Recompile every workflow lock file on main and gate merges on check-workflow-drift — committed .lock.yml files carry a frontmatter hash that no longer matches their .md at HEAD, hard-failing 4+ workflows at the activation step before any agent runs.
Problem statement
Multiple agentic workflows fail immediately at the Check workflow lock file activation step with:
ERR_CONFIG: Lock file '.github/workflows/<name>.lock.yml' is outdated!
The workflow file '<name>.md' frontmatter has changed. Run 'gh aw compile' to regenerate the lock file.
The agent job is skipped; no work is done and no safe output is produced. This is not self-healing: the Copilot PR NLP Analysis run failed at 66c9b2f — the current main HEAD — so the drift is committed, not a stale-checkout artifact.
Failure cluster (last 6h)
| Workflow |
Run |
Branch / event |
SHA |
| Copilot PR Conversation NLP Analysis |
§29088572869 |
main / schedule |
66c9b2f (HEAD) |
| [aw] Failure Investigator (6h) |
§29078152775 |
main / schedule |
d520c64b |
| Auto-Triage Issues |
§29077454106 |
main / schedule |
d520c64b |
| Daily News |
§29084339057 |
main / schedule |
d520c64b |
| Impeccable Skills Reviewer |
§29086731450 |
PR / pull_request |
0fcba1de |
Parent #44684 tracks only the Failure Investigator instance; this issue captures the systemic blast radius across ≥4 workflows on main.
Probable root cause
Committed .lock.yml files were regenerated by a gh aw compiler version whose frontmatter-hash output differs from the version CI uses to validate — or a .md frontmatter edit landed without a matching recompile. .md and .lock.yml share a commit, so the lightweight scripts/check-stale-lock-files.sh gate (which only inspects working-tree/staged edits) does not catch this committed hash drift; only the recompile-based check-workflow-drift.sh does.
Proposed remediation
- Run
gh aw compile on main with the CI-pinned gh-aw version, commit the regenerated .lock.yml files.
- Wire
make check-workflow-drift (full recompile diff) into required CI so committed lock/hash drift blocks merge — the current check-stale-lock-files early gate misses it.
- Pin/align the
gh-aw compiler version between local recompiles and CI validation to stop hash-algorithm drift from re-introducing this.
Success criteria / verification
- Every agentic workflow passes the
Check workflow lock file activation step on main.
make check-workflow-drift is green in CI and required for merge.
- No new
ERR_CONFIG stale-lock activation failures in the next 6h investigator window.
Analyzed runs: 29088572869, 29078152775, 29077454106, 29084339057, 29086731450. Parent: #44684.
Related to #44684
Generated by 🔍 [aw] Failure Investigator (6h) · 205.7 AIC · ⌖ 17.7 AIC · ⊞ 5.2K · ◷
Recompile every workflow lock file on
mainand gate merges oncheck-workflow-drift— committed.lock.ymlfiles carry a frontmatter hash that no longer matches their.mdat HEAD, hard-failing 4+ workflows at the activation step before any agent runs.Problem statement
Multiple agentic workflows fail immediately at the
Check workflow lock fileactivation step with:The agent job is
skipped; no work is done and no safe output is produced. This is not self-healing: the Copilot PR NLP Analysis run failed at66c9b2f— the currentmainHEAD — so the drift is committed, not a stale-checkout artifact.Failure cluster (last 6h)
66c9b2f(HEAD)d520c64bd520c64bd520c64b0fcba1deParent #44684 tracks only the Failure Investigator instance; this issue captures the systemic blast radius across ≥4 workflows on
main.Probable root cause
Committed
.lock.ymlfiles were regenerated by agh awcompiler version whose frontmatter-hash output differs from the version CI uses to validate — or a.mdfrontmatter edit landed without a matching recompile..mdand.lock.ymlshare a commit, so the lightweightscripts/check-stale-lock-files.shgate (which only inspects working-tree/staged edits) does not catch this committed hash drift; only the recompile-basedcheck-workflow-drift.shdoes.Proposed remediation
gh aw compileonmainwith the CI-pinnedgh-awversion, commit the regenerated.lock.ymlfiles.make check-workflow-drift(full recompile diff) into required CI so committed lock/hash drift blocks merge — the currentcheck-stale-lock-filesearly gate misses it.gh-awcompiler version between local recompiles and CI validation to stop hash-algorithm drift from re-introducing this.Success criteria / verification
Check workflow lock fileactivation step onmain.make check-workflow-driftis green in CI and required for merge.ERR_CONFIGstale-lock activation failures in the next 6h investigator window.Analyzed runs: 29088572869, 29078152775, 29077454106, 29084339057, 29086731450. Parent: #44684.
Related to #44684