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
§31797105737 — failed 2026-08-14T11:39:34Z at "Execute Claude Code CLI"; agentic-workflows audit reports ##[error]Process completed with exit code 159.
§31777889016 — failed 2026-08-14T07:04Z, same step, self-filed as [aw] Avenger failed #52665 ("claude engine terminated unexpectedly", no exit code captured in that issue).
Root cause (partial — needs a full stack trace from the next occurrence)
Exit 159 (128+31, SIGSYS/bad-syscall under the standard shell signal-exit convention) is a different signature from the tracked exit-139 (SIGSEGV) pattern in #51789 — that issue is scoped to the copilot engine; this is the claude engine. audit on §31797105737 shows the network/firewall layer healthy up to the crash (102 allowed requests, 0 blocked, all expected domains), so this looks like a sandbox/runtime-level fault rather than network or auth. The run log tail only captures the last agent JSON message, not the actual crash frame, so root cause isn't fully confirmed yet.
Proposed remediation
Add core-dump / signal capture (ulimit -c unlimited or equivalent) to the Claude harness wrapper so the next crash preserves a stack trace instead of just the last streamed message.
Check whether a seccomp/sandbox policy change coincides with the first occurrence at 07:04 UTC today.
Fix the Claude Code CLI crash in Avenger — exit code 159, now 2 occurrences in one day.
Affected workflow: Avenger (
.github/workflows/avenger.lock.yml)Runs analyzed:
agentic-workflows auditreports##[error]Process completed with exit code 159.Root cause (partial — needs a full stack trace from the next occurrence)
Exit 159 (128+31, SIGSYS/bad-syscall under the standard shell signal-exit convention) is a different signature from the tracked exit-139 (SIGSEGV) pattern in #51789 — that issue is scoped to the
copilotengine; this is theclaudeengine.auditon §31797105737 shows the network/firewall layer healthy up to the crash (102 allowed requests, 0 blocked, all expected domains), so this looks like a sandbox/runtime-level fault rather than network or auth. The run log tail only captures the last agent JSON message, not the actual crash frame, so root cause isn't fully confirmed yet.Proposed remediation
ulimit -c unlimitedor equivalent) to the Claude harness wrapper so the next crash preserves a stack trace instead of just the last streamed message.claudevscopilot) and different exit code, even though both are "engine terminated unexpectedly" crashes.Success criteria
3 consecutive scheduled Avenger runs complete without an "engine terminated unexpectedly" / non-zero-exit crash in the "Execute Claude Code CLI" step.
Related to #52665, #51789
References:
Related to [aw-failures] [aw] Failure Investigator Report — 2026-08-13 19:10 UTC (6h) #52570