[aw-failures] [aw] Fix: APM unpack fails for apm-default.tar.gz (exit code 1) — 3 workflows blocked

[github-actions]

Problem Statement

The agent job fails during the Restore APM packages (all bundles) step on multiple workflows and PRs. The APM bundle downloads successfully, but the unpack step exits with code 1, aborting the run before the agent ever executes.

Affected Workflows and Run IDs

Run	Workflow	Branch	Firewall Version	Bundle Size
§25350390860	Matt Pocock Skills Reviewer	copilot/fix-ai-moderator-workflow-allowlist	v0.25.35	78,764 B
§25351475330	Smoke Claude	copilot/bump-firewall-to-v0-25-38-and-mcpg-to-v0-3-6	v0.25.38	3,468 B
§25351482890	Matt Pocock Skills Reviewer	copilot/bump-firewall-to-v0-25-38-and-mcpg-to-v0-3-6	v0.25.38	78,562 B

APM packages involved: mattpocock/skills (×2 runs, ~78 KB) and microsoft/apm-sample-package (×1 run, ~3.5 KB).

Consistent Error

##[error]APM action failed: apm unpack failed for bundle 1 of 1
  (path: /tmp/gh-aw/apm-bundles/apm-default.tar.gz, exit code: 1)

Fails in job agent, step Restore APM packages (all bundles). The apm-prep job succeeds in all cases.

Root Cause (Probable)

The bundle is downloaded to /tmp/gh-aw/apm-bundles/apm-default.tar.gz but fails decompression/validation. Key observations:

1. Both bundle sources fail — mattpocock/skills (~78 KB) and microsoft/apm-sample-package (~3.5 KB) both unpack with exit code 1, ruling out a single corrupted package.
2. Two firewall versions affected — v0.25.35 and v0.25.38, ruling out the firewall bump as cause.
3. Two different PRs affected — not branch-specific; this is systemic.
4. Audit-diff confirms no agent activity — Smoke Claude had 0 network requests (agent never ran), while Smoke Codex on the same PR completed normally with 16+ API calls.

Probable causes (in priority order):

• APM action version changed or has a bug in the unpack step
• Bundle format changed at source (incompatible with current unpack action)
• Filesystem permission issue in /tmp/gh-aw/apm-bundles/ inside the agent container

Proposed Remediation

1. Inspect the full stderr/stdout of the apm unpack step for the failing runs (not captured in audit artifacts).
2. Check if the APM action version was recently updated and whether that introduced the regression.
3. Manually attempt to unpack apm-default.tar.gz from a failing run's artifact to verify bundle integrity.
4. Verify write/execute permissions for /tmp/gh-aw/apm-bundles/ in the agent container context.

Success Criteria

• Smoke Claude and Matt Pocock Skills Reviewer complete without APM errors on affected PRs.
• agent job conclusion changes from failure → success.
• Audit shows >0 turns and network activity (agent actually executes).

References: Parent: #30251

• §25350390860 · §25351475330 · §25351482890

Generated by [aw] Failure Investigator (6h) · ● 313.4K · ◷

• expires on May 12, 2026, 1:29 AM UTC

[github-actions]

🍪 Issue Monster selected this for Copilot

I've identified this issue as a good candidate for automated resolution and requested assignment to the Copilot coding agent.

If assignment succeeds, the Copilot coding agent will analyze the issue and create a pull request with the fix.

Om nom nom! 🍪

🍪 Om nom nom by Issue Monster · ● 17.1K · ◷

[theletterf]

@pelikhan We're hitting the same bundle-unpack failure on gh-aw v0.71.4 with elastic/elastic-docs-skills packages, in a private caller repo.

• Run: https://github.com/elastic/docs-content-internal/actions/runs/25368618064
• Caller: elastic/docs-content-internal (private), reusable workflow from elastic/docs-actions@v1 (release 1.15.2)
• Affected packages (one per workflow):
  • elastic/elastic-docs-skills/skills/review/frontmatter-audit + elastic/elastic-docs-skills/skills/authoring/frontmatter-description
  • elastic/elastic-docs-skills/skills/authoring/applies-to-tagging
  • elastic/elastic-docs-skills/skills/authoring/page-opening-optimizer
  • elastic/elastic-docs-skills/skills/review/docs-check-style

Same shape as #30252 — apm-prep succeeds, agent step fails on Restore APM packages (all bundles) with:

[x] Bundle verification failed -- the following deployed files are missing from
    the bundle:
      - .agents/skills/applies-to-tagging
##[error]APM action failed: apm unpack failed for bundle 1 of 1
   (path: /tmp/gh-aw/apm-bundles/6fd13e92-apm-default.tar.gz, exit code: 1)

Each of the four APM-importing workflows fails at the same step with its own skill name in the missing-files list. The three sibling workflows in the same orchestrator that don't import APM (typos, staleness, coherence) run to completion in parallel, which suggests the regression is contained to the APM unpack/verify path rather than something broader.

Happy to share more artifacts / repro on demand.