Claude Code will automatically block this request: what it means
Why Claude Code counts down two minutes and then blocks an rm, which versions added it, and how to answer it in time from the terminal or your phone.

"Claude Code will automatically block this request in 1:58, to avoid holding up an unattended session" means Claude Code stopped before an rm it will not run without a person's answer, and it is giving you two minutes to answer. If nobody does, it refuses the step and carries on. You see it when Claude tries to delete a protected folder, such as everything in your project with rm -rf ./*.
Key takeaways
- The countdown covers
rmandrmdircommands aimed at a critical path: the working folder or its parents, your home directory, a top-level folder or a drive root. No allow rule and no PreToolUse hook allow approves one, even with--dangerously-skip-permissions. - Claude Code 2.1.326 (16 November 2026) and 2.1.330 (20 November) made more steps ask in bypass mode.
- At zero the step is refused, not run, and the run continues. Press any key to stop the countdown and keep the prompt open.
- Answer it in time, turn on
permissions.denyInsteadOfPromptingto refuse at once, or get the question on your phone before the countdown starts.
What the message means
The prompt in the terminal looks like this:
Bash command
Delete all files in the current directory
rm -rf ./*
│ Dangerous rm operation on working directory or its ancestor: /path/to/your/project/*
⚠ Claude Code will automatically block this request in 1:58, to avoid holding up an unattended session
Do you want to proceed?
❯ 1. Yes
2. NoClaude Code puts a timer on this prompt because the session might have nobody in front of it. 2.1.281 "changed the dangerous rm prompt in --dangerously-skip-permissions and auto mode to wait 2 minutes for an answer, then deny the command with a rewrite hint so unattended sessions keep going." Before that, a run could sit on this question until someone came back.
The step is one Claude Code will not approve by rule. This is what it told the model in our claude -p run: "This command would remove a workspace directory (the working directory, an additional working directory, or one of their parent directories). This requires explicit approval and cannot be auto-allowed by permission rules."
Why it started showing up in November
Two releases widened the list of steps that ask even in bypass mode:
- 2.1.326, 16 November 2026: "Changed some tool calls in bypass permissions mode to ask for approval, or be denied where no one can be asked." The same release asks about more PowerShell commands, even in bypass mode.
- 2.1.330, 20 November 2026: "Changed more rm commands aimed at the root, home, a top-level folder or the working folder to ask for approval; where nobody can answer, they're refused."
So more cleanup steps in a run you left with --dangerously-skip-permissions now stop at this prompt. Our --dangerously-skip-permissions guide covers what else the flag still leaves on.
What happens when nobody answers
We ran it on 22 November 2026 with Claude Code 2.1.330 in bypass mode, asking it to empty a scratch folder with rm -rf ./*, and did not answer. Pushary's hooks were installed on their default setting, which lets the step through to Claude Code's own check:
- In an open terminal: the prompt counted down from 2:00. At zero, Claude Code printed "Not run · permission denied" with the reason "Permission for this command was denied by a built-in Claude Code safety check". Nothing was deleted. Claude told us the command was blocked and that we would need to run it ourselves.
- In
claude -p: there was no countdown. Nobody can answer a print-mode run, so the step was refused straight away.
Either way the run goes on without the step, and the work you asked for is not done. After three of these prompts run out unanswered in one session, Claude Code stops showing them and refuses further critical-path removals at once, until you send a new message. In the Desktop app and the VS Code chat panel, bypass mode refuses these removals at once (2.1.309), unless a PermissionRequest hook allows them.
How to answer it in time
Be at the terminal. Pick Yes or No within two minutes. Press any key while the prompt is open to stop the countdown and keep it waiting for you. If the rm is broader than you meant, pick No and ask Claude to delete the specific subfolder instead.
Refuse it at once. If you would rather the run skip these steps than wait, set permissions.denyInsteadOfPrompting to true in ~/.claude/settings.json (Claude Code 2.1.319 or later). The step is denied immediately and the turn continues. The setting also takes AskUserQuestion and the plan mode tools out of the session, so Claude loses its built-in question tool, and a session in plan mode cannot leave it.
Answer from the Claude app. With Remote Control on and "Push when actions required" switched on in /config, the prompt reaches the Claude app on your phone. It still has to be answered inside the same two minutes. Our Claude Code notifications guide has the setup and its limits.
Get it on your phone before the countdown starts. Allow rules cannot answer this prompt, but a PermissionRequest hook can. Pushary's hook asks before Claude Code shows its own prompt, and Claude's two-minute timer does not run while that hook is waiting. If one of your Pushary rules names the command, or you chose "Stop me for risky steps" for sessions that skip permissions, the question goes to your phone as a push, and with calls turned on Pushary can ring you if you miss it. From Pushary CLI 1.9.238, a Yes you give on your phone also answers Claude Code's own prompt for the same step, so the step runs. A No refuses it. With the default "Never stop me" and no rule of yours naming the command, Pushary lets the step through, so Claude Code's own countdown is what you get.
Turn the prompt off. Since 2.1.311, starting Claude Code with CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS=1 in bypass mode runs these removals with no prompt. A command such as rm -rf ~ can then run without asking you, so Anthropic says to set it only in an isolated environment whose contents you can afford to lose. It has no effect while any PermissionRequest hook is installed, Pushary's included, or in the Desktop app or the VS Code chat panel.
Common questions
Why does Claude Code ask even with --dangerously-skip-permissions?
Claude Code does not auto-approve an rm or rmdir aimed at a critical path in any mode: the working folder or its parents, your home directory, a top-level folder or a drive root. No allow rule and no PreToolUse hook allow approves one. Version 2.1.326 made more tool calls in bypass mode ask, or be denied where no one can be asked, and 2.1.330 added more rm commands. Since 2.1.311 you can turn the rm prompt off in bypass mode with CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS=1, which Anthropic says to use only in an isolated environment you can afford to lose. It has no effect while any PermissionRequest hook is installed, Pushary's included, or in the Desktop app or the VS Code chat panel.
What happens when the countdown reaches zero?
Claude Code refuses the step, tells Claude what to do instead, and the run continues without it. In our test it showed "Not run · permission denied", with the reason that a built-in Claude Code safety check denied the command. Nothing was deleted. After three of these prompts run out unanswered in one session, Claude Code refuses further critical-path removals at once until you send a new message.
Can I make Claude Code refuse these steps at once instead of waiting two minutes?
Yes. Turn on permissions.denyInsteadOfPrompting (Claude Code 2.1.319 or later). A tool call that would wait at a permission prompt is denied and the turn continues. It also takes AskUserQuestion and the plan mode tools out of the session, so Claude loses its built-in question tool, and a session in plan mode cannot leave it.
Does Claude Code wait the full two minutes in claude -p?
No. In a claude -p run with no permission prompt tool, Claude Code refuses the step straight away unless a PermissionRequest hook allows it, and the run continues. The Desktop app and the VS Code chat panel do the same in bypass mode.
Frequently asked questions
Why does Claude Code ask even with --dangerously-skip-permissions?
Claude Code does not auto-approve an rm or rmdir aimed at a critical path in any mode: the working folder or its parents, your home directory, a top-level folder or a drive root. No allow rule and no PreToolUse hook allow approves one. Version 2.1.326 (16 November 2026) made more tool calls in bypass mode ask, or be denied where no one can be asked, and 2.1.330 (20 November) added more rm commands. Since 2.1.311 you can turn the rm prompt off in bypass mode with CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS=1, which Anthropic says to use only in an isolated environment you can afford to lose. It has no effect while any PermissionRequest hook is installed, Pushary's included, or in the Desktop app or the VS Code chat panel.
What happens when the countdown reaches zero?
Claude Code refuses the step, tells Claude what to do instead, and the run continues without it. In our test it showed Not run, permission denied, with the reason that a built-in Claude Code safety check denied the command. Nothing was deleted. After three of these prompts run out unanswered in one session, Claude Code refuses further critical-path removals at once until you send a new message.
Can I make Claude Code refuse these steps at once instead of waiting two minutes?
Yes. Claude Code 2.1.319 added permissions.denyInsteadOfPrompting. With it on, a tool call that would wait at a permission prompt is denied and the turn continues. It also takes AskUserQuestion and the plan mode tools out of the session, so Claude loses its built-in question tool, and a session in plan mode cannot leave it.
Does Claude Code wait the full two minutes in claude -p?
No. In a claude -p run with no permission prompt tool, Claude Code refuses the step straight away unless a PermissionRequest hook allows it, and the run continues. The Desktop app and the VS Code chat panel do the same in bypass mode.

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.