auto mode temporarily unavailable - cannot enter auto mode via Shift+Tab or defaultMode setting for Max user

+1 happening to me too

Same issue here.

• Claude Code: 2.1.76
• OS: macOS
• Launch method: claude --enable-auto-mode

"auto mode temporarily unavailable" is displayed persistently, not intermittently.

Same for me, Debian Linux, displayed intermittently.

I checked ~/.claude.json and found the relevant feature flag:

jq '.cachedGrowthBookFeatures.tengu_auto_mode_config' ~/.claude.json

{
  "enabled": "disabled",
  "twoStageClassifier": true
}

Looks like auto mode is controlled server-side via tengu_auto_mode_config.enabled. Mine is still "disabled".

For anyone who has auto mode working, could you share what your tengu_auto_mode_config looks like? Curious if enabled changes to "enabled" or some other value when it's rolled out.

Environment: Claude Code 2.1.74 / macOS / Max subscriber

Same for me. Found this issue via investigation with my own Claude. Seems it's likely simply due to the phased rollout. Too bad that it's not hitting our accounts sooner; I could really use it to simplify approval of long &&'d commands to build, test, and run C++ related tasks.

Thanks @madebyjun for the investigation! Confirmed tengu_auto_mode_config.enabled is "disabled" on my end too. This appears to be a phased rollout rather than a bug. Should we close this issue?

@madebyjun
Thanks! I'm still disabled too.

{
  "enabled": "disabled",
  "twoStageClassifier": true
}

Same here. Windows 11, Claude Code 2.1.79, Max subscriber.

tengu_auto_mode_config.enabled is "disabled" on my end as well. Confirmed that removing defaultMode, skipAutoPermissionPrompt, and launching with --permission-mode auto has no effect.

tengu_auto_mode_config.enabled is still "disabled" on my account as of today (Mar 19). This issue has been open for a week with multiple users reporting the same flag state, but no official response yet. Any update on the rollout status? At this point it's hard to tell if we're just waiting in the queue or if something is stuck.

Same issue here.
Claude Code 2.1.81, macOS, Max subscriber. Confirmed tengu_auto_mode_config.enabled is "disabled". Launched with claude --enable-auto-mode but Shift+Tab only cycles between Plan and Accept Edits — Auto mode never appears.
Would appreciate any update on the rollout timeline.

It's still unavailable for me...

Same issue. Really annoying how long the message persists in the location it is in as it prevents everything except the first line of my statusbar from displaying until it goes away. They should really move the message location for this to all the wasted empty space above the prompt fold.

same here!

This seems to work in 2.1.71.

You can downgrade using:

curl -fsSL https://claude.ai/install.sh | sh -s -- 2.1.71

Then link using ln -sf ~/.local/share/claude/versions/2.1.71 ~/.local/bin/claude

Edit: You may also want to remove any newer versions in ~/.local/share/claude/versions

~~Officially released!! Gonna use the --enable-auto-mode option and see how it works.~~

~~https://x.com/claudeai/status/2036503582166393240?s=46&t=T0gFj_4yBz7zXQvOC5DeBA~~

---

Still not work...

<img width="751" height="191" alt="Image" src="https://github.com/user-attachments/assets/6351d07c-bf01-4e5a-ba5c-f89ce505703f" />

Still not work...

+1

+1

This might not be a bug but a plan-level restriction.

As of March 24, 2026, auto mode is currently available only as a research preview for Team plan subscribers. The rollout roadmap is:

1. Team plan — available now (research preview)
2. Enterprise plan & API users — rolling out in the coming days
3. Individual plans (Pro / Max) — no confirmed timeline yet

If you're on a Pro or Max plan, the auto mode temporarily unavailable message is likely expected behavior rather than a bug. There's been no official announcement on whether or when auto mode will be extended to individual subscribers.

References:

• Official docs: https://code.claude.com/docs/en/permission-modes
• https://x.com/claudeai/status/2036503582166393240

For what it's worth, auto mode is working as expected in my Team plan environment, which further suggests this is a plan-level restriction rather than a platform-wide bug.

Hope this helps clarify the situation!

@smorimoto Thx for the info!! I'll just wait then for Max plan👍

Just leaving a quick note on how I’m setting things up so CC can auto-approve more stuff (without using --dangerously-skip-permission) 📝

• In settings.json, I put stuff like rm -rf in deny (anything obviously risky)
• Then I set allow to Bash(*)

This way it’s basically: “everything is allowed except what’s explicitly denied”

Kinda obvious setup, but it’s something I learned recently.

Any update on this for ubuntu?

This is still broken. When will this be fixed?

This issue has essentially been resolved in the latest release.

“Auto mode now shows ‘unavailable for your plan’ when disabled by plan restrictions (was ‘temporarily unavailable’)”

Ref: https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md#2186

For most users at present, this behaviour is not a bug but a specification. So, it is not broken but functioning as intended (at least for now).

Same issue here. I'm on Max 20x plan (Opus 4.6, 1M context) and auto mode shows "unavailable for your plan". Previously it said "temporarily unavailable", now the message changed making it clear it's a plan restriction.

It's frustrating that Max subscribers — who pay a premium — don't have access to a feature that Team plans already have. Max plan gives access to Claude Code with Opus 4.6 but not to auto mode? That doesn't make sense.

Please expand auto mode availability to Max and Pro plans. We're paying customers too.

Environment:

• Claude Code v2.1.86
• Ubuntu 24.04
• Model: Claude Opus 4.6 (1M context)
• Plan: Max 20x

While the native auto mode availability depends on internal Claude Code logic, you can approximate auto-mode behavior with hooks that auto-approve everything:

INPUT=$(cat)
TOOL=$(echo "$INPUT" | jq -r '.tool_name // empty' 2>/dev/null)
case "$TOOL" in
    Bash)
        COMMAND=$(echo "$INPUT" | jq -r '.tool_input.command // empty' 2>/dev/null)
        if echo "$COMMAND" | grep -qE '(rm\s+-rf\s+/|DROP\s+TABLE|curl.*\|\s*bash|shutdown|reboot)'; then
            exit 0  # Let normal permission flow handle it
        fi
        ;;
esac
jq -n "{hookSpecificOutput:{hookEventName:\"PreToolUse\",permissionDecision:\"allow\",permissionDecisionReason:\"Hook-based auto mode\"}}"
exit 0

{
  "hooks": {
    "PreToolUse": [{
      "hooks": [{ "type": "command", "command": "bash ~/.claude/hooks/auto-mode.sh" }]
    }]
  }
}

To make it toggleable without editing settings:

TOGGLE="$HOME/.claude/.auto-mode"
if [ ! -f "$TOGGLE" ]; then
    exit 0  # Auto mode disabled — normal permission flow
fi
INPUT=$(cat)
COMMAND=$(echo "$INPUT" | jq -r '.tool_input.command // empty' 2>/dev/null)
if echo "$COMMAND" | grep -qE '(rm\s+-rf\s+/|DROP\s+TABLE)'; then
    exit 0
fi
jq -n '{hookSpecificOutput:{hookEventName:"PreToolUse",permissionDecision:"allow",permissionDecisionReason:"Hook-based auto mode"}}'
exit 0

Toggle: touch ~/.claude/.auto-mode to enable, rm ~/.claude/.auto-mode to disable.
This gives you auto-mode behavior regardless of whether the native auto mode is available, with the added benefit of a customizable deny list.

A note of caution regarding the workarounds suggested above (broad allow rules in settings.json, or PreToolUse hooks that auto-approve everything):

These are not workarounds for auto mode. They simply skip the permission prompt for all tool calls, which is functionally equivalent to --dangerously-skip-permissions. This has nothing to do with auto mode availability — you could have done this before auto mode existed.

Real auto mode uses an AI-based classifier (twoStageClassifier in the feature flag) that evaluates the risk of each action in context. The approaches above replace that with hand-written regex deny lists, which are trivially bypassable (e.g., rm -rf / is blocked but find / -delete or chmod -R 000 / passes through).

If you want fewer permission prompts while waiting for auto mode on your plan, consider selectively allowing specific tools you trust (e.g., Read, Write, Glob, Grep) rather than blanket-allowing everything. That gives you a better workflow without sacrificing safety.

Update: https://x.com/claudeai/status/2038693742094246032?s=46

Why not for Max/Pro plan 😭😭😭

It is a kind of sequence. And speaking as someone who uses it, it is still quite unstable. Internally, Sonnet is used, but it frequently gets caught up in system outages, so Auto Mode cannot be used very often. Given that it is being rolled out so gradually and yet remains in this state, that is reason enough to make this decision...

I'd love to see auto mode available on Max plans. Right now the only way to get uninterrupted workflow is --dangerously-skip-permissions, which is a blunt instrument — it bypasses all safety checks entirely.

Auto mode is the better answer here. It still runs actions through an AI safety classifier, so you get the productivity of not being prompted on every file edit while keeping a meaningful safety net. That's a much more responsible default than "skip everything."

For users who work solo on private repos and just want to stay in flow, the current choice is either click-approve every action or disable all guardrails. Auto mode fills the gap perfectly — please make it available beyond Team/Enterprise.

Why even remove such a feature?

<img width="863" height="268" alt="Image" src="https://github.com/user-attachments/assets/77fe24aa-c5db-443a-b736-8ba4702b5195" />
YOOOOOOOO

https://claude.com/blog/best-practices-for-using-claude-opus-4-7-with-claude-code

Available for Max users!!!

Finally available for Max users! 🎉 Closing!! Thanks everyone for the investigation and comments along the way. Happy auto-moding! 🚀

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.