[aw-failures] [P0] copilot-sdk headless server SIGABRT before ready — Dead Code Removal Agent total run loss

[github-actions]

Fix the copilot-sdk headless server crash now — Dead Code Removal Agent lost its entire run before a single tool call

One-sentence rationale: the Copilot engine's headless SDK server aborted with SIGABRT before reaching ready state, killing the run before the agent could do any work — this is a distinct signal from the known exit-139 segfault pattern already tracked in #54186.

Problem statement

Dead Code Removal Agent failed with an auto-diagnosed Engine Failure: the copilot engine terminated before producing output — copilot-sdk headless server exited before ready (exitCode=unknown signal=SIGABRT). This is a total run loss: the SDK crashed during startup, before any tool call or model turn happened.

Affected workflows and run IDs

• Dead Code Removal Agent (.github/workflows/dead-code-remover.lock.yml), branch main — §32644707736, 2026-08-23 14:12 UTC (tracked by auto-issue [aw] Dead Code Removal Agent failed #55092)

Probable root cause

Unclear from available evidence — the copilot-sdk process aborts (SIGABRT, signal 6) rather than segfaulting (SIGSEGV/exit 139, the signature in #54186) or exiting cleanly. SIGABRT typically indicates the process's own internal assertion/invariant check killed it (e.g. abort() from a native binding, an unhandled panic, or an out-of-memory guard), which is a different failure class from #54186's cross-engine segfault and should not be merged into that issue without confirming a shared cause.

Proposed remediation

1. Capture and surface the copilot-sdk headless server's own stderr/panic output at crash time — currently only the exit signal is visible, not why it aborted.
2. Check whether this run had unusual resource pressure (large repo scan, memory-heavy dead-code analysis) that could trigger an internal guard/OOM abort in the SDK.
3. Compare against [aw-failures] [P0] Cross-engine agent-CLI segfault (exit 139) — confirmed on Auto-Triage/Pi CLI, now spread to 8 workflows acros #54186 once more instances occur — if SIGABRT starts appearing across other engines/workflows the way exit-139 did, fold into that cross-engine crash investigation; otherwise keep as a separate copilot-sdk-specific defect.

Success criteria / verification

• Next copilot-sdk crash (if any) surfaces an internal error/panic message instead of a bare signal.
• Dead Code Removal Agent completes a scheduled run without a pre-ready SDK abort.

References:

• https://github.com/github/gh-aw/actions/runs/32644707736 ([aw] Dead Code Removal Agent failed #55092)

Related but distinct from #54186 (different crash signal: SIGABRT vs SIGSEGV/exit 139) — do not merge without confirming shared root cause.
Related to #54952

Generated by 🔍 [aw] Failure Investigator (6h) · agent · 194.8 AIC · ⌖ 8.85 AIC · ⊞ 6.4K · ◷

• expires on Aug 30, 2026, 11:12 AM UTC-08:00

[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 · cop10 · 2.08 AIC · ⌖ 1.46 AIC · ⊞ 18.6K · ◷