What probably matters to youSection of the release
Unclear It is not clear whether the timer itself now covers these cases or only its help text was updated.
What
The help text for --startup-timeout-min has been rewritten. It is the command-line flag that sets how long a self-hosted runner waits on a session child, meaning the separate Claude Code process the runner starts for each session. The help used to say the timer only caught a child that hung while restoring an earlier conversation with --resume. It now says the timer also covers:
connecting to MCP servers (outside programs that give Claude Code extra tools)
a session that has been assigned but has no input waiting
It also says the timer starts again each time the child writes output.
Separately, Claude Code refuses web fetch responses that are too large. That check now skips responses with certain status codes, in the same way it already skipped HEAD requests.
Why
If you run a self-hosted runner, the timeout can now apply in more situations than the old help described. A child that is slow to connect to MCP servers, or that sits idle with nothing to do, can be stopped when the timer runs out. A child that keeps writing output keeps resetting the timer.
How sure we are
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubtIt is not clear whether the timer itself now covers these cases or only its help text was updated.
Anthropic's release notes agreeFixed bare mode notices in background sessions telling you to restart without --bare, which keeps the mode; they now name /stop and claude…
