You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 0664044
Browse filesBrowse the repository at this point in the historyBrowse files
fix(server): usage-limit recovery survives notification wakes that did no work
A Claude thread that hits its limit while background subagents run gets one
root run per failed subagent, each started by a notification and completed
within milliseconds. The newest of those became the thread's latest executed
run, so the limited run dropped out of the recovery candidates and the thread
was never armed or resumed (or, if already armed, never resumed at the reset).
A run that a notification started and that completed before the newest
usage-limit failure resets no longer replaces that failure as the latest run.
The SQL shell and candidate queries, the shared usageLimitBlockedRun used by
the server and clients, arming, manual resume, and Stop all use that rule.
Fixes#16815
0 commit comments