Auto mode: distinguish unpersuadable hard-category denials from approvable prompts

Status Open
Maintainer reply None cached
Activity 0 comments · opened Jul 24, 2026

Problem. In auto permission mode, some denials are hard-category classifier blocks (force-push, production deploy, destructive ops outside the working dir, secret/data exfiltration, self-modification of agent config) that will never clear no matter what the user says, while others are ordinary approvable prompts. The two look similar in the UI, so users spend cycles re-prompting and rephrasing against a hard block that has a 0%-by-design chance of clearing, before giving up and switching to bypassPermissions.

Request. When a denial is a hard-category block, say so in the denial message — e.g. a line like "This is a policy-category block (force-push) and cannot be approved in auto mode; switch to bypassPermissions to proceed." Ideally also surface the suggested next action (Shift+Tab to bypass).

Why. It turns a trial-and-error dead end into a one-step decision. Users keep the classifier's protection for everything else and only escalate deliberately when they hit a wall they now know is a wall.

View original on GitHub ↗