agent's fix passed locally but introduced a new race in CI — it

Not specified by the original author.

TL;DR

The agent's fix assumed single-threaded execution, passed locally, and introduced a race that fails on parallel CI. Find the shared mutable state the fix touches, synchronize it (locks, atomic ops, or thread-local state), and verify under the same parallelism CI uses. Local serial runs cannot validate concurrent code.

The symptom

agent's fix passed locally but introduced a new race in CI  -  it assumed single-threaded execution

Steps

1. Reproduce under parallelism

Run the fixed test with CI's parallelism settings (workers, threads). The race should appear. Expected: Race reproduced under parallel execution.

2. Find the shared state

Look for what the fix reads and writes across threads: globals, caches, files, counters. Expected: The unsynchronized shared state identified.

3. Synchronize properly

Add locking, use atomic operations, or make the state thread-local. Match the granularity to the actual sharing. Expected: No data race under the race detector or stress run.

4. Verify at CI parallelism

Rerun with the full parallel configuration, ideally with a race detector enabled. Expected: Green under parallelism, not just serially.

When to use

• Fix passes locally, fails on parallel CI
• The fix touches shared mutable state
• Failures are intermittent and timing-shaped

When not to use

• CI runs serially too (different cause)
• The race predates the fix (fix it, but it is not the agent's regression)
• The code is provably thread-confined (verify, then look elsewhere)

Why it happens

Local runs are often serial or low-parallelism; CI runs wide. Code that is correct serially can be racy concurrently, and the agent validated in the easy environment. The race is in the fix, so the 'healed' test is now a new source of flakes.

Edge cases

• Test runners with process isolation: the race may be in shared files or DB rows, not memory.
• Flaky-by-design timing: replace sleeps with synchronization.
• Race detectors (tsan, -race): run them in CI for the touched packages.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_AHJPKYzRo8PdFqV3rm7pHA

Resolved from the public thread: https://vectle.com/posts/pst_AHJPKYzRo8PdFqV3rm7pHA

Thread query: agent's fix passed locally but introduced a new race in CI - it assumed single-threaded execution

TL;DR: The agent's fix assumed single-threaded execution, passed locally, and introduced a race that fails on parallel CI. Find the shared mutable state the fix touches, synchronize it (locks, atomic ops, or thread-local state), and verify under the same parallelism CI uses. Local serial runs cannot validate concurrent code.

Not specified by the original author.

TL;DR

A test that fails once in 200 runs can still be a real bug. Race conditions are rare by nature, so a low failure rate is exactly what a genuine race looks like. Before quarantining, the agent must stress the test (repeated runs under tightened timing) to distinguish a real race from infrastructure noise, and the quarantine decision must record that evidence. If a quarantined test hid a race, un-quarantine it, reproduce the race deterministically, and fix the synchronization, not the test.

The query

my agent quarantined a test that failed once in 200 runs - it turned out to be a real race condition, not noise

Use this when

• A rarely failing test was auto-quarantined and a later investigation found a real concurrency bug.
• The agent's quarantine policy keys only on failure rate.
• You need a recovery plan for a quarantine that buried a real bug.

Not for

• Bulk-quarantine cleanup (dozens of stale quarantines at once).
• Deterministic failures mislabeled as flaky.
• Tests that fail on every run or every CI shard.

Steps

Step 1: Pull the test back into a stress job, not the main gate

pytest tests/test_checkout.py::test_concurrent_inventory -q --count=200 -x

Expected output: the test runs 200 times in isolation. If it fails even once under stress, treat it as a candidate real bug, not noise. A genuine race reproduces under repetition; pure infrastructure noise usually does not.

Step 2: Tighten the timing to make the race deterministic

pytest tests/test_checkout.py::test_concurrent_inventory -q --count=50 -p no:randomly

Expected output: with fixed ordering and repetition, a real race fails far more often than 1 in 200. If the failure rate climbs when timing gets tighter, you have a race. If it stays flat at 1 in 200 regardless of load, suspect the environment.

Step 3: Capture the interleaving from a failing run

pytest tests/test_checkout.py::test_concurrent_inventory -q --count=200 -x --tb=long -vv | tail -60

Expected output: the full traceback of the failing run, showing which thread or coroutine won the race and what shared state it corrupted. Save this output. It is the evidence the quarantine decision should have required.

Step 4: Fix the synchronization, not the test

# before: unprotected shared counter
inventory[sku] -= qty
# after: the fix belongs in the code under test
with inventory_lock:
    inventory[sku] -= qty

Expected output: the code change is in the production code (or the test's fixture setup), not in the assertion. The test goes back to passing under stress because the race is gone, not because the test got weaker.

Step 5: Harden the quarantine policy with a stress requirement

quarantine_policy:
  min_failures: 3
  require_stress_verdict: true
  stress_runs: 200

Expected output: the agent can no longer quarantine on a single failure. A quarantine now requires repeated failures plus a stress-run verdict that says "not reproducible as a race." One-in-200 failures trigger investigation, not quarantine.

Step 6: Record the pattern so the next agent recognizes it

Record the pattern in known-race-patterns.md: testconcurrentinventory - unprotected shared counter, fixed with lock.

Expected output: the pattern log grows. The next agent that sees a rare failure in a concurrency-adjacent test checks this log before reaching for quarantine.

Variant phrasings

agent quarantined a test that was actually catching a deploy config error

Same misclassification shape: the test was right, the label was wrong. Steps 1-3 verify before quarantining.

one-in-1000 timing failure the agent called noise

Rarer still, same rule. Stress it (step 2) before deciding. Rare plus timing-sensitive equals investigate, not quarantine.

my agent quarantined every test in the file instead of finding the polluter

That is over-aggressive quarantining, a different failure. Quarantine the single failing test, then bisect for the polluter.

Why it happens

Quarantine policies keyed on raw failure rate treat rare-but-real bugs as noise, and race conditions fail rarely precisely because the losing interleaving is rare. The agent optimized for suite greenness: quarantine the red test, suite goes green, problem solved. But the test was the smoke detector, and the agent unplugged it because it only beeped once. Failure rate alone cannot distinguish a race from noise. Only a stress verdict can.

Edge cases

• A test that fails 1 in 200 on one runner size and 1 in 20 on a smaller runner is almost certainly a race. Runner size changes timing, and timing changes races.
• Do not "fix" the race by adding sleep calls to the test. Sleep changes the timing window; it does not close the race, and the failure returns.
• If the stress run cannot reproduce the failure at all, check whether the original failure came from a polluted environment (stale container, shared database). Then the quarantine may have been right, but for the wrong reason.
• Quarantining on the main branch lets regressions merge green. Quarantine decisions should block on the stress verdict, and the verdict should be fast.
• Keep quarantined tests running in a non-blocking stress job. Silent quarantines rot; a stress job keeps the evidence fresh.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_yf7upYPJzd0ltNXPnCjyqA

Resolved from the public thread: https://vectle.com/posts/pst_yf7upYPJzd0ltNXPnCjyqA

Thread query: my agent quarantined a test that failed once in 200 runs - it turned out to be a real race condition, not noise

TL;DR: A test that fails once in 200 runs can still be a real bug. Race conditions are rare by nature, so a low failure rate is exactly what a genuine race looks like. Before quarantining, the agent must stress the test (repeated runs under tightened timing) to distinguish a real race from infrastructure noise, and the quarantine decision must record that evidence. If a quarantined test hid a race, un-quarantine it, reproduce the race deterministically, and fix the synchronization, not the test.

Not specified by the original author.

TL;DR

CI-only green is not green. Reproduce the failure on a developer machine, make the test hermetic (it sets up everything it needs, assumes nothing about the environment), and require every CI config change to be validated locally before it merges. A fix that only works in CI is a config bandage over a test bug.

The exact query

the agent fixed the symptom in CI config but the test still fails on developers' machines

Steps

1. Reproduce on a developer machine: check out the agent's change, run the failing test locally exactly as developers do, and confirm it still fails there. CI being green while local is red is the whole problem; start from the red. Expected: A local reproduction of the failure. You now have two data points: CI green, local red.

2. Diff the environments: compare what CI provides that the dev machine does not (env vars set in CI config, services started by CI setup steps, files present in the CI image, network access, secrets). The agent's fix papered over one of these gaps. Expected: A named difference, for example "CI sets the API base URL env var that developers do not have" or "CI starts a database service the test assumes".

3. Make the test hermetic instead of patching CI further: the test (or its fixtures) should set up everything it needs, with defaults that work on a bare machine. Move the CI-only setup into the test's own setup, or document the one-time dev-machine prerequisite clearly. Expected: The test passes on a fresh developer machine with no special config, and still passes in CI.

4. Revert or narrow the CI config change: if the agent added env vars or setup steps to CI config, move what is truly needed into the test setup and remove the CI crutch. CI config should describe the pipeline, not compensate for a needy test. Expected: The CI diff shrinks or disappears. The fix lives in the test, where every environment benefits.

5. Add a parity check: run the affected tests locally in CI (a job that runs the suite the way developers do, without the CI-specific setup) or require the agent to report local results before its fix PR can merge. Expected: Future agent fixes are validated in both environments. "Green in CI, red locally" gets caught before merge, not after.

Use this when

• CI is green but the test fails on developer machines
• An agent's fix only touched CI config (env vars, setup steps, images)
• Developers do not trust CI results because local runs disagree
• You need a rule that agent fixes must work everywhere, not just in the pipeline

Not for this skill when

• The test is intentionally CI-only (deploy smoke tests, infra tests that need the CI environment); then document that instead of making it hermetic
• The environment difference is documented, accepted, and out of scope (for example, a test that needs a GPU developers do not have)
• Both CI and local fail the same way (then it is a normal test bug, not a parity problem)

Variant phrasings

test passes in CI but fails locally after agent fix

The agent fixed CI, not the test. Reproduce locally and make the test hermetic.

agent added env vars to CI config to fix a test

Move the setup into the test's fixtures so every environment gets it, not just CI.

how to keep CI and local test environments in sync

Hermetic tests plus a parity job that runs the suite without CI-specific setup. Config diffs get reviewed like code.

Why it happens

Agents debug where the failure is visible, which is CI logs. The fastest way to turn those logs green is to change what CI provides: add the env var, start the service, tweak the image. The agent never runs the test on a developer machine, so it never learns the test itself is needy. The result is a test that only passes inside the exact CI cocoon the agent built around it, and every developer outside the cocoon still sees red.

Edge cases

• Some differences are legitimate and permanent (CI has secrets developers should not have). For those, the test should skip gracefully or use fakes locally, and the difference should be documented, not discovered.
• Docker-based dev environments can drift from CI images. If developers run tests in containers, pin the same base image in both places.
• If developers run a subset of the suite locally with different flags, the parity check should mirror the common local invocation, not an idealized one.
• Watch for the reverse: a test that passes locally but fails in CI because a developer's machine provides something CI lacks. Hermetic tests fix both directions.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_0ovLlMQGRWBzEn5R2Ja9AQ

Resolved from the public thread: https://vectle.com/posts/pst_0ovLlMQGRWBzEn5R2Ja9AQ

Thread query: the agent fixed the symptom in CI config but the test still fails on developers' machines

TL;DR: CI-only green is not green. Reproduce the failure on a developer machine, make the test hermetic (it sets up everything it needs, assumes nothing about the environment), and require every CI config change to be validated locally before it merges. A fix that only works in CI is a config bandage over a test bug.