[safe-output-health] 🏥 Safe Output Health Report - 2026-06-23 #40961
Closed
Replies: 3 comments
-
|
"Smoke run 28004944791 checked discussion commenting at 2026-06-23T05:48:58Z ✅"
|
Beta Was this translation helpful? Give feedback.
0 replies
-
|
Smoke test run initiated. ✅
|
Beta Was this translation helpful? Give feedback.
0 replies
-
|
This discussion was automatically closed because it expired on 2026-06-24T05:46:12.606Z.
|
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Executive Summary
2026-06-23T05:15Z → 05:32Z, 12 run directories)engine.driverfield with @earendil-works/pi-agent-core built-in driver #40897 /copilot/add-sample-pi-agent-core-driver)patch-format: bundlebranch-pin family)Safe Output Job Statistics
push_to_pull_request_branch, bundle)Error Cluster:
changeset_generator_push_to_pull_request_branch_bundle_job_hardfailpull_request(labeled), branchcopilot/add-sample-pi-agent-core-driverfailurepush-to-pull-request-branch(patch-format: bundle,allowed-files: .changeset/**,protected-files: blocked) +update-pull-request(append).changesetfile was not pushed to the PR branch → missing release-note entry on that PR (low-to-medium production impact)Root cause (inferred — see data limitation)
This is the same family as the 2026-06-17
create_pull_request_branch_pin_dubious_ownership_bridge_processcluster — thepatch-format: bundle/ branch-pin transport path — but at a different layer:create_pull_request(bundle)push_to_pull_request_branch(bundle)Key isolation: in the same window, Design Decision Gate (§28004370976) used
push_to_pull_request_branchwith the default patch-format and succeeded. This strongly implicates the bundle path specifically, notpush_to_pull_request_branchin general.The 06-17 root cause was the bundle branch-pin git step hitting
fatal: detected dubious ownershipbecause the safeoutputs bridge runs outside the container as a different user/HOME than the/home/runner/work/gh-aw/gh-awcheckout, and the agent's in-containergit config --global safe.directorycan't reach the bridge's gitconfig. The 06-23 JOB-layer failure is consistent with the same bundle branch-pin transport being exercised at the downstream push job.Process Safe Outputsstep logs were pre-bundled for any run today (job_details[]gives JOB conclusions only).agenticworkflows auditMCP tool (artifacts: true) did not surface the failed JOB's step stdout/stderr; itskey_findingseven mislabeled this as "failed before agent activation" — the failure was atsafe_outputs, well after agent success.patch-format: bundle+push_to_pull_request_branch) and family history, not directly read.run-28003923242Process Safe Outputs step log directly (Actions UI /gh run view) to confirm.Recurring Clusters Status
..._branch_pin_dubious_ownership_bridge_process(bundle)review_path_unresolved_422(Path variant)pr_review_buffer.cjs:554UNVALIDATED 26th consecutive auditadd_comment_discussion_resource_not_accessible_by_integrationtarget_star_review_comment_no_pr_number_fallbacktarget_star_add_labels_no_item_number_hardfailupdate_issue_target_triggering/assign_to_agent_*/hide_comment_intPositive Signals
add_comment→discussion+target_staron 06-14/06-15 did not fail its JOB today.push_to_pull_request_branch(default patch-format) CLEAN — isolates the failure to the bundle path.push_experiments_state,update_cache_memory,send_slack_message,trufflehog_scan,upload_code_scanning_sarifall success.Recommendations / Work Items
Work Item 1 — Investigate the
patch-format: bundlebranch-pin push path (Priority: High)run-28003923242Process Safe Outputs step error.push_to_pull_request_branchbundle path applies-c safe.directory=/home/runner/work/gh-aw/gh-aw(orsafe.directory='*') / correctHOME/GIT_CONFIG_GLOBALto the bridge git invocation — same fix area as 06-17.Work Item 2 — Improve safe_outputs failure observability (Priority: Medium)
auditMCP tool reports a generic "failed before agent activation" for asafe_outputs-stage failure and does not bundle/surface theProcess Safe Outputsstep log. This blocks root-cause analysis for exactly the failures this monitor exists to triage.Process Safe Outputsstep stdout for failedsafe_outputsjobs.key_findingsso asafe_outputs-stage failure isn't labeled as a pre-activation failure.Work Item 3 — Validate the long-standing
review_path_unresolved_422Path-variant fix (Priority: Low/Medium)pr_review_buffer.cjs:554Path-variant fallback unvalidated for 26 consecutive audits (no production 422 has fired it). Carryover — no new signal today.Next Steps
Process Safe Outputsstep log for run28003923242and confirm/replace the inferred root cause.safe.directoryfix at the downstream push path if confirmed.patch-format: bundle) for recurrence.References:
Beta Was this translation helpful? Give feedback.
All reactions