Agent Quality Metrics (2026-10-01 → 2026-10-31)
Generated by node scripts/agent-metrics.mjs on 2026-10-21 09:48 UTC.
Escaped defect = a later fix: commit within 7 days touching a code file the run changed.
Summary
Metric
Value
Agent runs found
16 (0 with a run-metrics file)
Escaped defect rate
6% (1 of 16 runs needed a follow-up fix)
Reverts of agent commits
0
Plan + code review approved in round 1
n/a of runs with metrics
Runs blocked or stopped
0
Code review skipped
0
Flaky tests seen in QA ×2
0
Cost (local runs with a figure)
n/a
Headless PRs
12 (closed unmerged 1, with /revise 0)
Median PR open → merge
7 min
CI build failure rate (all pushes)
6%
CI build failure rate (agent head commits)
0% of 6
Headless autopilot jobs
success 27, skipped 9, cancelled 4
By week
Week (Mon)
Agent runs
Runs with escaped defect
CI builds
CI failures
2026-09-28
0
0
3
0
2026-10-12
0
0
16
1
2026-10-19
16
1
15
1
Escaped defects
2026-10-19-dynamic-cron-jobs → 62526c97 fix(jobs): job bodies propagate failures so Admin shows FAILED (SJ-1) (api/src/main/java/com/canhlabs/funnyapp/jobs/AppScheduler.java)
Reverts
Runs
Run
Last commit
Commits
Code files
Rounds (plan/code/qa)
Result
2026-10-19-suggestion-rereview
2026-10-19
1
0
n/a
n/a
2026-10-19-suggestion-transitions
2026-10-19
1
0
n/a
n/a
pr-27
2026-10-19
1
1
n/a
n/a
pr-29
2026-10-19
1
3
n/a
n/a
pr-31
2026-10-19
1
1
n/a
n/a
pr-33
2026-10-19
1
3
n/a
n/a
2026-10-19-dynamic-cron-jobs
2026-10-19
3
19
n/a
n/a
pr-35
2026-10-20
1
2
n/a
n/a
pr-37
2026-10-20
1
3
n/a
n/a
pr-39
2026-10-20
1
2
n/a
n/a
pr-43
2026-10-20
1
2
n/a
n/a
pr-45
2026-10-21
1
3
n/a
n/a
2026-10-21-report-webhook-state
2026-10-21
3
16
n/a
n/a
pr-47
2026-10-21
1
1
n/a
n/a
2026-10-21-report-closed-notify
2026-10-21
3
13
n/a
n/a
pr-49
2026-10-21
1
1
n/a
n/a
Data gaps
Runs before the Autopilot-Run trailer are found only through their report hashes or the headless PR title; local lite runs from that time are missed.
A follow-up fix done by a later autopilot run whose subject is not fix: is not counted.
Runs that ended less than 7 days before the report date have not had their full fix window yet.
Headless runs have no cost in their run-metrics file; the workflow posts it on the PR.