Pushary
Blog
Guides

Claude Cowork stuck waiting for permission? Stop babysitting it

Cowork pauses on questions and approvals, and if you stepped away it just sits there. Route those moments to your phone so tasks finish while you are gone.

AG
Aadil Ghani
Founder, Pushary
Jul 10, 20264 min read
Share

When Claude Cowork stops mid-task, it is almost always waiting on you: a permission prompt in "Ask before acting" mode, a clarifying question, or a go-ahead before something irreversible. If you are at the screen, you tap and it continues. If you walked away, the task sits frozen exactly where you left it. The fix is to route those moments to your phone, so a question costs you ten seconds from a lock screen instead of an afternoon of not knowing.

Key takeaways

  • Cowork stalls are usually a waiting question, not a crash. The task resumes the moment someone answers.
  • Anthropic's own notifications cover part of this: the Claude app can push when a task finishes or needs a go-ahead, depending on your plan and rollout.
  • A Pushary connector puts Cowork's questions on your phone as push notifications you answer from the lock screen, alongside every other agent you run, with a record of what was decided.

Why Cowork freezes at step two

Cowork runs long, multi-step tasks, and it is built to pause rather than guess. In "Ask before acting" mode it checks in before actions; even in "Act without asking" it still requires explicit permission before things like permanently deleting files. Add ordinary clarifying questions ("two vendors match this brief, which one?") and a several-step task has several chances to stop.

None of that is a bug. A task that pauses before overwriting your onboarding checklist is doing its job. The problem is purely logistical: the pause happens on a screen you are no longer looking at. People describe handing Cowork twenty minutes of work, coming back after lunch, and finding it stopped at the second step with a question that took five seconds to answer.

What Anthropic gives you out of the box

Be fair to the built-in loop: it exists and it is improving. The Claude mobile app can send a push when a Cowork task finishes or when Claude needs your go-ahead, and with cloud sessions you can open the same session from your phone to answer. As of July 2026 the web and mobile experience is rolling out Max-first, with other plans following.

If Cowork is the only agent you run and you are happy answering inside the Claude app, that may be all you need. The gaps show up when Cowork is one of several agents in your day, when you want to answer from the lock screen without opening an app, or when you want a record afterwards of what was asked and what you approved. We keep an honest side-by-side on the Pushary vs Claude Cowork comparison.

Route the questions to your phone

Pushary connects to Cowork as a custom connector, which works on every paid Claude plan. Setup is a paste, not an install:

  1. Get your connector link from Settings, Connections in the Pushary dashboard.
  2. In Claude, open Settings, then Connectors, then Add custom connector, and paste the link. Connectors are account level, so the same connector is available inside Cowork. Enable it in a session under Customize, then Connectors.
  3. Copy the standing instructions block from the same Connections page into Claude Settings, then Cowork. This is the part that makes it proactive: every session now asks you before risky steps and pings you when a task finishes, without you prompting it each time.

From then on, a question arrives as a push notification. Tap it and you land on a decision page: yes or no, a choice, or a short text answer. The answer flows back into the running session and the task continues. Missed a question? It stays answerable for ten minutes, and pending questions sit in your inbox when you next pick up the phone.

One honest limitation

Cowork has no hook system, so nothing, Pushary included, can physically block it mid-action the way the CLI hooks block Claude Code or Codex. The Cowork integration is cooperative: Claude decides when to ask, guided by your standing instructions. In practice that covers the moments that matter, but if you want a hard gate that stops a command until you answer, that lives with Claude Code, Codex, Gemini CLI, Cursor, and Hermes.

One more edge to know about: some Cowork permission prompts are native operating system dialogs on the desktop, and those never leave your machine, no matter which notification tool you use.

The payoff

The point of delegating to Cowork is that the work continues while you do something else. A phone loop is what makes that literal: reads and safe steps run on their own, the genuinely important questions find you wherever you are, and you stop planning your day around checking a screen. If you run other agents too, the same inbox covers all of them at once.

Setup takes about two minutes: connect Claude Cowork.

AG
Aadil Ghani
Founder, Pushary

Building Pushary so an AI agent can reach you on your phone and wait for a yes before it does something you would not want.

Read next

Guides

How to approve Claude Cowork tasks from your phone

Paste one connector link, add standing instructions, and Cowork's questions arrive as push notifications you answer from the lock screen.

Jul 10, 20263 min readAadil Ghani
Guides

An audit trail for Claude Cowork: what did your AI coworker decide while you were out?

Delegation without a record is hope. Keep every Cowork question and answer in an immutable, exportable trail, next to every other agent you run.

Jul 10, 20264 min readAadil Ghani
Guides

What an AI agent audit log should capture for teams and compliance

The fields a coding-agent audit record needs to be worth keeping, and the honest line on what GDPR-aligned and self-assessed actually means.

Jun 27, 20265 min readAadil Ghani

Get a push the moment your agent needs you

Approvals, done alerts, and a kill switch for Claude Code, Codex, Cursor, and the rest. It takes a couple of minutes to set up.