Agent Boundary Quickscan 🦋

This free scan runs entirely in your browser. Nothing is submitted or stored. It does not certify safety or compliance. It identifies durable-state surfaces that may deserve a closer look.

1. Does the agent keep persistent memory or user profiles across sessions?

2. Can the agent take external actions such as sending, purchasing, posting, or changing records?

3. Are some actions gated by human approval?

4. Can an approved workflow pause, serialize, resume, retry, or continue later?

5. Can approval ever become reusable/sticky for later actions?

6. Does the system summarize, infer, consolidate, or rewrite memory?

7. Can the same persistent agent move between models/providers/runtimes?

8. Are there multiple users, tenants, threads, or long-lived work contexts?

9. Can internal state become public output or publication?

10. Is authority-bearing serialized state authenticated before restoration?

11. Can important memories/decisions be traced to their source or provenance?

12. Do regression tests prove one decision/container cannot silently authorize another?