fix(ci): run product-invariants on main pushes with the merge-commit body (#9744)

[aryanorastar]

Fixes #9744

What changed and why

repo-checks.yml ran the deterministic check manifest on main pushes with --skip-pr-body-checks, so product-invariants (and failure-class-protocol) were silently skipped on every post-merge run — even though GitHub folds the PR description, including locked invariant citations like INV-CHAT-1, into the merge commit message. A main-push Hygiene run could not verify citations at all; #9734 merged with the IDs in its body yet the post-merge run reported them missing.

Now the main-push path captures the merge-commit body (git log -1 --format=%B HEAD) and passes it through --pr-body-file, so body-requiring checks run with the real citations. Fail-closed is preserved: a direct push whose commit message omits required invariant IDs still fails.

Product invariants affected

None (CI workflow + test wiring only). This fixes the enforcement of INV-* citations on main pushes.

How it was verified

• actionlint .github/workflows/repo-checks.yml — clean.
• python3 .github/scripts/test_run_checks.py — the two wiring regression guards updated/passing; only pre-existing sparse-checkout failures remain (identical on pristine main).

Tests

• test_ci_lane_is_reachable_from_repo_checks — now asserts the main-push path passes --pr-body-file with the merge-commit body and no longer uses --skip-pr-body-checks (the wiring regression guard this issue asks for).
• test_main_push_includes_pr_body_checks_when_body_supplied / test_main_push_without_body_still_excludes_pr_body_checks — body-supplied runs select product-invariants; body-less runs stay fail-closed.

Failure class (fixes)

Failure-Class: none

Scoped cleanups

None.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[cubic-dev-ai]

No issues found across 2 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}

[Git-on-my-level]

Thanks for fixing this — the change is tightly scoped and the direction looks right to me.

I reviewed the workflow/script diff statically. Passing the main-push merge commit body into run_checks.py --pr-body-file should restore the body-required manifest checks on post-merge Hygiene runs, and the added regression tests cover both sides of the behavior: body supplied selects product-invariants / failure-class-protocol, while body-less runs still skip those PR-body-required checks.

Because this changes CI workflow behavior, I’m leaving it for human maintainer review rather than treating it as an automated approval. I don’t see a concrete blocker from this review, but a maintainer should confirm this is the desired post-merge enforcement behavior for direct pushes and merge commits before merging.

---

by AI on behalf of David — if you need David’s attention urgently, please @Git-on-my-level and escalate with need human response.

[aryanorastar]

Independently verified the final head (ff04f08):

• actionlint: clean on repo-checks.yml (no new findings).
• Contract tests: swapped the branch's repo-checks.yml + test_run_checks.py into the working tree and ran the Repo Checks drops merged PR invariant citations on main pushes #9744 regression suite → 4 passed (0 failed), covering both sides: test_main_push_includes_pr_body_checks_when_body_supplied (body-supplied → product-invariants/failure-class-protocol run) and test_main_push_without_body_still_excludes_pr_body_checks (no body → fail-closed, excluded). Working tree restored clean afterward.
• Fix is correct: main pushes now capture the merge commit body (git log -1 --format=%B HEAD) and pass it via --pr-body-file instead of --skip-pr-body-checks. GitHub folds the PR description (with INV-* citations) into merge commits, so product-invariants can now see the citations on post-merge Hygiene runs. The pull_request lane and --skip-pr-body-checks behavior are untouched.
• Fail-closed preserved: body-less runs still exclude PR-body-required checks.
• CI green on this head (17 success / 11 skipped).

@cubic-dev-ai's 5/5 is correct; this is a tightly scoped workflow wiring fix. @Git-on-my-level confirmed no concrete blocker and CI is green.

This restores post-merge invariant-citation enforcement on main pushes. Requesting your human maintainer sign-off on the CI workflow behavior change (needs-maintainer-review / workflow-review).

[cubic-dev-ai]

Independently verified the final head (ff04f08):

• actionlint: clean on repo-checks.yml (no new findings).
• Contract tests: swapped the branch's repo-checks.yml + test_run_checks.py into the working tree and ran the Repo Checks drops merged PR invariant citations on main pushes #9744 regression suite → 4 passed (0 failed), covering both sides: test_main_push_includes_pr_body_checks_when_body_supplied (body-supplied → product-invariants/failure-class-protocol run) and test_main_push_without_body_still_excludes_pr_body_checks (no body → fail-closed, excluded). Working tree restored clean afterward.
• Fix is correct: main pushes now capture the merge commit body (git log -1 --format=%B HEAD) and pass it via --pr-body-file instead of --skip-pr-body-checks. GitHub folds the PR description (with INV-* citations) into merge commits, so product-invariants can now see the citations on post-merge Hygiene runs. The pull_request lane and --skip-pr-body-checks behavior are untouched.
  ...

@aryanorastar The change is sound and I'm comfortable signing off from the review side. The wiring matches what the issue asked for: main-push runs now feed git log -1 --format=%B HEAD through --pr-body-file instead of skipping body-required checks wholesale, so product-invariants and failure-class-protocol get the citations GitHub folds into merge commits. The two new regression guards cover both sides (body supplied → checks selected; body-less → still excluded), and removing --skip-pr-body-checks from the workflow with the assertNotIn guard in test_ci_lane_is_reachable_from_repo_checks correctly prevents the silent-skip regression from creeping back.

One thing worth keeping straight, since you asked for sign-off: the "human maintainer" gate here is separate from my review. I can confirm there's no blocker from the code/diff side, but Git-on-my-level already flagged this as a CI workflow behavior change that needs human maintainer review — that approval has to come from a maintainer, not from me or CI. My review sign-off doesn't substitute for the needs-maintainer-review gate.

Everything else checks out: pull_request lane untouched, fail-closed behavior preserved for direct pushes without citations, and the verification you ran (actionlint clean, 4/4 regression tests passing, CI green) matches what I'd expect from the diff. No outstanding concerns.

_{Tip: get faster answers by chatting with cubic’s review copilot in the review UI}

[aryanorastar]

@Git-on-my-level this CI workflow fix has your positive signal and cubic's 5/5 with no issues — requesting your maintainer sign-off on the needs-maintainer-review gate. Ready to merge when you have a moment.

[Git-on-my-level]

Re-reviewed the current head (2f3504106baaff0b5d270a935a30c2f98debc11d). I still don’t see a code-level blocker: the main-push workflow now feeds the merge commit body through run_checks.py --pr-body-file, the PR lane remains separate, and the regression tests cover both the body-supplied path (product-invariants / failure-class-protocol selected) and the body-less fail-closed path.

I also ran the full .github/scripts/test_run_checks.py suite locally on this head: 41 tests passed.

Keeping this as human maintainer review rather than an automated approval because it changes CI workflow behavior. The existing needs-maintainer-review / workflow-review labels are still the right gate; no additional blocker from this review.

---

by AI on behalf of David — if you need David’s attention urgently, please @Git-on-my-level and escalate with need human response.

[aryanorastar]

@Git-on-my-level need human response — this is still blocked on code-owner review for CI/workflow behavior.

Your latest AI pass on 2f350410 was again a positive signal (no code-level blocker): main-push feeds the merge-commit body through run_checks.py --pr-body-file, PR lane stays separate, regression tests cover body-supplied + body-less fail-closed, and you ran full test_run_checks.py (41 passed). Cubic is 5/5. Current head is CI green (17 pass).

What’s left is the formal human / workflow-review / needs-maintainer-review sign-off, or an explicit hold if something else is blocking. Happy to follow up immediately either way.

[aryanorastar]

@Git-on-my-level follow-up: PR #10965 remains CI-green at 2f35041 with your latest positive re-review and 41/41 local contract tests passed. The only remaining blocker is the required workflow code-owner approval; no code or CI changes are pending.

[aryanorastar]

@Git-on-my-level need human response — workflow code-owner APPROVE still the only merge blocker on #10965.

Refreshed tip merges current main onto the main-push --pr-body-file wiring (no conflicts). Still:

• main-push captures git log -1 --format=%B HEAD → --pr-body-file (not --skip-pr-body-checks)
• PR lane untouched; body-less stays fail-closed
• regression guards cover body-supplied + body-less paths
• Failure-Class: none

Your AI re-review on 2f350410 found no code-level blocker (41/41 test_run_checks.py). Cubic 5/5. Please human / workflow-review APPROVE (or explicit hold) when you can.

[aryanorastar]

@Git-on-my-level need human response — Desktop Swift red on prior tip was the same stale glass tripwires as #11142/#10964 (not this workflow wiring).

Merged current main on tip 16475fba7330. Main-push --pr-body-file fix unchanged:

• git log -1 --format=%B HEAD → --pr-body-file
• PR lane untouched; body-less fail-closed
• Failure-Class: none

Please human / workflow-review APPROVE (or explicit hold) when CI re-greens.

[aryanorastar]

@Git-on-my-level need human response — #10965 still blocked only on CI/workflow code-owner APPROVE.

Author-clear on tip 16475fba7330:

• CI green (Desktop Swift after main refresh cleared glass tripwires)
• MERGEABLE
• Your AI re-reviews: no code-level blocker (--pr-body-file from merge-commit body; PR lane untouched; body-less fail-closed; 41/41 test_run_checks.py)
• Cubic 5/5; Failure-Class: none
• No author tip pending

Please human / workflow-review / needs-maintainer-review APPROVE (or explicit hold) when you can.

[Git-on-my-level]

Thanks for keeping this fix focused. I rechecked the current head and don’t see a code-level blocker.

Specific notes from the changed files:

• .github/workflows/repo-checks.yml now keeps the PR lane separate and changes only the main-push Hygiene path: it writes git log -1 --format=%B HEAD to PR_BODY_FILE and passes that to run_checks.py --lane ci --pr-body-file, so product-invariants / failure-class-protocol can inspect the merge-commit text instead of being skipped wholesale. The workflow checkout in this job still uses fetch-depth: 0, which supports reading the merge commit body.
• .github/scripts/test_run_checks.py updates the CI reachability guard to reject --skip-pr-body-checks in repo-checks.yml, assert the git log -1 --format=%B HEAD / --pr-body-file wiring, and adds coverage for both body-supplied selection (product-invariants, failure-class-protocol, diff-hygiene) and the existing body-less skip behavior.

CI is green on this head. Because this changes CI/workflow enforcement behavior, the existing workflow-review / needs-maintainer-review gate still needs human workflow maintainer sign-off before merge; from this review, the implementation itself looks sound.

---

by AI on behalf of David — if you need David’s attention urgently, please @Git-on-my-level and escalate with need human response.

[aryanorastar]

@Git-on-my-level need human response — thanks for the fresh positive-signal recheck on tip 16475fba7330.

Author-clear / CI green / MERGEABLE. Your latest pass again found no code-level blocker (main-push --pr-body-file from merge-commit body; PR lane separate; body-less fail-closed; regression guards present).

Only remaining merge blocker is formal workflow code-owner APPROVE (workflow-review / needs-maintainer-review). Please APPROVE (or explicit hold) when you can — happy to act immediately either way.

[aryanorastar]

@Git-on-my-level workflow-review ping: current head is MERGEABLE and the product-invariants main-push guard has no author-side blocker. Please provide the required approval or state an explicit hold.

[aryanorastar]

@Git-on-my-level workflow CO ping: still MERGEABLE, author-side clear. Waiting maintainer APPROVE.

[aryanorastar]

@Git-on-my-level workflow CO ping (>1d): still MERGEABLE, CI green, author-clear. Please APPROVE or explicit hold.

[aryanorastar]

@Git-on-my-level Workflow CO APPROVE ping — tip 16475fba MERGEABLE; author-side clear. Please APPROVE when convenient.

[aryanorastar]

@Git-on-my-level @kodjima33 Workflow review ping: tip 16475fba is MERGEABLE with 35/35 checks green (positive-signal). Runs product-invariants on main pushes with merge-commit body (#9744). Ready for maintainer review/merge.

[undivisible]

Cline review agrees: main-push Hygiene now passes the merge-commit body via --pr-body-file so product-invariants can see citations, with fail-closed preserved when the body is empty. CI-only, tests updated. Merge.