Example: 'run_task: Launch a background task with optional model and strategy. Returns task_id for polling. Use check_task() to monitor progress.'
• Document all parameter descriptions with format, range, and valid values. Example: 'strategy: Execution strategy, "single" (one-shot) or "auto" (multi-turn reasoning). Default: single.'
• Add output schema documentation for each tool. Example: 'run_task returns {status: string, task_id: string, error?: string}'. Include all fields the LLM might need for chaining.
• Enhance error messages with recovery guidance. Example: 'Task not found. Verify task_id with check_task() or list recent tasks.' instead of bare 'task not found'.
• Add tool annotations via fastmcp decorators: @mcp.tool(readOnlyHint=True) for check_task, destructiveHint=True for run_task/cancel_task, idempotentHint=True for check_task.