[FEATURE] [Feature Request] Allow autonomous (no-prompt) sessions to run and be monitored from Android/mobile Cowork

Open 💬 0 comments Opened Jul 9, 2026 by servicecyclehq

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

Filing this as a feature request that ties together a few related gaps I've seen referenced in other open issues (#29319, #29565, #39889).

Use case: I run long coding tasks that don't need me in the seat — I kick them off, walk away, and review results when I'm back. For that to work, I need two things at once: (1) the session running with prompts disabled so it doesn't stall waiting on an approval I'm not there to give, and (2) the ability to open that same session on Android to check progress, nudge it, or fix something if it's gone sideways.

The problem: these two don't currently fit together.

A session started on desktop doesn't sync to the Android app in a way that lets me monitor, prompt, or course-correct it remotely.

Working around this by starting the session from Android first (so the mobile bridge is open) hits a wall: cloud-backed sessions don't expose a true "skip all approvals" mode. The mode dropdown only offers Accept Edits, Plan, and Auto — no bypass — confirmed in the permission-modes docs, which explicitly state bypassPermissions "is not available" for cloud sessions. Anything not covered by Auto's classifier just stalls waiting for a tap.

So today the only real options are: babysit approval prompts on my phone all day (defeats "fully autonomous"), or run unattended on desktop and lose the ability to check in or steer remotely from mobile.

Proposed Solution

Ask: either allow a bypass-permissions-equivalent for cloud/mobile sessions (even gated behind an explicit risk acknowledgment, similar to how local --dangerously-skip-permissions works), or make Auto mode robust enough to not need human taps for routine long-running coding work — paired with real desktop-to-mobile session sync so a session started on one device is actually controllable from the other, not just visible.

Cross-device continuity is clearly the intended Cowork story — this is the piece that would make "walk away and let it run" and "check in from my phone" both true at the same time.

Alternative Solutions

_No response_

Priority

Critical - Blocking my work

Feature Category

CLI commands and flags

Use Case Example

I'm working on a feature branch that needs a big refactor — say, migrating a set of API routes to a new auth pattern across 40+ files, plus updating the corresponding tests. It's mechanical enough that Claude can execute it end-to-end, but it takes 45+ minutes of tool calls (reading files, editing, running tests, fixing failures, iterating).

Step by step:

Before leaving my desk, I open Cowork on Android first (so the mobile-desktop bridge session exists), which spins up a linked session in desktop Cowork.

On desktop, I hand it the full task: "Migrate all routes in /api to the new auth middleware, update tests, make sure the suite passes." I want this to run unattended, so I try to switch the session to a no-prompt mode — this is where I hit the wall, since bypass isn't offered on cloud sessions and Auto still stalls on some actions.

I step away — errands, a meeting, whatever — for an hour or two.

While I'm out, I pull out my phone and open Cowork on Android to check in: has it finished the first 10 files? Did it hit a test failure it's stuck on? Is it waiting on an approval it can't get past?

If it's stalled on something (e.g., "this edit touches a protected path, approve?"), I want to just tap approve from my phone and let it keep going — without needing to be at my laptop.

If it's gone off track (e.g., editing files outside the intended scope), I want to send it a quick correction from mobile — "stop, don't touch the payment routes" — the same way I'd type into the desktop session.

By the time I'm back at my desk, the refactor is either done and I review the diff, or it's most of the way there and I finish it up locally.

The value is entirely in steps 4–6: being able to both let it run unattended and check/steer it from my phone in the gaps between other things I'm doing, rather than choosing one or the other.

Additional Context

_No response_

View original on GitHub ↗