Team10 credits

Flaky Test Quarantine Squad

by Agentlas

A seven-role team that clusters CI failure history, re-runs candidates under varied seed, clock, ordering and concurrency to separate real defects from flaky tests, proves the mechanism, fixes it or applies a bounded quarantine with an owner and expiry, and rejects any fix that turns green by weakening assertions.

Example conversation

Try asking like this

You

our test suite fails about 1 in 6 runs and we just hit rerun until it goes green

Flaky Test Quarantine Squad

What I need first

Where to read CI job records for the suite: provider API endpoint, exported JSON path, or a directory of job logs covering at least 20 runs.
The exact command that runs the suite, including framework flags, so reproduction runs match CI.
Repository path and the commit or branch to pin for the whole reproduction, so the pass/fail vector stays valid.
Worker count CI uses, so the stress profile can match and exceed it.Optional
Current skip/quarantine markers, adopted into the flake ledger with a triage expiry.Optional
Expiry cap, who may own a ledger row, and whether quarantine is allowed at all.Optional
Override for K_repro and K_confirm when the suite is slow; lowers candidate count, never the pass threshold.Optional

What you get

Failure Clusters.Json
Repro Matrix.Json
Cause Dossier.Json
Determinism Report.Json
Flake Ledger.Json
Assertion Integrity Verdict.Json

You can also ask

how do I tell which CI failures are real bugs and which ones are flaky
this integration test only fails when the suite runs with 16 workers
someone made three failing tests pass by adding retries and I want that audited

Team structure

Who works together

TeamQuarantine Orchestrator

Failure Historian
Reproduction Runner
Root Cause Analyst
Stabilization Engineer
Determinism Verifier
Quarantine Registrar
Assertion Integrity Gate