[BUG] Auto mode blocks all Bash actions with endless retries when Opus 4.8 classifier is temporarily unavailable

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

When the selected model (Opus 4.8) is temporarily unavailable due to capacity, the auto-mode safety classifier — which itself runs on the same model — cannot evaluate Bash actions. Instead of failing gracefully or falling back, every Bash command enters an endless retry loop ("Retrying in Ns · attempt x/10") and the session hangs for minutes at a time.

The blocking error is:

claude-opus-4-8[1m] is temporarily unavailable, so auto mode cannot determine the safety of Bash right now. Wait briefly and then try this action again. If it keeps failing, continue with other tasks that don't require this action and come back to it later.

This effectively makes the whole session unusable for any task involving Bash, even though it's a transient capacity issue and not an account/request problem. The status page showed "All Systems Operational" the entire time, so there was no incident to wait on. Switching to an Opus variant (4.7) did not help since it shares the same capacity pool; only switching to Sonnet resolved it.

What Should Happen?

When the model backing the auto-mode safety classifier is temporarily unavailable, Claude Code should:

1. Not enter a silent 10-attempt retry loop that hangs the session for minutes.
2. Fail fast with a clear, actionable message after 1–2 attempts.
3. Ideally fall back to an available model for the safety classification (or offer to), so Bash actions aren't fully blocked just because the chat model is at capacity.
4. Make it obvious that switching to a model on a different capacity pool (e.g. Sonnet) is the workaround.

Error Messages/Logs

Error: claude-opus-4-8[1m] is temporarily unavailable, so auto mode cannot determine the safety of Bash right now. Wait briefly and then try this action again. If it keeps failing, continue with other tasks that don't require this action and come back to it later.
Note: reading files, searching code, and other read-only operations do not require the classifier and can still be used.

Spinner: Retrying in 0s · attempt 1/10

Steps to Reproduce

1. Set the model to Opus 4.8 (/model claude-opus-4-8) during a period when Opus is at capacity.
2. Ask Claude Code to run any Bash command (e.g. a git/worktree inspection).
3. Observe that the auto-mode safety classifier cannot evaluate the action.
4. Observe the session entering "Retrying in Ns · attempt x/10" loops and hanging for 30s–2min per attempt.
5. Note: switching to Opus 4.7 does not help (same capacity pool); switching to Sonnet 4.6 resolves it immediately.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.158

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

• claude --version: 2.1.158 (well above the v2.1.154 minimum for Opus 4.8)
• Anthropic status page showed "All Systems Operational" throughout, so this transient capacity condition is not surfaced as an incident.
• Core issue is the coupling between the auto-mode safety classifier and the chat model: when the chat model is at capacity, the classifier is too, which blocks even unrelated Bash work. Decoupling the classifier model (or a fallback) would prevent full-session blocking.