Problem
Fix the Copilot CLI harness segfault first — it kills the entire run before Copilot even starts, and nothing in the retry loop catches it.
Affected workflows and runs
- Agent Performance Analyzer - Meta-Orchestrator (
.github/workflows/agent-performance-analyzer.lock.yml) — §31390296030, 2026-08-10T12:55Z. First failure in the last 11 daily runs; most recent success §31314425506 (2026-08-09).
Probable root cause
The node process running the harness segfaults (exit 139 = SIGSEGV) right after the pre-flight /reflect fetch and before the Copilot subprocess spawns:
[copilot-harness] awf-reflect: fetching (apiproxy/redacted) (timeout=60000ms)
/tmp/awf-cmd-1.sh: line 29: 80 Segmentation fault /bin/bash -c '... node ... copilot_harness.cjs ... copilot_sdk_driver.cjs ...'
[WARN] Command completed with exit code: 139
audit-diff against the nearest successful run ([§31314425506]) shows zero firewall/network drift (0 new domains, 0 status changes, 0 anomalies) — this rules out a sandbox/policy change as the trigger. The crash is in the node process itself (harness or sdk-driver), not bash, and it happens before maxRetries=3 can engage because the harness process dies outright instead of raising a catchable error.
Proposed remediation
- Reproduce locally with the same Node version (v22.23.1): run
copilot_harness.cjs through the /reflect fetch path under ulimit -c unlimited to capture a core dump and confirm whether the fault is in Node itself, a native addon, or the sdk-driver's reflect-handling code.
- Wrap the harness process spawn so a SIGSEGV/non-zero-from-signal exit is treated as retryable, not a hard job failure — currently the retry loop never gets a chance to run.
- Add basic resource/memory monitoring around the
/reflect fetch step to rule out OOM as the trigger.
Success criteria
- Next 3 scheduled runs of Agent Performance Analyzer - Meta-Orchestrator complete without exit code 139.
- If the harness does segfault again, the job retries automatically instead of failing outright.
Generated by 🔍 [aw] Failure Investigator (6h) · agent · 176.3 AIC · ⌖ 43.7 AIC · ⊞ 5.3K · ◷
Problem
Fix the Copilot CLI harness segfault first — it kills the entire run before Copilot even starts, and nothing in the retry loop catches it.
Affected workflows and runs
.github/workflows/agent-performance-analyzer.lock.yml) — §31390296030, 2026-08-10T12:55Z. First failure in the last 11 daily runs; most recent success §31314425506 (2026-08-09).Probable root cause
The node process running the harness segfaults (exit 139 = SIGSEGV) right after the pre-flight
/reflectfetch and before the Copilot subprocess spawns:audit-diffagainst the nearest successful run ([§31314425506]) shows zero firewall/network drift (0 new domains, 0 status changes, 0 anomalies) — this rules out a sandbox/policy change as the trigger. The crash is in the node process itself (harness or sdk-driver), not bash, and it happens beforemaxRetries=3can engage because the harness process dies outright instead of raising a catchable error.Proposed remediation
copilot_harness.cjsthrough the/reflectfetch path underulimit -c unlimitedto capture a core dump and confirm whether the fault is in Node itself, a native addon, or the sdk-driver's reflect-handling code./reflectfetch step to rule out OOM as the trigger.Success criteria