Обнаружение нестабильных тестов в CI

testing

test-flakiness is a Claude Code skill that detects non-deterministic (flaky) tests by parsing CI run logs and test result history across multiple runs. It aggregates per-test pass rates, identifies intermittent failures, and classifies likely root causes — including timing and async races, order dependencies, random seed usage, resource leaks, and floating-point comparisons. Three modes are available: analyzing a specific log file, scanning all logs in `.github/` or `test-results/`, and reviewing already-flagged tests from the quarantine section of `tests/regression-suite.md`. Findings are written to an optional `production/qa/flakiness-report-[date].md`. The skill is most effective during a project's polish phase or when developers start dismissing CI failures as "probably flaky" — the pattern that lets real bugs slip through unnoticed.