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
Pin the CI Gate to a pre-0.1.11 pr-monitor; the rolling v1 fails closed on dirsql's dynamic matrices #1015
The CI Gate job in .github/workflows/pr-monitor.yml uses the rolling tag thekevinscott/pr-monitor@v1. Since 2026-08-21 that tag installs willfire 0.1.11, which is an exact fail-closed gate: it refuses to gate a check set containing entries it cannot resolve. Every dirsql PR now goes red on the gate regardless of whether the PR itself is sound.
Observed on #1014 (run 32499025181), where all 27 other checks pass and only the gate fails:
Two distinct sources, and only one of them has a fix available:
testing-conventions contributes twelve of the thirteen. Resolvable today with an execute: thekevinscott/testing-conventions:detect grant.
putitoutthere's build.ymlplan job contributes the last one and checks out with fetch-depth: 0. The willfire executor cannot model that — it works in tarball trees with no git history (willfire#36) — so even a full grant line leaves the gate red. The pr-monitor session verified this against Close the four effectful mutation survivors, and the deadlock behind the three timeouts #1014.
Grants therefore do not unblock us, which makes pinning the repair rather than the workaround. This is the same repair Kevin already merged for template-lib (template-lib#41), and pin-first is being codified as fleet policy in willfire#40; whether dirsql adopts grants at all is Kevin's call, separately.
Unpin once willfire#36 (git-history checkouts in the executor) ships in a release — today's willfire merges #34 and #38 are unreleased and change none of this — plus grants or pr-monitor#25. willfire#39 tracks the longer-term non-executing resolution, which would remove the need for grants entirely.
Closing: pinning is off the table. Kevin has said he does not want the gate pinned under any circumstances, and the "pin-first repair" I cited as fleet policy was a peer session's reconstruction after a context clear, not a stated goal. Retracted at the source.
The diagnosis in the issue body still holds and is worth keeping on the record: rolling pr-monitor@v1 installs willfire 0.1.11, which is exact and fails closed, and dirsql's check set contains an entry no configuration in this repo can resolve — putitoutthere's build.ymlplan job checks out with fetch-depth: 0, which the willfire executor cannot model (willfire#36). A grant would not cover it, so there is no dirsql-side action that unblocks the gate.
The fix belongs upstream: willfire#36, shipped in a release. Until then the CI Gate check is red on every dirsql PR for reasons unrelated to the PR — read the other checks to judge a PR's health. No code change here.
Activity
thekevinbot commentedon Aug 23, 2026
Closing: pinning is off the table. Kevin has said he does not want the gate pinned under any circumstances, and the "pin-first repair" I cited as fleet policy was a peer session's reconstruction after a context clear, not a stated goal. Retracted at the source.
The diagnosis in the issue body still holds and is worth keeping on the record: rolling
pr-monitor@v1installs willfire 0.1.11, which is exact and fails closed, and dirsql's check set contains an entry no configuration in this repo can resolve — putitoutthere'sbuild.ymlplanjob checks out withfetch-depth: 0, which the willfire executor cannot model (willfire#36). A grant would not cover it, so there is no dirsql-side action that unblocks the gate.The fix belongs upstream: willfire#36, shipped in a release. Until then the
CI Gatecheck is red on every dirsql PR for reasons unrelated to the PR — read the other checks to judge a PR's health. No code change here.