PermissionRequest runs when Claude Code is about to show the user a permission dialog. The hook can intercept - auto-approving or auto-denying based on custom logic.

What it does

The hook sees the tool, arguments, and reason for the prompt. It can allow, deny, or pass through to the user. This is how you implement policy that's smarter than static permission rules - decisions based on the current branch, time of day, remote git state, or anything else you can script.

When to use it

Conditional auto-approval (only on feature branches, never on main).
Integrating with an external approval system for sensitive tasks.
Enforcing team policies that vary by project phase.
Building smarter "auto mode" behavior than the built-in classifier.

Gotchas

Automating approvals removes the human safety net. Log every decision.
Long-running hooks stall the session while the user waits.
A hook that auto-denies too aggressively will be worse than the default prompts.

Official docs: https://code.claude.com/docs/en/hooks.md#permissionrequest