Guide
By default Claude Code asks before it edits files or runs commands. The --dangerously-skip-permissions flag turns every one of those prompts off. That is what lets an agent work through a long task unattended, and it is why the flag has the word "dangerously" in it.
Get startedStarted with claude --dangerously-skip-permissions, Claude Code runs every tool call without asking: it writes and deletes files, runs shell commands and installs packages as the user that launched it. Nothing about the model changes; only the confirmation step is gone.
A misread instruction can delete files, rewrite git history or drop a local database, and there is no prompt to catch it.
On a laptop the agent can read anything you can: SSH keys, cloud credentials, browser profiles and every other project in your home directory.
Text in a web page, issue or dependency can tell the agent to do something you never asked for. With prompts on, you would see the command first.
Commands run with your network access, including anything reachable inside your office or home network.
Claude Code can allow specific tools without allowing everything. Allow rules in its settings file approve particular commands, such as your test runner, while it still asks about the rest, and a mode that auto-accepts file edits removes most prompts on a normal coding task. Check Claude Code’s own documentation for the current option names.
Run it somewhere a bad command cannot reach anything that matters: a container at minimum, and ideally a virtual machine that holds only this project.
Give the environment only the tokens the task needs, scoped to one repository and short-lived where possible.
Work on a branch, commit often and push. If the agent wrecks the checkout, you throw it away and clone again.
Let the agent work without prompts, but read the diff before it merges. A pull request is a better checkpoint than a hundred confirmations.
Every Claude Code tile and lane in YOLO Studio starts with --dangerously-skip-permissions. The checklist above is how the workspace is built: each one is a virtual machine holding one repository, the agent runs as a non-root user, network rules keep it away from other workspaces, credentials arrive per session, and each lane is its own git clone on its own branch. Results can land as pull requests for review.
In YOLO Studio, YOLO Mode is the name for agents running without permission prompts. For Claude Code that is this flag; Codex gets the same effect from its approval and sandbox settings.
Claude Code always starts with the flag in a YOLO Studio workspace. The workspace itself is the safety boundary.
Sign in with GitHub, open a workspace on your repository, and put your first agent to work in a few minutes.
Get started
