Auto-Approve
Control how aggressively the agent auto-approves tool calls per session.
Location: Shield icon in the chat input (right side)
Every tool call has a risk level. The auto-approve shield controls the threshold. Anything at or below the selected level runs without asking. You're always in control of where that line sits.
Severity levels
| Level | Icon | Behavior |
|---|---|---|
| Off | Shield (dimmed) | Always ask for approval. No auto-approve. |
| None | Shield (gray) | Auto-approve permissions with none risk only. |
| Low | Shield (green) | Auto-approve low risk and below. |
| Medium | Shield (yellow) | Auto-approve medium risk and below. |
| High | Shield (red) | Auto-approve high risk and below. Almost everything runs automatically. |
The default is Low. Read operations and safe tools run without asking, anything that writes or executes asks first.
Per-session
Auto-approve is set per session. New sessions inherit the default. Change it mid-session and it applies immediately. No restart.
How it interacts with permissions
Workspace permission grants are separate from auto-approve. Grants are saved approvals for specific tools. Auto-approve is a blanket severity threshold. If auto-approve is set to a level that covers a tool's risk, the tool runs, regardless of whether a grant exists.
What's next
- Chat Controls: all the selectors in the chat input
- Permissions: saved tool approval grants
- Capabilities: capability-specific risk levels