A forked agent is a separate copy of Claude that Claude Code starts to handle part of a task, carrying on from the main conversation's history. When Claude Code starts one, it can now pass two more things:
- A count of API attempts left, which the forked agent uses for its own requests to the model.
- A set of tool replacements, where some of the main conversation's tools are swapped for copies that behave differently when run but are otherwise identical.
Claude Code refuses to start the fork, with an error, if the fork does not pick up exactly where the main conversation left off, or if a replacement is not a genuine copy of one of the main conversation's tools. A related relay mode for forks remains behind a switch that is off unless it is turned on remotely.
Passing on a count of attempts left fits with retries after a dropped connection now drawing from the same budget as a request's other retries, so a forked agent cannot quietly start its retry count over.
tengu_wild_tome Off by default, switched on for this accountThe shipped code defaults this off, and the flag server returned on for the one account this site reads on this version. That is the reading that makes the entry above worth a second look, and it still says nothing about your account.
This account: on · anonymous baseline: on · compiled default in v2.1.286: off
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 v2.1.286. It isn't a statement about your account. What a flag value here can and cannot tell you
It is not clear which parts of Claude Code pass tool replacements to a forked agent, or whether this is in use yet.