A user report says Claude Code puts the account email in the model's context, with no documented off switch

The reporter says Claude Code adds a "# userEmail" block, "The user's email address is ...", to every session. They say the value comes from the OAuth account in ~/.claude.json, that no setting turns it off, and that IS_DEMO=1 hides the address from the banner and /status but not from the model. These findings come from their reading of the decompiled bundle. We have not reproduced them, and the page text we read shows no response from Anthropic.

The reporter also describes two incidents. They say the model printed the address unprompted, and that it passed the address as git -c user.email on ten commits, overriding a configured noreply identity until GitHub's push protection rejected the push. They list four earlier issues on the same behavior, each closed for inactivity or as a duplicate. The issue had 74 thumbs-up reactions when we read it. A Hacker News post pointing to it has 24 points and 3 comments, and links a Reddit thread alleging the model sent the address to websites; we did not open that thread.

Why it matters: Anything in an agent's context can end up in a commit, a pull request body or an outbound tool call, and an unattended run has no reviewer to catch it. If these claims hold, an author-identity check before push is a cheap guard, and a CLAUDE.md instruction does not remove the data from context.