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
Failures in window: 10 (9 in a single PR cluster, 1 scheduled run already fixed)
Root cause (cluster A): AWF firewall image v0.25.47 ships with a broken Node bundle — oidc-token-provider.js requires ./oidc-token-provider-base which is not present in /app/, so awf-api-proxy crashes at startup and the agent is never invoked.
All 9 cluster-A runs share head_sha=4be2b6b9852ffe06efc8f59c2c5bdf1b5db12668 from branch copilot/bump-firewall-mcpg-versions (PR #32503, "Bump default AWF to v0.25.47 and MCP Gateway to v0.3.10"). The failure occurs before the agent runs — awf-api-proxy exits non-zero on startup with the same Node module-resolution error in every workflow.
Note: Smoke Pi, Smoke Codex, and Changeset Generator do not have individual auto-generated tracking issues; they share the same root cause as the linked sub-issues.
Evidence (cluster B)
Run §25947409109 (Daily Model Inventory Checker, scheduled, on main). The Copilot CLI spawned via copilot_harness.cjs exited with code 1 after ~1s, producing zero stdout/stderr — a silent startup crash. The harness flagged: no output produced — not retrying (possible causes: binary not found, permission denied, auth failure, or silent startup crash).
The spawn command included a deeply quoted --allow-tool 'shell(jq ".endpoints[] | select(.provider == \"copilot\") | .models" /tmp/gh-aw/model-inventory/reflect.json)', which matches the fix landed in commit f91a078 ("Stabilize Daily Model Inventory Copilot startup by removing fragile quoted jq allow-tool (#32505)"). The fix merged at 2026-05-16T01:04:37Z, after the 00:10Z failed run. No subsequent Daily Model Inventory run exists in the window to confirm the fix end-to-end, but the symptom matches exactly.
No prior open agentic-workflows issue describes the AWF v0.25.47 bundle bug as an upstream/release problem — only per-workflow symptom reports.
Cluster B has no separate tracking issue; the fix has already landed.
Proposed fix roadmap
P0 — Block ✅ already in place: the smoke gate caught the bad bump, PR #32503 closed without merging. No action.
P1 — Upstream fix needed before retrying v0.25.47:
File or escalate an upstream issue against the AWF firewall image build pipeline: the oidc-token-provider-base.js (or its compiled counterpart) is missing from /app/ in the published v0.25.47 container image. The require call at /app/oidc-token-provider.js:22 resolves to nothing.
Until upstream resolves, do not re-attempt the bump to v0.25.47. If a future PR retries the bump without upstream proof of fix, expect the same cluster of smoke failures.
P2 — Verification:
Confirm that the next scheduled Daily Model Inventory Checker run (post f91a078) succeeds on main. If it fails again with zero stdout/stderr, re-open cluster B as a separate investigation.
Sub-issues linked
The 6 auto-generated per-workflow failure issues are linked as sub-issues of this report. They are kept open so each affected workflow's per-engine surface (copilot, claude, gemini) remains visible; closing is deferred to maintainers once the upstream image is confirmed fixed.
Confidence and unknowns
High confidence that all 9 cluster-A failures share the same root cause (identical stack trace, identical SHA, identical container image).
High confidence that the failure is in the upstream image, not in gh-aw code (the missing file is inside /app/ of awf-api-proxy, which is the published AWF v0.25.47 container).
Medium confidence that cluster B is fully resolved by f91a078 — symptom matches the fix description but no post-fix run exists yet to verify.
Unknown: whether the upstream image bug has already been reported externally.
Summary
oidc-token-provider.jsrequires./oidc-token-provider-basewhich is not present in/app/, soawf-api-proxycrashes at startup and the agent is never invoked.mainis not currently affected. The smoke gate worked as designed.f91a078(PR Stabilize Daily Model Inventory Copilot startup by removing fragile quoted jq allow-tool #32505), merged at 2026-05-16T01:04:37Z — after the failed run started at 00:10Z.Failure clusters
MODULE_NOT_FOUND4be2b6b9(PR #32503, closed)890db390(main, scheduled)f91a078(#32505)Evidence (cluster A)
All 9 cluster-A runs share
head_sha=4be2b6b9852ffe06efc8f59c2c5bdf1b5db12668from branchcopilot/bump-firewall-mcpg-versions(PR #32503, "Bump default AWF to v0.25.47 and MCP Gateway to v0.3.10"). The failure occurs before the agent runs —awf-api-proxyexits non-zero on startup with the same Node module-resolution error in every workflow.Representative stack trace (Smoke OTEL, [§25948486975](https://github.com/github/gh-aw/actions/runs/25948486975))
All 9 failed runs in cluster A
Note: Smoke Pi, Smoke Codex, and Changeset Generator do not have individual auto-generated tracking issues; they share the same root cause as the linked sub-issues.
Evidence (cluster B)
Run §25947409109 (Daily Model Inventory Checker, scheduled, on
main). The Copilot CLI spawned viacopilot_harness.cjsexited with code 1 after ~1s, producing zero stdout/stderr — a silent startup crash. The harness flagged:no output produced — not retrying (possible causes: binary not found, permission denied, auth failure, or silent startup crash).The spawn command included a deeply quoted
--allow-tool 'shell(jq ".endpoints[] | select(.provider == \"copilot\") | .models" /tmp/gh-aw/model-inventory/reflect.json)', which matches the fix landed in commitf91a078("Stabilize Daily Model Inventory Copilot startup by removing fragile quoted jq allow-tool (#32505)"). The fix merged at 2026-05-16T01:04:37Z, after the 00:10Z failed run. No subsequent Daily Model Inventory run exists in the window to confirm the fix end-to-end, but the symptom matches exactly.Existing issue correlation
Cannot find module './oidc-token-provider-base'error against PR Bump default AWF to v0.25.47 and MCP Gateway to v0.3.10 #32503's run IDs. They are all symptoms of this single upstream image bug. They are linked below as sub-issues.agentic-workflowsissue describes the AWF v0.25.47 bundle bug as an upstream/release problem — only per-workflow symptom reports.Proposed fix roadmap
P0 — Block ✅ already in place: the smoke gate caught the bad bump, PR #32503 closed without merging. No action.
P1 — Upstream fix needed before retrying v0.25.47:
oidc-token-provider-base.js(or its compiled counterpart) is missing from/app/in the publishedv0.25.47container image. The require call at/app/oidc-token-provider.js:22resolves to nothing.P2 — Verification:
f91a078) succeeds onmain. If it fails again with zero stdout/stderr, re-open cluster B as a separate investigation.Sub-issues linked
The 6 auto-generated per-workflow failure issues are linked as sub-issues of this report. They are kept open so each affected workflow's per-engine surface (
copilot,claude,gemini) remains visible; closing is deferred to maintainers once the upstream image is confirmed fixed.Confidence and unknowns
/app/ofawf-api-proxy, which is the published AWF v0.25.47 container).f91a078— symptom matches the fix description but no post-fix run exists yet to verify.References:
f91a078, merged)