Stabilize flaky tests with minimal assertion-preserving fixes that remove nondeterminism.
npx skills add https://github.com/QwenLM/qwen-code --skill deflake
Please help me install this Agent Skill. Skill: deflake Source: https://github.com/QwenLM/qwen-code/tree/main/.qwen/skills/deflake Command: npx skills add https://github.com/QwenLM/qwen-code --skill deflake
Flaky tests that pass and fail on the same commit erode trust in CI and waste developer time on reruns. This Skill diagnoses the nondeterminism behind a named flaky test and applies the smallest fix that makes it deterministic without weakening or deleting any assertion.
Ask the agent to deflake the named failing test from the issue, keeping every assertion intact and verifying it passes on repeated runs.
High-intent search queries and answers about installing and using this skill.
Identify the source of nondeterminism, then apply the smallest targeted fix: raise the timeout budget, await the real condition instead of sleeping, seed randomness or fake timers, or isolate shared resources. Every assertion and input stays exactly as written.
Give the test a generous per-test testTimeout as the third argument to it(), or replace a fixed-iteration poll loop with a real wall-clock budget. Fixed counts of setImmediate turns elapse in milliseconds and race real I/O under CI contention.
Retries hide the flake instead of fixing it and can mask a real intermittent product bug. The correct approach is to remove the nondeterminism itself through timing, randomness, or isolation fixes while preserving every assertion.
Yes. Using vi.useFakeTimers(), mocking Date.now, or seeding the RNG pins values that would otherwise drift with the real clock or Math.random, making the test produce the same result on every run.
Stop and document the finding in a failure report instead of patching the test. If none of the four allowed fixes applies or the failure looks like genuine intermittent product behavior, a human should investigate the underlying bug.
jmaglyAnalyze CI history and test execution patterns to identify flaky tests.
jacobcyReview source diffs and pull requests for correctness, regressions, and security issues.
otto-poblyshRequire fresh verification command output before asserting task completion.
luxunxianshengTrace software bug root causes through a structured four-phase investigation process.
