Fix Claude Code CLI abnormal exit (code 159): classify fatal-signal crashes in harness retry logic by Copilot · Pull Request #52837 · github/gh-aw

Copilot · 2026-08-15T05:17:05Z

Avenger's Claude Code CLI step crashed with exit code 159 (128+31 = SIGSYS) twice in one day. The harness only special-cased SIGKILL(137)/SIGTERM(143) as non---continue-able termination signals; any other fatal-signal crash (SIGSYS, SIGSEGV, etc.) fell through to the generic "partial execution" retry path, which resumes the exact on-disk session via --continue — risking an immediate repeat of whatever crashed it.

Changes

New crash-signal classification in actions/setup/js/claude_harness.cjs: added a CRASH_SIGNAL_EXIT_CODES map for known fatal-signal exit codes (SIGILL=132, SIGABRT=134, SIGBUS=135, SIGFPE=136, SIGSEGV=139, SIGSYS=159), with isCrashSignalExitCode() and crashSignalNameForExitCode() helpers.
Retry decision update: shouldRetryWithContinue() now treats these crash codes like the existing SIGKILL/SIGTERM handling — --continue is disabled and the next attempt starts a fresh session instead of resuming corrupted state.
Improved diagnostics: retry logs now surface the resolved signal name (e.g. crashSignal=SIGSYS) alongside the exit code, giving future occurrences a clearer trail without requiring a core dump.
Tests: added coverage for the new helpers and the updated shouldRetryWithContinue behavior on crash exit codes (134/139/159).

// Before: exit 159 fell into the generic "partial execution" path → retried with --continue
// After: recognized as a fatal-signal crash → --continue disabled, fresh-run retry
isCrashSignalExitCode(159); // true (SIGSYS)
crashSignalNameForExitCode(159); // "SIGSYS"

Fixes [aw-failures] [P0] Avenger: Claude Code CLI abnormal exit (code 159) — 2nd crash today, distinct from tracked exit-139 #52727

branch refresh requested from run https://github.com/github/gh-aw/actions/runs/31886878354> Generated by 👨‍🍳 PR Sous Chef · gpt54 · 11.7 AIC · ⌖ 5.9 AIC · ⊞ 8.7K · ◷

_{Comment /souschef to run again}

…e harness retry logic Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

github-actions · 2026-08-15T06:26:17Z

PR Triage

Category: bug | Risk: medium | Score: 66/100

Impact: 28/50 (fixes harness mis-classification of fatal-signal crashes, improves retry reliability)
Urgency: 20/30 (recurring crash signature, no CI signal yet)
Quality: 18/20 (small targeted diff, clear root cause, no reviews yet)

Recommended action: fast_track
Targeted reliability fix for harness crash classification. Batched with reliability-fixes cluster (#52757, #52836, #52854).

Generated by 🔧 PR Triage Agent · auto · 46.6 AIC · ⌖ 2.63 AIC · ⊞ 8.1K · ◷

github-actions · 2026-08-15T12:25:46Z

PR Triage

Category: bug
Risk: medium
Priority: medium
Score: 68/100
Recommended action: fast_track
Batch: reliability-fixes

Improves harness crash-signal classification to avoid corrupted-session retries. CI unstable — verify before merge.

_{Automated triage — see [PR Triage Report] for full context.}

Generated by 🔧 PR Triage Agent · auto · 62.5 AIC · ⌖ 2.53 AIC · ⊞ 8.1K · ◷

pelikhan · 2026-08-15T12:55:16Z

@copilot generalize to all harnesses

Copilot

Pull request overview

Adds fatal-signal classification to Claude harness retry logic so crashes should restart with a fresh session.

Changes:

Maps known fatal-signal exit codes to signal names.
Disables --continue after classified crashes.
Adds helper and retry-decision tests.

Show a summary per file

File	Description
`actions/setup/js/claude_harness.cjs`	Adds crash classification, logging, and fresh-session retries.
`actions/setup/js/claude_harness.test.cjs`	Tests crash-code helpers and retry behavior.

Review details

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.

Files reviewed: 2/2 changed files
Comments generated: 1
Review effort level: Balanced

    return false;
  }
-  if (isSignalTerminationExitCode(exitCode)) {
+  if (isSignalTerminationExitCode(exitCode) || isCrashSignalExitCode(exitCode)) {


Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

Copilot · 2026-08-15T13:15:18Z

@copilot generalize to all harnesses

Generalized in 8a1c0e6: extracted the crash-signal classification (SIGILL/SIGABRT/SIGBUS/SIGFPE/SIGSEGV/SIGSYS) into a shared harness_crash_signals.cjs module and applied it to the Copilot harness's CLI-mode --continue retry decision, so a fatal-signal crash there also permanently disables --continue and forces a fresh-run retry. Codex harness needs no change since it never supports --continue — every retry there is already a fresh run.

gh-aw-bot · 2026-08-15T13:25:31Z

@copilot Quick triage for maintainer-ready follow-up:

Please refresh the branch if needed, verify the current unstable status is fully explained or resolved, and run the pr-finisher skill before handing this PR back to maintainers.

Outstanding review items:

confirm the generalized fatal-signal handling is correct across all harnesses that can reuse this classifier, and summarize why no additional harness changes are needed

Failed checks from the compact candidate set:

None listed.

Branch update was requested automatically for this run when GitHub allows it.
Run context: https://github.com/github/gh-aw/actions/runs/31886878354

Generated by 👨‍🍳 PR Sous Chef · gpt54 · 11.7 AIC · ⌖ 5.9 AIC · ⊞ 8.7K · ◷
_{Comment /souschef to run again}

Co-authored-by: gh-aw-bot <259018956+gh-aw-bot@users.noreply.github.com>

Copilot · 2026-08-15T13:40:33Z

@copilot Quick triage for maintainer-ready follow-up:

Please refresh the branch if needed, verify the current unstable status is fully explained or resolved, and run the pr-finisher skill before ha...

Fixed the outstanding blocking review item in 6f... (see latest commit): runProcess in process_runner.cjs now synthesizes the conventional 128+signal exit code when Node reports code=null, signal="SIGSYS" (or any other signal), instead of collapsing it to exitCode=1. This means isCrashSignalExitCode/shouldRetryWithContinue in both claude_harness.cjs and copilot_harness.cjs now actually observe signal-killed subprocesses, closing the gap the reviewer flagged. Added an integration-style test that signal-kills a stub child (SIGSEGV) and asserts the synthesized exit code (139) and log output. No other harness needs a change: Codex never supports --continue, so all its retries are already fresh runs regardless of exit code/signal. Full actions/setup/js suite (392 files / 12996 tests) passes locally.

pelikhan · 2026-08-15T13:52:03Z

@copilot fix https://github.com/github/gh-aw/actions/runs/31887826567/job/95020049821

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

github-actions · 2026-08-15T17:24:50Z

🎉 This pull request is included in a new release.

Release: v0.86.3

Initial plan

d669bf5

Copilot AI assigned Copilot and pelikhan Aug 15, 2026

Copilot started work on behalf of pelikhan August 15, 2026 05:17 View session

Copilot AI linked an issue Aug 15, 2026 that may be closed by this pull request

[aw-failures] [P0] Avenger: Claude Code CLI abnormal exit (code 159) — 2nd crash today, distinct from tracked exit-139 #52727

Closed

Classify fatal-signal crash exit codes (SIGSEGV/SIGSYS/etc.) in Claud…

46b0fb1

…e harness retry logic Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

Copilot AI changed the title ~~[WIP] Fix Claude Code CLI abnormal exit (code 159)~~ Fix Claude Code CLI abnormal exit (code 159): classify fatal-signal crashes in harness retry logic Aug 15, 2026

Copilot finished work on behalf of pelikhan August 15, 2026 05:33

Copilot AI requested a review from pelikhan August 15, 2026 05:33

github-actions Bot mentioned this pull request Aug 15, 2026

[pr-review] Daily PR Code Quality Review — 31867888921 #52856

Open

github-actions Bot added pr-action:fast_track pr-agent:copilot pr-batch:reliability-fixes pr-priority:medium pr-risk:medium pr-type:bug labels Aug 15, 2026

This was referenced Aug 15, 2026

Fix Design Decision Gate false failures caused by the LLM invocation cap #52836

Merged

Fail fast when required MCP servers are omitted #52854

Closed

[PR Triage Report] PR Triage Report — 2026-08-15 (Run 31869042445) #52862

Closed

github-actions Bot mentioned this pull request Aug 15, 2026

[PR Triage Report] PR Triage Report — 2026-08-15 (run 31884164165) #52888

Closed

pelikhan marked this pull request as ready for review August 15, 2026 12:54

Copilot AI balanced review requested due to automatic review settings August 15, 2026 12:54

Copilot started reviewing on behalf of pelikhan August 15, 2026 12:55 View session

Copilot AI reviewed Aug 15, 2026

View reviewed changes

Comment thread actions/setup/js/claude_harness.cjs

return false;

}

if (isSignalTerminationExitCode(exitCode)) {

if (isSignalTerminationExitCode(exitCode) || isCrashSignalExitCode(exitCode)) {

Copilot started work on behalf of pelikhan August 15, 2026 13:02 View session

Generalize fatal-signal crash classification to Copilot harness

2070ca9

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

Copilot finished work on behalf of pelikhan August 15, 2026 13:15

Merge branch 'main' into copilot/aw-failures-fix-code-cli-crash

d2b9db9

github-actions Bot mentioned this pull request Aug 15, 2026

[pr-sous-chef] PR Sous Chef run report #52798

Open

Copilot started work on behalf of gh-aw-bot August 15, 2026 13:25 View session

Preserve child signal in runProcess for crash-signal classification

b688609

Co-authored-by: gh-aw-bot <259018956+gh-aw-bot@users.noreply.github.com>

Copilot finished work on behalf of gh-aw-bot August 15, 2026 13:41

Copilot AI requested a review from gh-aw-bot August 15, 2026 13:41

Copilot started work on behalf of pelikhan August 15, 2026 13:52 View session

Fix prettier formatting in eslint-factory lastIndex-reset rule

d27061a

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

pelikhan merged commit 6f73444 into main Aug 15, 2026
9 checks passed

pelikhan deleted the copilot/aw-failures-fix-code-cli-crash branch August 15, 2026 14:05

Navigation Menu

Uh oh!

Fix Claude Code CLI abnormal exit (code 159): classify fatal-signal crashes in harness retry logic - #52837

Fix Claude Code CLI abnormal exit (code 159): classify fatal-signal crashes in harness retry logic#52837
pelikhan merged 6 commits into
mainfrom
copilot/aw-failures-fix-code-cli-crash

Copilot AI commented Aug 15, 2026 •

edited by github-actions Bot

Loading

Uh oh!

github-actions Bot commented Aug 15, 2026

Uh oh!

github-actions Bot commented Aug 15, 2026

Uh oh!

pelikhan commented Aug 15, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI commented Aug 15, 2026

Uh oh!

gh-aw-bot commented Aug 15, 2026

Uh oh!

Copilot AI commented Aug 15, 2026

Uh oh!

pelikhan commented Aug 15, 2026

Uh oh!

Uh oh!

github-actions Bot commented Aug 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Conversation

Copilot AI commented Aug 15, 2026 • edited by github-actions Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changes

Uh oh!

github-actions Bot commented Aug 15, 2026

PR Triage

Uh oh!

github-actions Bot commented Aug 15, 2026

PR Triage

Uh oh!

pelikhan commented Aug 15, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Review details

Uh oh!

Copilot AI commented Aug 15, 2026

Uh oh!

gh-aw-bot commented Aug 15, 2026

Uh oh!

Copilot AI commented Aug 15, 2026

Uh oh!

pelikhan commented Aug 15, 2026

Uh oh!

Uh oh!

github-actions Bot commented Aug 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Copilot AI commented Aug 15, 2026 •

edited by github-actions Bot

Loading