Permission Prompt

What is a Permission Prompt

A permission prompt is a pause the harness inserts before a potentially risky or irreversible action, a destructive command, a force push, a payment, asking the human to explicitly approve it before the agent proceeds. It is the harness's default answer to genuine uncertainty about whether an action should be allowed to happen automatically.

This differs from a hook in an important way: a hook applies a fixed rule automatically, with no human involved at all, while a permission prompt exists specifically because the harness does not have a fixed rule confident enough to decide alone. It hands the decision to the one party who can actually weigh the context, the human.

Permission prompts are not meant to interrupt every action, only ones judged consequential enough to warrant a pause. Many setups reduce unnecessary interruptions by explicitly allowlisting routine, low-risk commands ahead of time, so the prompt only fires for the genuinely uncertain cases rather than for everything.