[FEATURE] Agent view dispatch should support --allow-dangerously-skip-permissions
Status Fixed / completed
Maintainer reply ✓ Yes — bogini
Workaround ✓ Mentioned in description ↑
Activity 10 comments · opened May 12, 2026 · closed May 13, 2026
💡 Likely answer: A maintainer (bogini, collaborator)
responded on this thread — see the highlighted reply below.
Preflight Checklist
- [x] I have searched existing requests and this feature hasn't been requested yet
- [x] This is a single feature request (not multiple features)
Problem Statement
There's no way to set the permission mode when dispatching a session from the agent view input. If I want to run a session with --allow-dangerously-skip-permissions, it's just not possible from agent view.
The current workarounds are:
- Set
defaultModein settings to apply globally, which is too broad - Set
permissionModein a subagent's frontmatter, which requires creating a subagent definition for what should be a one-off dispatch
Proposed Solution
Support --allow-dangerously-skip-permissions (or a shorthand) as a modifier when dispatching from the agent view input. Could be:
- A prefix/flag in the dispatch input itself
- A keybinding that toggles the permission mode for the next dispatch (e.g.,
Ctrl+Pto cycle through modes before hitting Enter) - A persistent setting within agent view that applies to all new dispatches from that session
The existing guard (requiring the user to have accepted bypass mode interactively at least once) should still apply.
Priority
Medium - Would be very helpful
Feature Category
Interactive mode (TUI) / Agent view
Additional Context
Related to #39889, which covers the same gap for mobile/desktop dispatch. This is specifically about the CLI claude agents TUI.
10 Comments
Found 1 possible duplicate issue:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
Not a duplicate. #39889 is about mobile/desktop dispatch ignoring the user's saved model and permission-mode preferences. This is about the CLI
claude agentsTUI having no mechanism at all to specify--allow-dangerously-skip-permissionswhen dispatching a session from the input.+1 on this one.
Also: once you're inside a session that was dispatched from agent view (without the flag), Shift+Tab cycling doesn't include bypass permissions either (since its never started that way), so there's no in-session recovery.
Anything started from agent view is effectively locked out from using --dangerously-skip-permissions, which greatly reduces it's usefulness.
Simplest fix imo would have it be persistent. If I have a session running with certain permissions, I would rarely want to change them and would almost always prefer to start again with the last permissions used (and can easily cycle to something different inside the session in the few occasions when I don't).
Same here, agents is pretty much unusable for me without this feature, I guess I'll stick to 6 terminal tabs for now.
+1. Adding another shape of this same problem:
If I launch the parent session with
claude --permission-mode bypassPermissionsin a separate terminaland then dispatch a child via
/bg, the child still prompts — consistently forevery Bash call, and intermittently for other tool uses (file edits etc.). The
parent's mode doesn't inherit; the child falls back to
defaultModefromsettings or the subagent's
permissionModefrontmatter, per the docs.That means even users who have already accepted bypass interactively (and
launched the parent with it) can't get a
/bg-dispatched session to rununattended without editing settings.json or creating a subagent definition for
what should be a one-off.
Two ways this could be solved, either would work for me:
bypassPermissions(and the user has accepted that mode previously),/bgdispatches inherit it by default.
in the dispatch input, or Ctrl+P to cycle modes before Enter.
(1) feels like the least-surprise behavior since the parent shell explicitly
opted in. Running multiple agent-view windows is the main use case the feature
seems designed for, and right now each
/bgdispatch needs the user back atthe keyboard for Bash calls in particular, which defeats the point.
Thanks for the request — landed.
claude agentsnow accepts--permission-mode,--model, and--effortto set defaults for every session you dispatch from that view:The active defaults appear in the footer below the dispatch input, and the header shows the model that dispatched sessions will use. The existing bypass/auto consent guard still applies — those modes are refused until you've accepted them once interactively.
Shipping in v2.1.142 (tomorrow).
@bogini,
claude agents --allow-dangerously-skip-permissionsdoes not work properly.bypass permissionsfor new chats, which is not the expected functionality for--allow-dangerously-skip-permissions. That is only expected for--dangeously-skip-permissions.bypass permissionsfor existing chats.@jonathanlaniado Found that you can run
claude agents --permission-mode bypassPermissionswhich allows you to dispatch agents with bypass permissions; FYI that any existing agents in the view will not inherit bypass permissions -- you will have to redispatch another agent to handle that task autonomously@goutham1220 Right,
--permission-mode bypassPermissionsworks for new sessions. The issue is specifically with--allow-dangerously-skip-permissions, which should only make bypass available in the Shift+Tab cycle without defaulting to it. Right now it behaves identically to--dangerously-skip-permissions, which defeats the purpose of having two separate flags.The existing session limitation is also a gap since there's no way to opt an already-running session into bypass after the fact.
Filed #59406 to track both.
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.