# Ticket: Tool-Calling Policy

## Ticket Information

- **ID**: TICKET-026
- **Title**: Tool-Calling Policy
- **Type**: Documentation
- **Priority**: High
- **Status**: Backlog
- **Track**: LLM Infra, Safety/Memory
- **Milestone**: Milestone 2 - Voice Chat MVP
- **Created**: 2024-01-XX

## Description

Define tool-calling policy:
- When model is encouraged/forbidden to call tools
- High-impact actions need confirmation
- Tool permission matrix
- Escalation rules

## Acceptance Criteria

- [ ] Tool-calling policy documented
- [ ] High-impact actions identified
- [ ] Confirmation requirements defined
- [ ] Permission matrix created
- [ ] Policy integrated into prompts

## Technical Details

Policy should cover:
- Low-risk tools: always allowed (weather, time)
- Medium-risk: require confirmation (tasks, notes)
- High-risk: explicit user approval (email, calendar, smart home)
- Tool categories and permissions

## Dependencies

- TICKET-003 (safety constraints)
- TICKET-025 (system prompts)
- MCP tool definitions (from MCP tickets)

## Related Files

- `docs/TOOL_CALLING_POLICY.md` (to be created)

## Notes

Ties into confirmation flows (TICKET-045). Should be defined before tool implementation.