Risk describes an operation's impact. Execution role describes which operating-system privileges it needs. A read-only tool can need administrator privileges to read protected information; a user-role tool can still make a consequential change.
| Risk level | Typical effect | Default execution setting |
|---|---|---|
| Read-only | Inspects status, logs, or configuration; output can contain device/user details | Auto-execute |
| Low | Small reversible changes, such as restarting an application or flushing a cache | Human-in-the-loop |
| Medium | Service/configuration changes or operations that can interrupt work or need recovery | Human-in-the-loop |
| High | Disruptive, destructive, security-sensitive, or difficult-to-reverse operations | Human-in-the-loop |
Auto-execute lets the agent dispatch an available tool at that risk level without a technician approving each plan. Human-in-the-loop holds it for review. The setting applies to all tools with that risk classification, including enabled custom tools. There is no per-tool approval override. Use Tools to remove a particular operation from availability.
The initial device environment probe is automatic even when ordinary read-only tools require review. Qualifying download-only and local-installer staging steps can also run automatically because they do not execute the installer. The later install still requires review. Generated scripts follow their separate session Code mode rather than these four risk rows.
Use the default split as a clear starting policy: the agent can gather ordinary diagnostics while the technician reviews changes. If you choose stricter diagnostic approval, expect more interruptions. If you permit changes to auto-execute, those changes can occur as soon as the agent selects them.
Software and driver tools are unavailable while High is set to Auto-execute. Switching High to Auto-execute does not create unattended installation capability; it disables that managed suite until high-risk approval is restored.