helpful in the future if we want more granularity for requesting escalated permissions: e.g when running in readonly sandbox, model can request to escalate to a sandbox that allows writes
<img width="485" height="600" alt="image" src="https://github.com/user-attachments/assets/4341740d-dd58-4a3e-b69a-33a3be0606c5" /> --------- Co-authored-by: jif-oai <jif@openai.com>