auto-rebase: unguarded conflict-comment aborts the whole run when a PR hits the 2500-comment cap — blocks the ring1→stable soak gate

Summary

A single comment-capped PR aborts every auto-rebase run on a repo. auto-rebase-reusable.yml's merge-conflict branch posts a conflict-resolution comment via gh pr comment (line ~235) without an error guard. When the target PR has hit GitHub's 2500-comment cap, that command exits non-zero and — because the step runs under bash -e (shell: /usr/bin/bash -e {0}) — the entire step fails, even though other PRs were rebased successfully earlier in the same run.

Evidence

petry-projects/bmad-bgreat-suite PR #205 (CI Failure Analyst…, OPEN) is at exactly 2500 comments. Every auto-rebase run that reaches it dies:

PR #205 (add/ci-failure-analyst-v2) is 8 commit(s) behind main — updating branch
  Update failed: {"message":"merge conflict between base and head", ... "status":"422"}
  Posting conflict resolution request
GraphQL: Commenting is disabled on issues with more than 2500 comments (addComment)
##[error]Process completed with exit code 1.

4 runs failed with this identical signature in the auto-rebase v2.1.0 soak window (all on bmad-bgreat-suite): 28484669277, 28481353704, 28314060392, 28311198868. PRs #341/#335/#334 were updated fine in the same runs — only the #205 comment killed them.

Impact

This is the canary-rollout blocker for auto-rebase (initiative #495, gate standard #548). The gate correctly classifies it BLOCKED / PRE_EXISTING (cum_fail=4, reusable byte-identical to prior), so it will not roll back — but auto-rebase can never reach stable while a capped PR keeps poisoning the ring1 (bmad) runs. Same failure class as the pr-review-mention comment-cap loop (#553).

Root cause

auto-rebase-reusable.yml line ~235:

gh pr comment "$PR_NUMBER" --repo "$REPO" --body "$CONFLICT_BODY"

The sibling delete on line ~217 is already guarded (... || true); the post is not. A best-effort conflict notification must not be fatal to the core function (rebasing the other PRs).

Fix (primary — the reusable)

Guard the post so a comment-side failure is logged, not fatal (mirrors #553):

gh pr comment "$PR_NUMBER" --repo "$REPO" --body "$CONFLICT_BODY" \
  || echo "::warning::could not post conflict comment on #$PR_NUMBER (comment cap / API error) — continuing"

Introduced by the #482 reusables initiative (auto-rebase-reusable.yml). Add a regression test to auto-rebase-tests.yml stubbing a non-zero gh pr comment and asserting the run still succeeds and continues to the next PR.

Secondary (the capped PR)

bmad-bgreat-suite #205 has 2500 comments and cannot be commented on again. Surface to the owner to resolve/close #205 (and audit what generated 2500 comments there). Not something the release automation should mutate.

Enhancement — auto-triage this class in the release workflow

The canary-rollout gate already classifies REGRESSION vs PRE_EXISTING and supports a per-agent benign_failure_classes allowlist (#1025 P2), but that matches on failed-step name only — too coarse here (the step name Update behind non-Dependabot PRs also covers real failures). Proposal: extend the gate's failure triage to inspect the log signature (error text) of an in-window failure and auto-classify known-environmental patterns — more than 2500 comments, secondary-rate-limit, HTTP 5xx — as benign/environmental, distinct from a genuine step failure. That would let a comment-cap failure be excluded from cum_fail automatically (defense-in-depth) even before the source fix lands, without masking real regressions. Tracked against canary-rollout #495 / soak standard #548.

Refs

#482 (introducing initiative) · #553 (precedent comment-cap fix) · #495 / #548 (canary-rollout + soak standard) · petry-projects/.github-private #1045 (promote path that surfaced this).

Dev-Lead Implementation Plan

Issue: #594 — auto-rebase: unguarded conflict-comment aborts the whole run when a PR hits the 2500-comment cap

Scope

Make the merge-conflict resolution comment in auto-rebase-reusable.yml best-effort so a comment-side failure (GitHub's 2500-comment cap, secondary rate limit, transient 5xx) is logged and swallowed instead of killing the whole bash -e step, letting the loop continue rebasing the other PRs.

Implementation Checklist

Add .github/scripts/auto-rebase/lib/comments.sh with a testable auto_rebase_post_comment_best_effort helper (posts via gh pr comment; on failure emits ::warning::… — continuing and returns 0) — mirrors the extract-logic-to-scripts convention so it can be bats-tested instead of trapped inline in YAML
Source comments.sh in .github/workflows/auto-rebase-reusable.yml and route the conflict post (line ~235) through the helper (fix mirrors fix(pr-review-mention): break the ack self-comment recursion #553)
Add comments.sh to the shellcheck step in .github/workflows/auto-rebase-tests.yml
Document the helper in .github/scripts/auto-rebase/README.md

Tests to Write

test/workflows/auto-rebase/comments.bats: stub a non-zero gh pr comment → helper returns 0 (non-fatal) and logs the ::warning::… continuing line
success case: stub gh exit 0 → returns 0, posts the comment, no warning
regression: a bash -e loop over two PRs where the first PR's comment fails still reaches and processes the second PR

Test Command

bats --print-output-on-failure test/workflows/auto-rebase/ (plus shellcheck -x on the auto-rebase libs and bash .dev-lead/scripts/dev-lead-lint.sh)

Note on scope

Primary (reusable) fix only. The capped PR (bmad-bgreat-suite #205) and the release-workflow log-signature auto-triage enhancement are separate follow-ups called out in the issue and are out of scope here.

Dev-Lead: Implementation Complete

Plan Execution

Added .github/scripts/auto-rebase/lib/comments.sh with best-effort auto_rebase_post_comment_best_effort helper
Routed the conflict-resolution post (reusable line ~235) through the helper — a comment-side failure is now logged, not fatal (mirrors fix(pr-review-mention): break the ack self-comment recursion #553)
Added comments.sh to the shellcheck step in auto-rebase-tests.yml
Documented the helper + rationale in the auto-rebase README
TDD regression tests written first (red), then green

Test Results

1..22
ok 1 post_comment_best_effort: comment-cap failure is non-fatal (returns 0)
ok 2 post_comment_best_effort: failure logs a continue warning naming the PR
ok 3 post_comment_best_effort: success returns 0, posts the comment, no warning
ok 4 post_comment_best_effort: a capped PR does not abort a bash -e loop over multiple PRs
ok 5 has_current_approval: single APPROVED review counts as approved
ok 6 has_current_approval: no reviews is not approved
ok 7 has_current_approval: only COMMENTED reviews are not approved
ok 8 has_current_approval: reviewer who approved then requested changes is not approved
ok 9 has_current_approval: reviewer who requested changes then approved is approved
ok 10 has_current_approval: a later COMMENTED review does not cancel an approval
ok 11 has_current_approval: a dismissed approval is not approved
ok 12 has_current_approval: one approver among several reviewers counts
ok 13 has_ready_label: label present
ok 14 has_ready_label: label absent
ok 15 has_ready_label: empty labels array
ok 16 pr_eligible review-ready: non-draft + approved is eligible
ok 17 pr_eligible review-ready: non-draft + ready label is eligible
ok 18 pr_eligible review-ready: non-draft but neither approved nor labelled is ineligible
ok 19 pr_eligible review-ready: draft is ineligible even when approved
ok 20 pr_eligible all: always eligible (legacy fan-out escape hatch)
ok 21 pr_eligible all: eligible even for drafts (preserves prior behavior)
ok 22 pr_eligible: unknown mode errors out (exit 2)

Also green: shellcheck -x on both auto-rebase libs (CI parity) and bash .dev-lead/scripts/dev-lead-lint.sh (all checks passed); both workflow YAMLs parse.

Files Changed

.github/scripts/auto-rebase/lib/comments.sh (new): auto_rebase_post_comment_best_effort PR_NUMBER REPO BODY — posts via gh pr comment; on any failure (2500-comment cap / secondary rate limit / transient 5xx) emits ::warning::… — continuing and returns 0.
.github/workflows/auto-rebase-reusable.yml: source comments.sh; the conflict post now calls the helper instead of a bare gh pr comment, so it can no longer abort the bash -e step and starve the other PRs of a rebase.
test/workflows/auto-rebase/comments.bats (new): stub a non-zero gh pr comment → non-fatal + warning; success case; and a bash -e loop over two PRs where the first (capped) PR's comment fails yet the loop still reaches the second PR.
.github/workflows/auto-rebase-tests.yml: shellcheck now also covers comments.sh.
.github/scripts/auto-rebase/README.md: documents the new helper and the auto-rebase: unguarded conflict-comment aborts the whole run when a PR hits the 2500-comment cap — blocks the ring1→stable soak gate #594 rationale.

Notes

Scope kept to the primary (reusable) fix. Out of scope per the issue: the capped PR bmad-bgreat-suite Implement Deepsec Org Scan #205 (owner must resolve/close it) and the release-workflow log-signature auto-triage enhancement (docs(standards): add Advanced Security (GHAS) org standard #495 / Define the release soak standard (ring-promotion gate): time floor + traffic-scaled sample + health #548).
Possible follow-up: the sibling blocked-comment post (reusable line ~194, workflows-permission branch) is the same unguarded class but was not in this issue's scope; it could reuse the same helper in a future change.

Second instance — same root cause blocks `pr-review-mention` too

The canary gate also holds pr-review-mention at ring1→stable (cum_fail=1), and it's the identical comment-cap class:

Run 28318567437 on petry-projects/bmad-bgreat-suite, step Post acknowledgement comment:

GraphQL: Commenting is disabled on issues with more than 2500 comments (addComment)
##[error]Process completed with exit code 1.

#553 added the marker-based trigger guard to stop the self-comment loop, but the ack-post itself is still unguarded, so a comment-capped target PR makes it exit non-zero → the run fails.

So both remaining #482 ring1→stable blockers are one bug class (unguarded best-effort comment under bash -e), both surfacing on bmad-bgreat-suite where PR #205 (and possibly others) sit at the 2500-comment cap:

auto-rebase-reusable.yml — conflict comment (~L235)
pr-review-mention reusable — Post acknowledgement comment step

Fix both: guard the gh pr comment/gh api ... addComment with || echo "::warning::…" so a comment-side failure is logged, not fatal. The proposed gate log-signature triage enhancement (auto-classify more than 2500 comments / rate-limit / 5xx as environmental) would unblock both reusables' soak automatically even before the source fixes land.

Once both reusables are guarded (and/or the enhancement lands), the canary timer (now armed, CANARY_AUTO_PROMOTE=true) will auto-promote auto-rebase + pr-review-mention to stable — completing the 6/6 #482 rollout with no manual step.

feat: implement issue #594 — auto-rebase: unguarded conflict-comment …

fix(pr-review-mention): guard ack-comment post against the 2500-comme…

Navigation Menu

Repository navigation

auto-rebase: unguarded conflict-comment aborts the whole run when a PR hits the 2500-comment cap — blocks the ring1→stable soak gate #594

Summary

Evidence

Impact

Root cause

Fix (primary — the reusable)

Secondary (the capped PR)

Enhancement — auto-triage this class in the release workflow

Refs

Dev-Lead Implementation Plan

Scope

Implementation Checklist

Tests to Write

Test Command

Note on scope

Dev-Lead: Implementation Complete

Plan Execution

Test Results

Files Changed

Notes

Second instance — same root cause blocks `pr-review-mention` too

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

auto-rebase: unguarded conflict-comment aborts the whole run when a PR hits the 2500-comment cap — blocks the ring1→stable soak gate #594

Description

Summary

Evidence

Impact

Root cause

Fix (primary — the reusable)

Secondary (the capped PR)

Enhancement — auto-triage this class in the release workflow

Refs

Activity

don-petry commented on Jul 3, 2026

Dev-Lead Implementation Plan

Scope

Implementation Checklist

Tests to Write

Test Command

Note on scope

don-petry commented on Jul 3, 2026

Dev-Lead: Implementation Complete

Plan Execution

Test Results

Files Changed

Notes

don-petry commented on Jul 3, 2026

Second instance — same root cause blocks pr-review-mention too

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Second instance — same root cause blocks `pr-review-mention` too