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
Header wording (.github/workflows/fork-secret-sweep.yml:20-22): it can be read as "a re-run repairs an undone reset". It does not; say "a new run by hand with reset_record, not a re-run". Also: the comment says "first attempt" where it means "newest earlier attempt". (A1, A4, B nit)
An undone reset is silent: a sweep that starts while a reset run is going on can undo the reset, and nothing warns about it. It never causes less scanning than before the reset, but the owner cannot see it happened. Options: a per-ref concurrency group on the sweep, or a run-name marking reset runs so a normal sweep that sees an unfinished earlier reset takes none failed. (A2)
Reset re-run can lose the reset: if attempt 1 kept no record and one read in attempt 2 (tag check or listing its own record) fails, the next sweep reads the record from before the reset. Two failures in a row, and the run goes red. Fix: in a reset run, fall back to the empty record on any read failure. (B1)
Timeout check skips quietly if the time-limit line it looks for is reworded; make it fail instead. (B nit)
Activity
scuttletoby commentedon Nov 22, 2026
Point 3 (live check of a re-run reading its own earlier record) did not confirm; it continues in #950. The other points are done in #948.
🤖 Generated with Claude Code
fix(secret-sweep): alert group never the workflow's; header says re-r…
fix(secret-sweep): sweeps of a ref one at a time; a reset run never f…