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
auto-rebase: roll the approval-survival ruleset fix out to the rest of the fleet #1438
Part of #1416 (AC2), blocked by the sibling spike/local-fix issue. Once the fix from that issue is confirmed working on this repo, apply the same pr-quality ruleset change to the other consumer repos currently carrying the dangerous dismiss_stale_reviews_on_push: true + require_last_push_approval: true combination: google-app-scripts, ContentTwin, incubator, TalkTerm, broodminder-export, bmad-bgreat-suite, repo-template, .github. (markets and broodly do not have this combination and don't need the change.)
Do not roll this out until the local fix in the blocking issue is verified — this is fleet-wide blast radius on a review-integrity setting.
Thanks for filing this rollout tracking issue — gating the fleet-wide change behind local verification is exactly the right call given the review-integrity blast radius. What's the acceptance signal on the blocking issue that will confirm the local fix is verified and safe to roll out, so this issue can be unblocked without ambiguity?
Dev-Lead: cannot proceed — blocking precondition not met (dev-lead:needs-human)
This is the fleet rollout tracker for AC2 of #1416. Its single hard gate — stated in this
issue's own body — is:
Do not roll this out until the local fix in the blocking issue is verified — this is
fleet-wide blast radius on a review-integrity setting.
That precondition is not satisfied, so there is nothing to roll out. I checked the live
state of the blocker before touching anything:
Why the gate is not cleared
The blocking issue #1437 closed carrying the dev-lead:needs-human label. It was a verification spike, and it resolved to the mandated escalation — no pr-quality ruleset
or workflow was changed on this repo (see docs/initiatives/auto-rebase-approval-survival-spike.md
and the #1437 completion comment). The decisive, live-verified finding:
dismiss_stale_reviews_on_push and require_last_push_approval are not separate rules —
they are parameters of one pull_request rule, bundled with required_approving_review_count, require_code_owner_review, and required_review_thread_resolution.
bypass_actors is a ruleset-level array with no per-rule/per-parameter bypass field.
So the narrow "stop erasing approvals for the auto-rebase identity" fix this rollout was meant to
propagate is not expressible in GitHub's ruleset model, and #1437 correctly stopped at dev-lead:needs-human rather than falling back to a fleet-wide relax. The epic's Decisions
(2026-08-02) record mandates exactly that halt: "if per-rule actor scoping isn't supported, do
not silently fall back to the fleet-wide relax — stop and flag dev-lead:needs-human so a human
picks the fallback explicitly." No human has since recorded an A/B/C decision on #1437 or #1416.
What rolling out now would actually do
There is no verified, committed fix artifact to replicate. The only "same change" available
today is adding the auto-rebase identity to each repo's pr-qualitybypass_actors — which, per
the finding above, would exempt automation from every PR-quality gate
(required_approving_review_count, code-owner review, thread resolution) across google-app-scripts, ContentTwin, incubator, TalkTerm, broodminder-export, bmad-bgreat-suite, repo-template, and .github. That is the explicitly non-chosen
fleet-wide review-integrity relaxation (option C in the spike), not a sanctioned fix.
Implementing it would violate both this issue's gate and #1416's decision.
Action taken
No source, ruleset, workflow, or fleet change made — correct outcome per the gate.
Unblock signal (answers the automated triage question on this issue)
This tracker becomes actionable only after a human, on #1437 / #1416, explicitly picks a
fallback and it is verified:
A — run the live approval-survival spike (§2 of the spike doc); if the approval survives
a merge-method update-branch, there is nothing to fix and this rollout issue can be closed
as not-needed.
B — split dismiss_stale_reviews_on_push / require_last_push_approval into their own
ruleset and grant the auto-rebase identity a bypass on only that ruleset (likely first
requires giving auto-rebase a dedicated app/PAT identity — it runs as generic GITHUB_TOKEN
today). Only then is there a concrete, verified change to replicate across the 8 repos, and this
issue should carry the exact per-repo ruleset spec before pickup.
Precondition resolved — proceeding directly rather than via dev-lead, since this is a repo-settings/ruleset mutation (not a code PR) across 9 repos.
Applied the fleet-wide relax (dismiss_stale_reviews_on_push: false, require_last_push_approval: false on the pr-quality ruleset's pull_request rule) to all 9 repos carrying the combination: .github-private, google-app-scripts, ContentTwin, incubator, TalkTerm, broodminder-export, bmad-bgreat-suite, repo-template, .github. Verified before/after on each — only the two target parameters changed; required_approving_review_count, require_code_owner_review, required_review_thread_resolution, and allowed_merge_methods are untouched.
Activity
github-actions commentedon Aug 2, 2026
Thanks for filing this rollout tracking issue — gating the fleet-wide change behind local verification is exactly the right call given the review-integrity blast radius. What's the acceptance signal on the blocking issue that will confirm the local fix is verified and safe to roll out, so this issue can be unblocked without ambiguity?
don-petry commentedon Aug 2, 2026
Dev-Lead: cannot proceed — blocking precondition not met (
dev-lead:needs-human)This is the fleet rollout tracker for AC2 of #1416. Its single hard gate — stated in this
issue's own body — is:
That precondition is not satisfied, so there is nothing to roll out. I checked the live
state of the blocker before touching anything:
Why the gate is not cleared
The blocking issue #1437 closed carrying the
dev-lead:needs-humanlabel. It was averification spike, and it resolved to the mandated escalation — no
pr-qualityrulesetor workflow was changed on this repo (see
docs/initiatives/auto-rebase-approval-survival-spike.mdand the #1437 completion comment). The decisive, live-verified finding:
dismiss_stale_reviews_on_pushandrequire_last_push_approvalare not separate rules —they are parameters of one
pull_requestrule, bundled withrequired_approving_review_count,require_code_owner_review, andrequired_review_thread_resolution.bypass_actorsis a ruleset-level array with no per-rule/per-parameter bypass field.So the narrow "stop erasing approvals for the auto-rebase identity" fix this rollout was meant to
propagate is not expressible in GitHub's ruleset model, and #1437 correctly stopped at
dev-lead:needs-humanrather than falling back to a fleet-wide relax. The epic's Decisions(2026-08-02) record mandates exactly that halt: "if per-rule actor scoping isn't supported, do
not silently fall back to the fleet-wide relax — stop and flag
dev-lead:needs-humanso a humanpicks the fallback explicitly." No human has since recorded an A/B/C decision on #1437 or #1416.
What rolling out now would actually do
There is no verified, committed fix artifact to replicate. The only "same change" available
today is adding the auto-rebase identity to each repo's
pr-qualitybypass_actors— which, perthe finding above, would exempt automation from every PR-quality gate
(
required_approving_review_count, code-owner review, thread resolution) acrossgoogle-app-scripts,ContentTwin,incubator,TalkTerm,broodminder-export,bmad-bgreat-suite,repo-template, and.github. That is the explicitly non-chosenfleet-wide review-integrity relaxation (option C in the spike), not a sanctioned fix.
Implementing it would violate both this issue's gate and #1416's decision.
Action taken
dev-lead:needs-humanand leaving this issue open (mirrors auto-rebase: run approval-survival spike; fix this repo's pr-quality ruleset if confirmed #1437).Unblock signal (answers the automated triage question on this issue)
This tracker becomes actionable only after a human, on #1437 / #1416, explicitly picks a
fallback and it is verified:
a merge-method
update-branch, there is nothing to fix and this rollout issue can be closedas not-needed.
dismiss_stale_reviews_on_push/require_last_push_approvalinto their ownruleset and grant the auto-rebase identity a bypass on only that ruleset (likely first
requires giving auto-rebase a dedicated app/PAT identity — it runs as generic
GITHUB_TOKENtoday). Only then is there a concrete, verified change to replicate across the 8 repos, and this
issue should carry the exact per-repo ruleset spec before pickup.
Once one of those is chosen and verified on this repo, re-label this issue for dev-lead pickup
with the concrete change to propagate.
don-petry commentedon Aug 3, 2026
Precondition resolved — proceeding directly rather than via dev-lead, since this is a repo-settings/ruleset mutation (not a code PR) across 9 repos.
Applied the fleet-wide relax (
dismiss_stale_reviews_on_push: false,require_last_push_approval: falseon thepr-qualityruleset'spull_requestrule) to all 9 repos carrying the combination: .github-private, google-app-scripts, ContentTwin, incubator, TalkTerm, broodminder-export, bmad-bgreat-suite, repo-template, .github. Verified before/after on each — only the two target parameters changed;required_approving_review_count,require_code_owner_review,required_review_thread_resolution, andallowed_merge_methodsare untouched.Closing.