Green checks can lie. See what your tests miss.
BonaFident makes small changes to the code in your pull request, the way a bug would, and checks whether your tests fail. It is mutation testing for JavaScript and TypeScript, scoped to the diff. It runs on your CI runner, and only the report is uploaded.
No account needed. Runs on your machine with Node 22+ and Jest or Vitest:
$ npx bonafident@latest audit --local
Runs on your CI runner. By default only the report is uploaded, never your source text.
-
GitHub Actions
Native merge gate (the BonaFident check), OIDC attestation, Evidence Ledger and a pull request comment. No API key.
-
GitLab CI
Merge request note and report upload. The job fails on a policy breach.
No merge gate check, attestation or ledger yet.
-
Bitbucket Pipelines and any CI with Node 22+
Runs a full audit, uploads the report and fails the job on a policy breach.
No pull request comment, attestation or ledger yet.
A passing build is a claim. This is the evidence.
Line coverage says the code ran. Mutation testing says whether a test would notice if the code were wrong. Drag the slider to compare both views of the same illustrative pull request.
The tests killed 34.1% of the mutants. The three below survived: the tests ran each changed line and still passed. In enforce mode the merge waits for a fix.
Jest or Vitest reports 14 passing tests. Most lines ran, so the coverage report looks healthy, but it cannot say whether a test would fail if a line were wrong.
BonaFident changed the code and at least one assertion failed. That is what a working test does.
The tests executed the changed code and passed anyway, for example because they assert only on their own mocks.
The changed lines never ran under any test, so BonaFident lists them separately instead of scoring them.
Illustrative example, not a measured run. The three shares add up to all the mutants in this example.
Don't ask one AI to grade another.
A second model can share the first one's blind spots, and it bills per token. BonaFident runs your own tests against small changes to your code, so there is no model to pay for.
No second AI grading the first
BonaFident does not send your repository to a model and ask whether a test looks right. It changes the code and lets your tests answer.
Your tests, on your runner
Every finding names the line, the change and the test result behind it, so you can check it yourself.
An audit that runs out of time says Incomplete, never Passed.
On a pull request BonaFident changes only the lines the diff touched and runs only the tests that cover them, within the time budget in your bonafident.yml.
Pull-request check
Every pull request gets its own verdict, scoped to the lines it changed.
Scheduled full sweep
A scheduled run over the whole codebase instead of one diff, for a periodic picture of where tests are thin.
Built for the stack you already ship.
BonaFident runs your Jest or Vitest tests in-process instead of restarting them for every change. That runtime is JavaScript and TypeScript today, on Node 22 or newer.
Not supported yet: Python, Go, Rust, Java. Tell us which one you need first.
Stop merging
on faith.
One command on your own machine. No GitHub App, no API key, and your source code never leaves it.