راقب اتجاهات التقنية العالمية، وجمع موارد المطورين

`If your team clicks Re-run failed jobs more than once a week, your end-to-end suite has stopped telling you the truth. Red no longer means "you broke something". It means "something was slow, probably".

I built QAFind to fix that. It's an open-source (MIT) CLI that attaches to the Playwright suite you already run. You don't rewrite specs, adopt a new runner or sign up for anything.

What it does, in one sentence each

attach wraps your existing playwright.config.ts. Your config and specs are read, never written.
verify-merge runs the specs a PR touches, reruns failures once, and fails CI only on a regression that reproduces.
flake aggregates runs and tells you which tests flake, the probable cause, how confident it is, and who owns them.
attach --determinism records and replays network traffic and freezes the browser clock, so fewer tests flake in the first place.
Signed evidence: every verdict can be signed (HMAC or Ed25519) and exported as an in-toto attestation you can verify offline.

60-second start

From the root of any Playwright project:

bash npx qafind attach . npx playwright test --config .qafind/playwright.qafind.config.ts npx qafind flake .qafind/reports --markdown

That's it. The first command writes a wrapper config into .qafind/. The second is your normal test run with one extra reporter. The third gives you a table like this:

text | Test | Runs | Flakes | Cause | Quarantine | | cart fixture returns totals | 3 | 1 | network_fixture | yes | | login selector exists | 3 | 1 | selector_drift | yes | | slow recommendations endpoint | 3 | 1 | timing | yes | | known checkout total mismatch | 3 | 3 | app_race | no |

Notice the last row: it failed every time, so it's a real failure and it is never suggested for quarantine.

The merge gate

Add one step to CI:

bash npx qafind verify-merge --run --base origin/main --head HEAD

First run	Rerun	Verdict	Blocks merge?
failed	failed	regression	yes
failed	passed	flake	no (but it's recorded)
failed	different failure	unreproduced	no

Flakes stop holding your PRs hostage. Real regressions still stop the merge. Add --markdown for a PR comment, or --intoto for a signed attestation.

Why not just use retries?

retries: 2 turns a flaky failure into a green exit code and throws away the information. Nobody counts how often a test needed a retry, and an intermittent bug your users will hit looks identical to a slow CI runner. QAFind keeps that information, counts it, and puts it in front of a human. Only the reproduced failures block.

What it isn't

I'd rather you trust it than be surprised by it:

Playwright only, for now.
Cause labels are suggestions, based on error text plus an optional request log. On a synthetic benchmark (400 tests × 5 runs) they're 87.8% accurate on text alone and 90.3% with the request log. Check before you quarantine.
WebSockets and service workers aren't recorded by the network replay yet.
Not a test generator. It makes the tests you already have more trustworthy.

Free for open source

The CLI is MIT and fully local: no account needed for attach, flake or verify-merge.

There's also an optional hosted GitHub App (QAFind CI) that runs QA on pull requests and comments the report on the PR. Public repositories are free with no daily quota.