Choosing and Running an Agent Sandbox Safely
A two-part playbook. Part one: write a threat model, match isolation to trust, score a shortlist and estimate cost. Part two: network, secret, state, approval, limit, log and incident controls.
27 USD
What's included:
- ✓A six-line threat model mapped to OWASP LLM risks
- ✓A six-question provider scorecard and a cost method
- ✓Run-time controls for egress, secrets, snapshots and approvals
- ✓An incident procedure and a quarterly retest routine