Description
PR Sous Chef's consolidated safe_outputs job ("Process Safe Outputs" step) has a recurring ~8% failure rate (2/24 in an 8h sample per today's Safe Output Health Report, discussion #54725) and continues generating fresh [aw] Failed jobs: PR Sous Chef auto-filed issues as recently as run #32543132845 (2026-08-22, issue #54685) and #32526302837. A prior root-cause attempt (#53615) was closed on 2026-08-18 without a durable fix, and the pattern persists. Upload Safe Outputs Items still succeeds after the failure, meaning already-valid items get uploaded — strongly suggesting one malformed item in this workflow's unusually wide multi-type batch (add-comment, approve-workflow-run, resolve-pull-request-review-thread, dismiss-pull-request-review, update-pull-request, push-to-pull-request-branch, create-issue) aborts the entire step instead of degrading per-item.
Expected Impact
Root-causing (or making validation degrade per-item instead of failing the whole batch) would stop a recurring ~15-minute-cycle disruption to one of the fleet's most actuation-heavy workflows (comments, branch pushes, workflow-run approvals) and stop the steady stream of duplicate [aw] Failed jobs: PR Sous Chef issues.
Suggested Agent
New Agent / general-purpose — needs access to raw Process Safe Outputs step logs (see related issue on exposing safe-output step logs via the audit/logs tooling) to pin down which of the 7 output types is failing validation.
Estimated Effort
Medium (1-4 hours)
Data Source
DeepReport analysis of Safe Output Health Report (discussion #54725, 2026-08-22), Work Item 1. Prior related closed issue: #53615.
Generated by 🔬 Deep Report · agent · 109.8 AIC · ⌖ 8.86 AIC · ⊞ 12.4K · ◷
Description
PR Sous Chef's consolidated
safe_outputsjob ("Process Safe Outputs" step) has a recurring ~8% failure rate (2/24 in an 8h sample per today's Safe Output Health Report, discussion #54725) and continues generating fresh[aw] Failed jobs: PR Sous Chefauto-filed issues as recently as run #32543132845 (2026-08-22, issue #54685) and #32526302837. A prior root-cause attempt (#53615) was closed on 2026-08-18 without a durable fix, and the pattern persists.Upload Safe Outputs Itemsstill succeeds after the failure, meaning already-valid items get uploaded — strongly suggesting one malformed item in this workflow's unusually wide multi-type batch (add-comment, approve-workflow-run, resolve-pull-request-review-thread, dismiss-pull-request-review, update-pull-request, push-to-pull-request-branch, create-issue) aborts the entire step instead of degrading per-item.Expected Impact
Root-causing (or making validation degrade per-item instead of failing the whole batch) would stop a recurring ~15-minute-cycle disruption to one of the fleet's most actuation-heavy workflows (comments, branch pushes, workflow-run approvals) and stop the steady stream of duplicate
[aw] Failed jobs: PR Sous Chefissues.Suggested Agent
New Agent / general-purpose — needs access to raw
Process Safe Outputsstep logs (see related issue on exposing safe-output step logs via the audit/logs tooling) to pin down which of the 7 output types is failing validation.Estimated Effort
Medium (1-4 hours)
Data Source
DeepReport analysis of Safe Output Health Report (discussion #54725, 2026-08-22), Work Item 1. Prior related closed issue: #53615.