In auto mode, a classifier (a model that judges whether an action is safe) usually checks each tool call, meaning each action Claude takes such as running a command or editing a file. Tool calls that Claude Code generates itself for a specific tool, for example ones marked synthesizedBy: "observer-gate", now skip that check and are allowed. The recorded reason starts with harness-synthesized:.
This happens only when both modes involved allow auto approval and no safety check has flagged the call. The existing shortcut for plugin-originated calls now shares the same underlying code.
Actions Claude Code starts on its own no longer wait for a classifier verdict. Any call that a safety check has flagged still goes through the normal checks.
tengu_structured_creek Not enough to sayNothing here resolved what this flag was doing on this version, so nothing here should be read as on or off.
This account: no value returned · anonymous baseline: no value returned · compiled default in v2.1.281: on
These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.
Read once, for one account on one subscription tier, against v2.1.281. It isn't a statement about your account. What a flag value here can and cannot tell you
New in this build: tengu_structured_creek