Workers

For teams with an existing Playwright suite

Move your Playwright tests to Kane. Prove nothing was lost.

Workers rewrites each test as plain English, breaks your app on purpose, and checks that the new tests catch the same bugs as the old ones.

01 The idea

One line, translated

A selector-based assertion goes in. A sentence a person can read comes out. The check stays the same.

From the demo suite

A Playwright assertion becomes a plain-English Kane line Two lines of Playwright code that read the todo counter and expect the text "3 items left", followed by the Kane line "Verify the exact visible text of the todo counter in the footer is 3 items left". Playwright, from the demo suite const todoCount = page.getByTestId('todo-count') await expect(todoCount).toHaveText('3 items left'); Kane CLI, same check in English Verify the exact visible text of the todo counter in the footer is "3 items left".
Real lines from the bundled demo suite and its ported Kane test.

02 The problem

Rewriting a test suite is slow, and AI rewrites are hard to trust

Selector tests are brittle. Rename a test id and the test fails, though nothing a user sees has changed.

Rewrites take weeks. Moving a suite to a new tool by hand is dull work that rarely gets scheduled.

AI rewrites are not trusted. A generated test that passes is not proof that it still catches bugs.

54/100trust in AI-generated tests

50%of AI-testing budget goes to generating tests

17%goes to verifying them

Source: TestMu AI, State of AI in Quality Engineering. Workers puts the effort on the verifying side.

03 Break it

Break the app. See who notices.

Workers applies one deliberate bug at a time (a mutant), runs the old Playwright test and the new Kane test, then undoes the bug. Pick a mutant to see what happened.

Replay Recorded results from a real run on 16 Nov 2026. Nothing here runs live.

Real run, 16 Nov

Bugs

Harmless refactors

Mini todo app used in the replay todos What needs to be done?
Clean app. Pick a mutant above.

No mutant applied

Each button applies one change from demo/workers.mutants.json and reveals the recorded verdicts below.

MutantTestPlaywrightKane

04 The repair loop

When the new test is weaker, the agent rewrites it

The agent sees what the old test asserted and what Kane did. It never sees the bug, so it cannot tune the test to one mutant.

  1. Miss

    1. Kane missed it

    Playwright failed on bug B3 (the counter says "3 item left"). The first Kane translation passed. A weaker test.

  2. Rewrite

    2. The agent rewrites it

    Input: Playwright's assertion and Kane's step results. Round 1 was rejected: it failed on the clean app. Round 2 fixed it.

  3. Caught

    3. Kane catches it

    The re-run passes on the clean app and fails on B3. Every other result held. Status: repaired.

Round 1, rejected (broke on the clean app)

Verify visible text "3 items left" is shown on the page.

Round 2, accepted (caught B3)

Verify the exact visible text of the todo counter in the footer is "3 items left".

Kane authoring is not deterministic, so the same translation can differ between runs. That is why Workers verifies instead of trusting.

05 The proof

What the real run showed

Demo app, 6 ported tests, 5 bugs and 2 harmless refactors. Run on 16 Nov 2026 with Kane CLI and a Codex agent.

  • 6tests ported
  • 7 of 7bug catches matched: whenever Playwright caught a bug, so did Kane, across 30 test-and-bug pairs
  • 2false alarms avoided: Playwright failed on a renamed test id, Kane stayed green
  • 1test repaired by the loop
  • 1Warning planted bug (B2, "1 items left") slipped past both suites. The report says so.
  • 389.8Kane credits for the whole run

Read the full report Matrix, repair log, Kane dashboard links and the generated test files.

Read the shop report A second app with page objects, fixtures and three spec files: 10 of 10 tests ported, 8 of 8 bug catches matched, 5 false alarms avoided.

06 Quickstart

Try it on the bundled demo

Install from npm in any empty folder. A dry run translates one test with your coding agent and runs Playwright on every break, but skips Kane, so it costs no Kane credits.

$ npm install workers-port
$ npx playwright install chromium
$ npx workers-port demo --dry-run --max-tests 1

Workers runs with Claude Code or Codex as the coding agent (--agent claude or --agent codex). A Claude Code skill in the repo's skill/ folder wraps the same command. Prefer the source? Clone the repo and run node bin/workers.js demo.

Built with

Kane CLI by TestMu AI runs the plain-English tests. Playwright runs the original suite and lists its tests. Claude Code or Codex writes and repairs the Kane tests.

See it on GitHub