Saboteur tests your tests

It plants real bugs in your web app one at a time, lets Kane CLI run your tests against each broken build, and shows which bugs slipped through. Then Claude writes a new Kane test for every miss, and Kane only keeps it if it passes on the correct app and fails on the bug.

Planted bugs caught
0/10 → 10/10
4 starter tests, then with Claude's Kane-proven tests
Duel: AI attacker vs the suite
Red 3 : Blue 3
Red found 3 bugs the suite missed; blue closed 3 with Kane-proven tests
Test suite
4 → 21
every added test proven by Kane CLI
Animated illustration of the real runs below.

How it works

1 · Plant

A small, realistic bug as a code patch: a wrong total, a dead button, a bad redirect. One at a time, on a fresh build.

2 · Hunt

Kane CLI replays your _test.md suite in real Chrome with healing off, so a broken page fails instead of being worked around.

3 · Score

Caught or slipped, per bug. Browser crashes and timeouts are errors, never catches, and every catch is re-run to confirm it.

4 · Close the gaps

Claude writes a test for each miss. Kane runs it on the correct app (must pass) and the bugged app (must fail). Kane's failure text goes back to Claude if not.

Run it yourself

git clone https://github.com/softalpha0/saboteur
cd saboteur
pnpm i
kane-cli login
pnpm saboteur doctor --replay
pnpm saboteur run --replay

Replay mode needs no Claude key: it uses the committed, Kane-proven tests. Point it at your own app with saboteur.config.json.