You'll noticeTier: how much it should matter to you
2Useful: my rating, 1 to 5
1Signal: worth watching, 1 to 5
Auto ModeArea: what it touches
ImprovementsKind: in v2.1.321,
ImprovementsSection of the release
What
In auto mode, Claude Code uses a classifier (a model that reviews actions) to decide whether something is safe to do without asking you. A subagent is a helper agent Claude starts to work on part of a task.
When the classifier cannot check a subagent's prompt, the subagent is now not started. The reason given says the conversation may be too long, or to try again in a few minutes.
The warning shown when a subagent's output is passed on without review has new wording. Which warning appears now depends on the environment variable CLAUDE_CODE_USER_DELEGATED_SESSION_DIRECT_API.
Why
In auto mode, starting a subagent can now fail when the classifier is down, so you may need to retry.
