False-positive Usage Policy block mid-session (req_011CbJudbehY5Yi6gtM4xko4)

What happened

Mid-session, Claude Code returned the following block message while I was working on a normal software engineering task (iOS Conversation Mode work in a private repo):

API Error: Claude Code is unable to respond to this request, which appears to violate our Usage Policy (https://www.anthropic.com/legal/aup). Please double press esc to edit your last message or start a new session for Claude Code to assist with a different task. Request ID: req_011CbJudbehY5Yi6gtM4xko4

The content being processed was ordinary application code — no policy-violating material. Appears to be a safety-classifier false positive.

Context

• Model: claude-opus-4-7 (1M context)
• Working directory contains: OAuth/JWT code, KeyVault secret references, Cosmos DB cleanup scripts ("wipe", "destroy" appear in CLAUDE.md as safety guidance), Zoom SDK integration. These are normal for the codebase and have been worked on across many prior sessions without issue.

UX issue with the block message itself

The message instructs the user to "double press esc to edit your last message." In practice this does not appear to unblock the session — the offending content remains in the transcript and re-sending hits the same classifier. Starting a new session is the only reliable workaround. The message should either:

1. Actually support inline edit-and-retry, or
2. Direct users to \/clear\ / new session as the primary remediation.

Request

Please review req_011CbJudbehY5Yi6gtM4xko4 for classifier tuning, and consider updating the user-facing message to reflect actual remediation steps.