Desktop browser: per-site approval cards ignore bypass-permissions mode — add a way to opt out

Status Open
Maintainer reply None cached
Activity 0 comments · opened Aug 6, 2026

Feedback / feature request

In the Claude Code desktop app, the built-in browser shows per-site approval cards (safety-classifier prompts for navigation and write actions on external pages) in every permission mode, including bypass permissions. There is no settings.json rule, trusted-origins list, or flag that pre-approves sites — the only option is clicking "Always allow" one domain at a time.

Why this is frustrating

Bypass permissions is an explicit, deliberate opt-in: "I understand the risks, stop prompting me." Having a second approval layer that this mode cannot turn off undercuts what the mode promises. It's my machine — if I've chosen to run the agent unattended with bypass permissions, I should be able to extend that choice to the built-in browser rather than babysitting a prompt per domain.

This particularly hurts autonomous/long-running sessions: the agent stalls on an approval card with nobody at the keyboard, which defeats the point of running in bypass mode.

Suggested options (any of these would help)

  • Honor bypass-permissions mode for the browser's per-site approval layer, or
  • A settings.json option for trusted origins (e.g. a browser.trustedOrigins allowlist or wildcard), or
  • An explicit global toggle ("don't ask for site approvals") with whatever scary warning is appropriate.

I understand the prompt-injection rationale for the extra layer — but it should be a default users can override, like every other permission in Claude Code, not a hard floor.

View original on GitHub ↗