Three tools with partial schemas and descriptions. start_call has a detailed, well-structured input schema with enums and nested objects (calendarSlot), but descriptions lack LLM-optimization guidance on when/why to use it. wait_for_call_result and send_sms_confirmation have minimal schemas (single string params) and descriptions that are functional but lack actionable context. No output schemas documented for any tool. Descriptions range 50 - 200 chars but lack dependency hints and recovery guidance. No error handling patterns visible. Tool names follow verb_noun convention (start_, wait_for_, send_) but descriptions do not explain prerequisites or failure modes.

• Document output schemas for all three tools. start_call should return {callId: string, dashboardUrl: string}; wait_for_call_result should return {transcript: string, receipt: {outcome: 'goal_met'|'voicemail_left'|'failed', details: string}}; send_sms_confirmation should return {success: boolean, messageId: string, timestamp: string}.
• Add error handling guidance to descriptions. E.g., start_call: 'Returns error if phoneNumber is invalid E.164 format or business is unreachable. Retry with a different number or check network connectivity.'
• Separate orchestration logic from tool descriptions. Move 'ALWAYS tell user dashboard URL' and 'FIRST check calendar' into agent system prompts, not tool docs. Keep tool descriptions focused on WHAT the tool does, not HOW the agent should use it.
• Document callType-to-calendarSlot dependency explicitly. E.g., 'calendarSlot is required for callType=booking or callType=reschedule; omit for callType=inquiry or callType=complaint.'
• Add format constraints to parameter descriptions. E.g., wait_for_call_result: 'callId: string returned by start_call (UUID format)'; send_sms_confirmation: 'toNumber: E.164 format, e.g. +15551234567'.