Agent permissions

The harness's check on every tool call before it runs - allow it, ask you, or refuse it - set by a permission mode and by allow, ask and deny rules (deny wins). A sandbox limits what an action can reach once it runs.

Depth 1 · Foundation

You can explain where permissions are checked, how deny and allow rules combine, and how to set them up so the questions you get are worth reading.

Depths of understanding

1
Foundation
You can explain where permissions are checked, how deny and allow rules combine, and how to set them up so the questions you get are worth reading.
Read it
- Asking PermissionAI Agents Level 1: How Agents Work
Opens after
Agent harnessPermissions are checked by the harness, the one place every tool call passes through on its way to your computer.
2
Deeper
You can build it: a permission check that runs before every tool call (inside the folder, a denied command, always allowed, ask the person), and a refusal sent back to the model as a tool_result with is_error and the reason, so the conversation stays valid and the model can change course.
Read it
- Asking Before ActingAI Agents Level 2: Build Your Own Agent
Master depth 1 first