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
Tests-fix fallback ladder rung 1 (CI-based temporary revert) has no interruption-safe checkpoint #84
Push a commit that reverts only the production fix while keeping the new/changed tests; confirm CI goes red naming exactly the new regression tests; then git reset --hard back to the verified fix commit and force-push, confirming CI goes green again.
There is no interruption-safe checkpoint between "push the revert" and "restore the fix." If the session is killed (host restart, OOM, timeout) between those two steps, a real, pushed commit is left on the PR branch shipping the pre-fix behavior — worse than "unverified," it's "verified broken and forgotten." A fresh/resuming session has to notice red CI, git show a suspicious-looking commit, and diff it against the prior good commit to work out that the fix (not the tests) needs restoring.
Confirmed on dmccoystephenson/medieval-factions-dev-loop-dev-loop: a 2026-07-29 gardener-dispatched cycle picked up an orphaned PR (Dans-Plugins/Medieval-Factions #1984) left mid-revert by a killed prior session, with no state signal beyond red CI and a TEMP: commit message.
Suggested action
Amend the ladder's rung 1 text:
Prefer a local, non-pushed technique (e.g. git checkout <merge-base-with-main> -- <src files>, run tests locally, then git checkout HEAD -- <src files> to restore) over the push-based CI technique whenever the local anchor can run at all — it never touches the pushed branch, so there is no window where the PR head is broken.
If the CI-based push technique is unavoidable (anchor genuinely unrunnable), require the revert commit's message to make recovery undoable by a future session without investigation, e.g.: TEMP: revert <fix-commit-sha> to prove regression tests fail — MUST be reverted before merge, see <verified-good-sha>
Add to Phase 1's orphaned-PR-continuation handling: a session resuming an orphaned PR should treat "red final CI run" + a TEMP:-prefixed HEAD commit as a specific, named signal to check for this exact stuck-mid-revert state before doing anything else — not just generic "CI is red, fix it."
Activity
Extend scratch-file workaround to dirs, harden revert-ladder rung 1 (#86