Checklist
Agent Safety Checklist
Permission, data, verification, logging, and rollback questions before an agent acts.
Core concept
CurrentNext review due 2026-10-21Content version 0.30.0
- Format
- Checklist
- Level
- Intermediate
- Audience
- Developer, Operator
- Owner
- Project42 Editorial
- Review cadence
- Every 90 days
- Prerequisites
- Basic familiarity with AI agents and tool use
01
Before granting tools
Limit tools and credentials to the task.
Separate read, reversible write, external communication, and destructive permissions.
Validate untrusted input and tool output.
Record material actions.
Require human approval and rollback for high-consequence changes.