Skip to content

[aw-failures] [aw] Failure Investigator Report — 2026-08-11 (6h) #51983

Description

@github-actions

Overview

Fix the Avenger workflow first — a deterministic Bun runtime segfault killed both of its runs in this window (8/8 execution attempts crashed), and it's the only P0 failure with zero existing tracking coverage. Everything else this cycle is a single-run, low-signal blip or a working-as-intended policy gate.

Fix roadmap

  • P0 — Avenger Bun segfault: pin/bisect the Bun runtime and fast-fail the harness retry loop instead of burning 4 attempts on a confirmed crash. Tracked in the sub-issue below.
  • P1: none identified this cycle.
  • P2 — watch, don't ticket yet: Daily Assign Issue To User, Daily PR Code Quality Review, Daily AgentRx Trace Optimizer each had one unexplained/opaque failure — re-open this as P1 if any recur tomorrow. Daily Container Image Security Scan failed because it found real critical CVEs — that's the gate doing its job, not a bug, no action needed.
Failure clusters (severity-ranked) and evidence
Cluster Severity Workflow Signature Runs Tracking
cluster-1 P0 Avenger Bun/Claude Code CLI SIGSEGV, 100% of attempts §31465993823, §31462571906 New gap → sub-issue
cluster-2 P2 Daily Assign Issue To User Copilot CLI exits 1 after 3 retries, no root-cause line captured §31463851158 Untracked, single occurrence
cluster-3 P2 Daily PR Code Quality Review Tool-denial threshold hit reading own generated diff file (/tmp/gh-aw/agent/pr51948_go.diff) §31463530917 Untracked — same failure class as #51547, different workflow/file; extend that issue's scope rather than duplicating
cluster-4 P2 Daily AgentRx Trace Optimizer Job failed, truncated log shows only post-job cleanup — cause not visible in captured tail §31462581875 Untracked, needs a full (untruncated) log pull if it recurs
cluster-5 P2 Daily Container Image Security Scan Policy gate exits 1: "Critical vulnerabilities detected in container images" §31462580227 Not a bug — scanner working as designed

Evidence — Avenger Bun segfault (cluster-1, P0):

  • Both Avenger runs in the window (§31465993823 and §31462571906) crashed on every one of their 4 retry attempts with panic(main thread): Segmentation fault, signal=SIGSEGV — 8/8 attempts, 0 successes. Crash address is different each time (0x12, 0x21, 0x0, 0xFFFE000000000015), which is the signature of heap/GC corruption rather than a deterministic input-triggered bug. Bun's own panic message states: "This indicates a bug in Bun, not your code."
  • audit-diff of the failed run against the nearest prior success (§31459752206) rules out a network/policy cause: the failed run successfully made 43 api.anthropic.com calls and 19 GitHub API calls before crashing, and there are no anomalous firewall/domain deltas. The crash happens after real, working API traffic — it's an in-process Bun runtime fault, not a sandbox or tool-permission issue.
  • Secondary harness bug found in §31462571906: attempt 2's --continue retry failed with "no deferred tool marker" and the harness responded by permanently disabling --continue (failure_reason=harness_retry_path_invalid), forcing full fresh-context restarts for attempts 3 and 4 — which then also segfaulted. This wastes ~90s of runtime per failed run and loses conversation context on every retry.

Existing issue correlation:

Sub-issues created

  • #aw_bunsegv — [P0] Fix Bun runtime segfault crashing Claude Code CLI in Avenger

References:

Generated by 🔍 [aw] Failure Investigator (6h) · agent · 123.6 AIC · ⌖ 41.5 AIC · ⊞ 5.3K ·

  • expires on Aug 17, 2026, 11:20 PM UTC-08:00

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions