Each night a regression run leaves hundreds of log lines. This topic walks through an agent that groups repeat failures, ranks suspect commits, and writes one evidence-backed card per group.
Lesson | What it covers |
|---|---|
Inputs, tools, and outputs of one triage pass. | |
Normalizing logs and grouping repeats, with a runnable model. | |
Weighted scoring, rerun arithmetic, and a runnable ranking model. | |
Least-privilege tools, approvals, and untrusted log text. | |
Precision, recall, top-1 accuracy, and operating cost. |
Ranking runs once per group instead of once per log line. Ten repeats of one bug need one suspect list, not ten.