Permission Mode
Mode Izin
The permission-gating slice of an agent mode: which tool calls trigger a request and which run automatically.
A permission mode is the part of an agent mode that decides which tool calls need a permission request and which run automatically. A harness might offer read-only, where everything that writes is gated; a normal mode, where edits go through but deletions ask; and a permissive mode, where the agent acts without stopping.
The modes are a risk dial, and the useful question is which one fits the task rather than which one is fastest. A read-only mode is right for a codebase you do not understand yet. A permissive mode is right for a throwaway directory you are about to delete. The wrong choice is the usual one: permissive on a repository with history, or read-only on a task whose whole point was to change something.
Narrow modes are not slower in a meaningful way, because the agent still works. What changes is the blast radius when it is wrong. A permission mode is a statement about what you are willing to have happen without you watching, which is a decision about your own risk tolerance rather than about the tool.
Related terms
Permission Request
What the harness shows the user before executing a tool call that is not pre-approved.
Agent Mode
A preset that bundles permissions and behaviour, like a profile for the agent.
Guardrails
Checks that stop or constrain an action before it happens.
Human in the Loop
A person who reviews, approves, or corrects the agent's work at key points.
Sandbox
An isolated environment the agent runs inside: container, VM, or restricted shell. It limits the blast radius.
Want this working inside your team?
It starts with one conversation about your workflow.