termdeck
Plan only
Claude explores, proposes a plan, and waits for your approval before it acts. Nothing is written while the mode is active.
Claude Code guide: permission modes
Claude Code permission modes, explained
A permission mode decides what Claude Code may do without stopping to ask you. Claude is the only one of the three engines that offers all five modes, and the only one where auto means a second model is doing the judging rather than a wrapper guessing on your behalf.
Anthropic
On Claude Code
Supervised
Every command and every file change stops for a decision. This is the mode to use on a repository you do not want touched by accident.
Accept edits
File changes go through without asking. Anything that executes still stops. Codex does not offer an equivalent, which is worth knowing before you switch engines mid-project.
Auto
A separate classifier model reviews each risky action rather than asking you every time. This is Anthropic behaviour built into Claude Code, not an approximation layered on top by a client. New chats start here.
Full access
Nothing is asked and nothing is blocked. Reserve it for a sandbox or a throwaway checkout, because there is no second gate behind it.
Compare
One vocabulary, three engines
| Claude Code | Codex CLI | Grok CLI | |
|---|---|---|---|
| Plan only | Plans, then waits for approval | Native plan mode, read-only sandbox | Not supported, behaves as read-only |
| Supervised | Asks before every action | Asks, and sandboxes to read-only | Refuses outright instead of asking |
| Accept edits | File edits pass, commands ask | Not offered | File edits pass, commands ask |
| Auto | A classifier reviews risky actions | Workspace-write sandbox, asks on request | Not offered |
| Full access | No prompts at all | No prompts, sandbox disabled | No prompts at all |
| Starts new chats in | Auto | Auto | Accept edits |
Termdeck drives all three from one console, so the differences above stop being three separate habits to remember.
FAQ
Claude Code permission modes, answered
What permission mode does Claude Code start in?
New chats start in auto, which follows the upstream Claude Code default. In auto a separate classifier model reviews risky actions instead of prompting you for each one.
What is the difference between auto and accept edits?
Accept edits is a fixed rule: file changes pass, anything that executes still asks. Auto is a judgement call made per action by a classifier model, so a risky edit can still be stopped and a routine command can still pass.
Is auto mode safe to leave running overnight?
It is the mode designed for it, but it is still a model making the call. On an unattended run the thing that matters more than the mode is being able to see what happened and stop it, which is what a remote console gives you.
Can I change the permission mode mid-session?
Yes. The mode applies to the next turn, not retroactively to work already done, so switching to supervised part way through does not undo anything the agent has already changed.
Claude Code, Codex and Grok in one browser console