Repository navigation
[agent-job-health] Agent Job Health 2026-11-14: 3.89% agent-job failure rate (10 of 257), partial coverage #77876
Closed
Replies: 1 comment
|
This discussion has been marked as outdated by Agent Job Health Monitor. A newer discussion is available at Discussion #78176. |
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.
Summary
Warning
Coverage is partial: about 9 of 24 hours were analysed, so the 15-hour gap is unknown. All 10 agent failures are unverified against open issues, so novel failures cannot be ruled out.
Failure Rate by Step
One run can fail more than one step, so shares sum to more than 100%.
Failure Accounting (44 completed failures)
The logs summary aftermath clusters (31 runs) overstate agent failures. Only 8 of those 31 runs are agent-job failures.
Tracked Failures
No cluster matched an open issue, but this is unverified. Issue search returned 9 to 10 integrity-filtered items per query. The only visible result (issue 24568, copilot-setup-steps building gh-aw from source) is unrelated. No issue was filed in this run because novelty could not be checked.
Conclusion-job failures with a successful agent job (not counted in agent_failures): Auditor 9 of 9 runs, Daily Go Test Parallelizer 5 of 5, Matt Pocock Skills Reviewer 4 of 22, Ponytail Reviewer 1 of 22, PR Sous Chef 1 of 52.
Novel Failure Clusters (unverified)
Blast radius: 10 distinct workflows, each failing once. Seven of the ten are smoke or engine-test workflows, so the signal concerns engine integration rather than production workflows. Nothing here indicates a fleet-wide regression.
Schedule Heartbeat
No blind spots detected. 242 schedule-triggered workflows were checked: 48 were seen in the local window and 194 were checked against the GitHub API, with 0 API errors. No schedule workflow is disabled.
Out of scope: eight smoke-* workflows (cursor, deepseek-harness, drive, multi-pr, opencode, project, pydantic, update-cross-repo-pr) concluded failure on scheduled runs on 2026-11-13 between 00:01 and 00:30 UTC. These fall before this window and are not counted.
The population of 242 is lower than the 253 recorded in the 2026-11-10 sweep. Reconciling the two is left for the next run.
Recommendations
Limitations
View per-workflow breakdown (workflows with any failure; 71 distinct workflows, 61 with zero failures)
View inspected run URLs (agent-job failures)
All reactions