# Claude Code Deletion Check

> Claude Code asks for confirmation before some deletions, such as rm -f * after a cd, even in autonomous mode, then blocks the command after two minutes.

- Area: Desktop app
- Plans: All plans
- Last checked against the product: 2026-11-07
- Web page: https://agentsroom.dev/docs/claude-removal-check

## What it does

Claude Code asks for confirmation before some deletions, such as `rm -f *` after a `cd`, even in autonomous mode, then blocks the command after two minutes. An agent working without you lost its step on a question nobody was there to answer. AgentsRoom now answers that question for autonomous agents, so they carry on, and keeps it for the deletions worth asking about.

## Where to find it

Settings > AI providers > Claude tab, under **Autonomous mode** when it is selected: **Answer Claude's deletion check**.

## How to use it

Nothing to do: it is on by default. Turn it off to keep Claude's question on every deletion it flags.

## Settings

- `skipClaudeRemovalCheck` (global, synced with your account, Settings > AI providers > Claude tab > **Answer Claude's deletion check**): `true` by default. `false` = Claude always asks. Read at each prompt, so switching it holds for agents already running.

## Agent tools (MCP)

`settings_get` / `settings_set` with the key `skipClaudeRemovalCheck`.

## Providers

Claude Code only: no other CLI has this check. Manual mode is never concerned: there, you approve what the agent does.

## Mobile

No setting on the phone. It applies to every autonomous Claude agent on the computer that runs it, whichever device started the agent. A question Claude keeps still reaches your phone as usual.

## Limits

- Claude still asks when the target is the root of a disk, a system folder, your home folder, the project folder or one of their parents; when a variable it cannot read, or a `cd` that may fail, could turn the target into one of those; and for `sudo`, `eval` or `sh -c`.
- Claude still asks when you set your own `permissions.ask` rule or hook on `Bash`, or when a managed policy applies: AgentsRoom never goes over a guard you or your company set.
- Consoles on a remote host over SSH run without AgentsRoom's hooks: Claude keeps asking there.
- It reaches an agent at its next launch: restart an agent that was already open before the update.
- This is not a sandbox: a script can always delete what it wants.

## Common questions

- **My agent stopped on "Compound command contains cd with write operation" and "Claude Code will automatically block this request". Why, in autonomous mode?** The question comes from Claude Code itself, and its wording is misleading: a `cd` followed by a write does not stop an autonomous agent, the deletion does. Since 2026-11-07 AgentsRoom answers it for autonomous agents.
- **I still see it. Why?** The target is worth the question (see Limits), you have your own ask rule or hook on `Bash`, the console runs over SSH, or the agent started before the update: restart it.
- **Is it safe?** It removes the question for ordinary deletions, not for your home folder, the project folder or a system folder. If you prefer to be asked every time, turn the setting off.

## Related

- [Multi-Provider Support](https://agentsroom.dev/docs/multi-provider.md): autonomous and manual modes, and Claude's account email check.
- [Needs Input Inbox](https://agentsroom.dev/docs/needs-input-inbox.md): how an agent waiting for you is flagged.
