[aw-failures] Conclusion job brittle when GitHub returns HTTP 500/DNS errors — cascades to `setup_globals.cjs` not found

[github-actions]

Problem

During a transient GitHub infrastructure outage (16:28–16:49 UTC on 2026-04-23), the conclusion job in multiple workflows failed with a two-stage cascade:

1. Primary error: git checkout fails with HTTP 500 or DNS: Could not resolve host: github.com
2. Cascade error: Error: Cannot find module '/home/runner/work/_temp/gh-aw/actions/setup_globals.cjs'

The cascade happens because handle_agent_failure.cjs imports setup_globals.cjs from the checked-out repo. When the checkout fails, that module is unavailable — turning a transient infra blip into a hard, unrecoverable conclusion job failure.

Affected Runs (last 6h)

Workflow	Run ID	Time (UTC)	Error
Smoke CI	§24846689722	16:29	HTTP 500 + setup_globals.cjs not found
Smoke CI	§24846759388	16:31	HTTP 500 + setup_globals.cjs not found
Smoke CI	§24846939615	16:35	HTTP 500 + setup_globals.cjs not found
Smoke CI	§24847116637	16:39	HTTP 500 + setup_globals.cjs not found
Smoke CI	§24852137215	18:32	DNS failure + setup_globals.cjs not found
Test Quality Sentinel	§24846641687	16:28	HTTP 500 + setup_globals.cjs not found
Design Decision Gate 🏗️	§24847007622	16:37	HTTP 500 + setup_globals.cjs not found
Slide Deck Maintainer	§24847273895	16:43	HTTP 500 + setup_globals.cjs not found

Note: in most of these runs, the agent job succeeded (noop or valid output) — only the conclusion cleanup step failed.

Root Cause

The conclusion job performs a fresh git checkout to access AWF action helpers (e.g., setup_globals.cjs, handle_agent_failure.cjs). When github.com is transiently unreachable (HTTP 500 or DNS failure), this checkout fails. The error cascades because handle_agent_failure.cjs cannot load its dependency, preventing graceful failure reporting.

Proposed Remediation

Option A (Preferred): Add retry with backoff to the conclusion job's git checkout

• 3 attempts with exponential backoff (e.g., 5s, 15s, 30s)
• Covers transient HTTP 500 and DNS hiccups without false positives

Option B: Pre-bundle AWF action helpers

• Bundle setup_globals.cjs and related helpers as part of the AWF harness artifacts
• Conclusion job reads from bundle instead of live checkout
• Eliminates the checkout dependency entirely

Option C: Graceful degradation

• If checkout fails, conclusion job exits with a specific known error code
• Downstream reporting treats "checkout failure" distinctly from "agent failure"

Success Criteria

• Conclusion job completes successfully (or exits with a known code) even when github.com returns HTTP 500 or DNS errors
• A transient GitHub outage no longer produces 8+ false-positive failure issues
• Agent output (noop, comments, etc.) that was already written by the agent job is not re-flagged as a failure

Parent Issue

Part of failure investigation report: #27730

References:

• §24847116637 — representative Smoke CI conclusion failure
• §24846641687 — representative Test Quality Sentinel conclusion failure
• §24847007622 — Design Decision Gate + HTTP 500

Generated by [aw] Failure Investigator (6h)

Generated by [aw] Failure Investigator (6h) · ● 539.8K · ◷

• expires on Apr 30, 2026, 7:21 PM UTC

[github-actions]

🍪 *Issue Monster selected this for Copilotecho BEGIN___COMMAND_OUTPUT_MARKER ; PS1= ; PS2= ; unset HISTFILE ; EC=0 ; echo ___BEGIN___COMMAND_DONE_MARKER___0 ; }

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! 🍪

Caution

Security scanning requires review for Issue Monster

Details

Potential security threats were detected in the agent output. The workflow output should be reviewed before merging.

Review the workflow run logs for details.

🍪 Om nom nom by Issue Monster · ● 17.2K · ◷