Video: Permission Modes

1. Video: Permission Modes

Permission modes lets you decide once what's safe to run without you. You already cycled through several of them with shift-tab, manual, acceptEdits, and plan. The rest are where hands-off Claude Code lives, and the one to lead with is auto. Claude runs without prompts while a separate classifier reviews each action before it executes, stepping in on ones it identifies as dangerous. So here are the permission modes. First we have manual, which reads only without prompting. Everything else asks. AcceptEdits reads file edits in common file system bash commands. This is for iterating on code that you review after the fact. Plan reads only, it researches and proposes without editing. Auto mode accepts everything with a classifier model reviewing each action before it runs. Don't ask, only pre-approved tools allowed, everything else auto-denied with no prompt. This is great for your CI pipelines. And then bypass permission skips all check. This is equivalent to the dangerously skip Only run it inside of an isolated container or virtual machine. Press shift-tab to cycle through manual, acceptEdits, plan, and auto. The status bar shows which one that you're in. Auto mode is where Claude runs, but before each action executes, a separate classifier model reviews it. The classifier guards intent. It blocks the moves that escalate beyond your request, production deploys and migrations, force pushing or piping download code into a shell, sending sensitive data to external endpoints and destroying files that exist for the session. It allows the everyday work, local edits in your project, installing dependencies from your log file, read only requests and pushing to your own branch. Now the classifier won't catch whether the code works or not. So if you ask Claude to refactor authentication and it writes broken authentication, the classifier waves it through because broken isn't dangerous. Use auto mode together with a stop hook that runs your tests. Auto mode watches what Claude is trying to do while the hook confirms that the code actually runs. Auto mode's guardrails are still evolving, so check the docs for the current block and allow lists. Don't ask is the right move whenever no human is there to approve prompts. CI, scheduled jobs, overnight batches. Anything off that list gets auto denied, so your pipeline keeps moving instead of hanging on an approval no one will give. There are several permission modes and you can reach the everyday ones by cycling shift tab. Auto is the hands-off default. The classifier checks intent before each action runs and a stop hook checks correctness after. Don't ask covers unattended pipelines and bypass permission belong only inside isolated containers and VMs. Match the mode to the job.

2. Let's practice!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.