Auto mode among the permission modes
Claude Code has six permission modes. They differ in what Claude may do without asking you:
| Mode | Runs without asking |
|---|---|
Manual (default) |
Reads only |
acceptEdits |
Reads, file edits and common file commands in the working directory |
plan |
Reads and exploration; no source edits until you approve a plan |
auto |
Everything, subject to permission rules, automatic approvals and classifier review |
dontAsk |
Only pre-approved tools; everything else is denied |
bypassPermissions |
Everything, with no check |
Deny rules block in every mode, including bypassPermissions. Explicit ask rules still force a prompt in auto mode.
How the check works
Actions that permission rules and automatic approval do not settle go to a separate classifier model before they run. Reads and file edits in the working directory, outside protected paths, skip it; mainly shell commands and network operations reach it. It blocks actions that escalate beyond what you asked for, that target infrastructure it does not recognize, or that appear driven by hostile content Claude read along the way. By default the classifier runs on Claude Sonnet 5.
The classifier sees your messages, Claude’s tool calls other than read-only lookups, and your CLAUDE.md. Tool results are stripped, so text in a file or web page cannot talk to the classifier directly.
Blocked by default, among others: downloading and executing code (curl | bash), sending sensitive data to outside endpoints, production deploys and migrations, force push, git reset --hard, terraform destroy, merging a pull request no human approved, and printing live credentials into the transcript.
Allowed by default, among others: local file work in the working directory, installing dependencies declared in lock files or manifests, read-only HTTP requests, and pushing to any branch of the current repository.
On entering auto mode, broad allow rules that grant arbitrary code execution, such as a blanket Bash(*), are dropped. Narrow rules like Bash(npm test) stay, and the dropped ones come back when you leave auto mode. Run claude auto-mode defaults to see the full rule lists.
How to turn it on, and availability
With Claude Code v2.1.283 or later, auto mode is the starting mode for interactive terminal and VS Code sessions on every plan and provider. Otherwise:
- Press Shift+Tab until the status bar shows
⏵⏵ auto mode on. - Or start with
claude --permission-mode auto. - Or set
"defaultMode": "auto"underpermissionsin~/.claude/settings.json. A project settings file will not do it. - For
claude -pand the Agent SDK, pass--permission-mode autoorpermissionMode: "auto".
It works on all plans. On Team and Enterprise an administrator can switch it off in managed settings. It needs a recent model: on the Anthropic API Opus 4.6 or Sonnet 4.6 and later, Haiku 5.5 or a Fable model; on Bedrock, Google Cloud’s Agent Platform and Foundry, Sonnet 5 or Opus 4.7 and later, Haiku 5.5 or Fable.
The risks
The docs say it plainly: auto mode reduces permission prompts but does not guarantee safety. Use it where you trust the general direction of the task, and keep human review for sensitive operations. In practice:
- The classifier is a model. It judges; it does not enforce. The Claude Code best practices put it in the non-deterministic column. A rule that may never bend belongs in a deny rule or a hook.
- Pushing is allowed. A push to any branch of the current repo goes through by default. Branch protection on
mainis what stops that from reaching production. - Trusted infrastructure is narrow at first. The classifier trusts your working directory and the configured git remotes. An administrator adds trusted repos, buckets and services through the
autoMode.environmentsetting; otherwise routine actions may get blocked. - Fewer questions. Auto mode also nudges Claude to keep working instead of stopping for clarifying questions. If a task is ambiguous, say so in the prompt or start in plan mode.
- Teammates follow the lead. In an agent team, an in-process teammate runs in the lead’s mode while the lead is in auto mode.
Compared with bypassPermissions, auto mode is the safer way to skip prompts: the docs say bypass offers no protection against prompt injection and belongs only in isolated containers or VMs. For a broader view, see is Claude Code safe for company code.
What this means for a team
Auto mode makes a session behave differently per developer: the classifier sees each person’s own messages and CLAUDE.md, and admins decide centrally whether it is allowed and what infrastructure counts as trusted. The team’s hard limits still belong in committed deny rules, hooks and branch protection, which work the same for everyone.
Where Bombos fits
The classifier judges actions against your request and CLAUDE.md. Bombos is a plugin that gives Claude Code and the team’s other AI tools one shared memory, so the agent works from the team’s actual knowledge and the rules learned from earlier corrections, each traceable to its source. One developer can try it on their own agents first.
Related questions
Which plans have auto mode?
All plans, per the permission modes docs. On Team and Enterprise it is on by default, and administrators can turn it off by setting permissions.disableAutoMode to "disable" in managed settings.
Which models support auto mode?
On the Anthropic API: Claude Opus 4.6 or later, Sonnet 4.6 or later, Haiku 5.5 or a Fable model. On Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry: Sonnet 5 or later, Opus 4.7 or later, Haiku 5.5 and the Fable models. Sonnet 4.5, Opus 4.5, Haiku 4.5 and older are not supported.
What happens when auto mode blocks something?
You get a notification and the action appears under Recently blocked in /permissions, where you can approve it so Claude may retry. After 3 blocks in a row or 20 in total, auto mode pauses and Claude Code asks you again.
Does the classifier cost tokens?
It runs on Claude Sonnet 5 by default. On Enterprise plans and on accounts that use the Claude API, Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry, classifier calls count toward your token usage.
Why does auto mode in my project settings not work?
A defaultMode of "auto" in a project's .claude/settings.json or settings.local.json does not take effect. Put it in ~/.claude/settings.json, or start with claude --permission-mode auto.
