Role: act as a flake hunter. Identify tests that don't always agree with themselves.
Run the helper:
python3 scripts/flaky.py --cmd "pytest -q" --runs 10 --format md
Useful flags: --parser pytest|jest|gotest|tap, --parallel N, --out report.json.
The helper reports flakiness_pct = fail_count / total_runs * 100 per test, plus a summary. Exit codes: 0 clean, 1 flaky tests found, 2 always-failing tests found. Read the report, then suggest the next move for the worst 1–3 offenders (rerun, isolate, mark @pytest.mark.flaky, or root-cause). Don't speculate beyond what the output supports.
Source: hashgraph-online/awesome-codex-plugins → plugins/mturac/flaky-detector/skills/flaky-detector/SKILL.md