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
Centralized workflow auto-rebase.yml references the reusable but is not pinned to @v1 (org standard). Replace with the canonical stub from standards/workflows/auto-rebase.yml which delegates to petry-projects/.github/.github/workflows/auto-rebase-reusable.yml@v1.
Dev-Lead: Finding is a false positive — @v2 pin is intentional
After investigating, I determined the literal remediation (downgrade auto-rebase.yml to the canonical @v1 stub) must not be applied — it is a behavioral regression for this repo. I documented the intentional divergence instead.
Why @v1 is wrong here
I diffed auto-rebase-reusable.yml@v1 vs @v2 in petry-projects/.github:
@v1: on merge conflict, posts a static "resolve manually" comment. github.token only.
@v2: posts a base-SHA-keyed conflict sentinel that triggers dev-lead's agentic rebase (the rebase intent in dev-lead.yml), accepts the optional GH_PAT_WORKFLOWS secret, and does smart delete/re-post to avoid cancelling in-flight rebase runs.
This repo has infrastructure built directly on the @v2 behavior, already documented in AGENTS.md:
auto-rebase-retry.yml — "self-heals a failed Auto-rebase run … the gap left by the conflict-only sentinel path"
auto-rebase-health.yml — "quantifies the agentic-conflict-resolution rate (conflict-sentinel fires vs. dev-lead rebase responses)"
The @v2 pin is also locked by the regression guard tests/dev-lead/integration/test_auto_rebase_stub.py, added for issue #139. Downgrading to @v1 would (a) break the dev-lead agentic-rebase trigger, (b) invalidate auto-rebase-health.yml's core metric, and (c) fail the #139 test.
Root cause of the finding
The org canonical stubstandards/workflows/auto-rebase.yml still pins @v1; this repo (home of dev-lead itself) intentionally runs the newer @v2. The divergence was real but undocumented as an AGENTS.md exception, so the auto-generated audit flagged it.
What I changed
AGENTS.md: added an **Exception:** entry under "Workflow Files" documenting that auto-rebase.yml deliberately pins @v2 ahead of the canonical @v1, why (agentic-rebase infra + Compliance: non-stub-auto-rebase.yml #139 test), and that it should defer to the template once the org stub is bumped to @v2.
No workflow change, no test change.
Verification
$ bash .dev-lead/scripts/dev-lead-lint.sh
[lint] shellcheck on 67 file(s)...
[lint] validating 4 agent profile(s)...
[lint] all checks passed
$ python3 tests/dev-lead/integration/test_auto_rebase_stub.py
PASS: .github/workflows/auto-rebase.yml correctly references ...auto-rebase-reusable.yml@v2
$ python3 tests/dev-lead/integration/test_caller_permissions.py
PASS: caller template grants >= every permission the reusable jobs request.
Recommended follow-up (org side)
Bump the canonical stub in petry-projects/.githubstandards/workflows/auto-rebase.yml to @v2. Once done, this repo's exception should be removed and the stub re-synced verbatim. Until then, this finding should be suppressed/acknowledged for this repo.
Phase 2 note
TDD phase skipped — this is a documentation-only change with no observable behavioral change (the existing #139 test already enforces the @v2 invariant and continues to pass).
Activity
don-petry commentedon Jun 19, 2026
Dev-Lead: Finding is a false positive —
@v2pin is intentionalAfter investigating, I determined the literal remediation (downgrade
auto-rebase.ymlto the canonical@v1stub) must not be applied — it is a behavioral regression for this repo. I documented the intentional divergence instead.Why
@v1is wrong hereI diffed
auto-rebase-reusable.yml@v1vs@v2inpetry-projects/.github:@v1: on merge conflict, posts a static "resolve manually" comment.github.tokenonly.@v2: posts a base-SHA-keyed conflict sentinel that triggers dev-lead's agentic rebase (therebaseintent indev-lead.yml), accepts the optionalGH_PAT_WORKFLOWSsecret, and does smart delete/re-post to avoid cancelling in-flight rebase runs.This repo has infrastructure built directly on the
@v2behavior, already documented inAGENTS.md:auto-rebase-retry.yml— "self-heals a failed Auto-rebase run … the gap left by the conflict-only sentinel path"auto-rebase-health.yml— "quantifies the agentic-conflict-resolution rate (conflict-sentinel fires vs. dev-leadrebaseresponses)"The
@v2pin is also locked by the regression guardtests/dev-lead/integration/test_auto_rebase_stub.py, added for issue #139. Downgrading to@v1would (a) break the dev-lead agentic-rebase trigger, (b) invalidateauto-rebase-health.yml's core metric, and (c) fail the #139 test.Root cause of the finding
The org canonical stub
standards/workflows/auto-rebase.ymlstill pins@v1; this repo (home of dev-lead itself) intentionally runs the newer@v2. The divergence was real but undocumented as anAGENTS.mdexception, so the auto-generated audit flagged it.What I changed
AGENTS.md: added an**Exception:**entry under "Workflow Files" documenting thatauto-rebase.ymldeliberately pins@v2ahead of the canonical@v1, why (agentic-rebase infra + Compliance: non-stub-auto-rebase.yml #139 test), and that it should defer to the template once the org stub is bumped to@v2.No workflow change, no test change.
Verification
Recommended follow-up (org side)
Bump the canonical stub in
petry-projects/.githubstandards/workflows/auto-rebase.ymlto@v2. Once done, this repo's exception should be removed and the stub re-synced verbatim. Until then, this finding should be suppressed/acknowledged for this repo.Phase 2 note
TDD phase skipped — this is a documentation-only change with no observable behavioral change (the existing #139 test already enforces the
@v2invariant and continues to pass).feat: implement issue #795 — Compliance: non-stub-auto-rebase.yml (#804)