Two tools with basic schemas and minimal descriptions. Both tools have input schemas with typed parameters, but descriptions are generic (under 50 chars) and lack context for LLM selection. No output schemas documented. No error handling guidance. No parameter descriptions beyond type. Missing idempotency hints for the write operation (apply_leave). Tool composition is reasonable but descriptions fail to explain WHEN to use each tool or what happens on failure.

• Expand tool descriptions to 50-200 chars. Example: 'get_leave_balance: Retrieve the current leave balance and name for an employee. Call this first to check available days before applying for leave. Returns {name, leave_balance}. If employee not found, try search_employees() with a partial name.'
• Add parameter descriptions with constraints. Example: 'emp_id (string, format: E001-E999): The employee ID. Must be a valid 4-character code starting with E.' 'days (integer, 1-365): Number of days to request. Must not exceed current leave_balance.'
• Document output schemas explicitly in tool descriptions or via JSON Schema. Example: 'Returns: {name: string, leave_balance: integer}' or 'Returns: {success: boolean, message: string, remaining_balance: integer}'.
• Add error recovery hints. Example: 'If employee not found, call search_employees() with a partial name. If insufficient balance, suggest checking leave_history first.'
• Mark apply_leave as destructive and idempotent (or not). Example: 'Destructive: modifies employee leave_balance. Idempotent: calling twice with same emp_id and days will fail on second call (insufficient balance). Supports dry_run parameter to preview without committing.'