Mutation testing, minus the pain

Break your tests before prod does.

BugBait plants realistic bugs in your code, runs your test suite against each one, and writes verified tests for every gap it finds. Your code is never touched — it all happens on throwaway copies.

Test your project View on GitHub

Free and open source. Your code never leaves your machine.

languages: 10
bug kinds: 9
lines of your code touched: 0

bugbait — my-project

$ python3 main.py ~/code/my-project --bugs 2
Baseline: 4 tests pass

Bug  1 SURVIVED  age >= 18  →  age > 18   [boundary, 5/5]
Bug  2 SURVIVED  or  →  and               [operator]

Score: 0/2 bugs caught (0%)

Fixer is writing tests for 2 survivors...

  [+] verified test added  boundary at 18
  [+] verified test added  broken validation

Score: 0% → 100%  (0/2 → 2/2 bugs caught)

Three steps to trustworthy tests

The Saboteur plants bugs

An AI agent studies your code and plants small, realistic mistakes — off-by-ones, flipped operators, dropped guards. Each gets a realism rating; artificial ones are thrown out.
The sandbox scores your tests

Every bug runs against your full suite in its own throwaway copy. Caught means a test failed; survived means a real gap. Your score is simply caught ÷ planted.
The Fixer writes the missing tests

A second agent writes a test per survivor — each verified to pass on your code and fail on the bug. Add them to your project or open a pull request.

Meet the bugs

Nine bug kinds, each a one-line change that keeps the code valid — the kind of thing that survives code review. Pick your language:

Kind	Python	JavaScript	TypeScript	Go	Java	Rust	Ruby	PHP	C#	Kotlin
off-by-one	`range(n)` → `range(n + 1)`	`i < len` → `i <= len`	`i < len` → `i <= len`	`i < n` → `i <= n`	`i < n` → `i <= n`	`i < n` → `i <= n`	`(0...n)` → `(0..n)`	`$i < $n` → `$i <= $n`	`i < n` → `i <= n`	`i < n` → `i <= n`
boundary	`age >= 18` → `age > 18`	`age >= 18` → `age > 18`	`age >= 18` → `age > 18`	`age >= 18` → `age > 18`	`age >= 18` → `age > 18`	`age >= 18` → `age > 18`	`age >= 18` → `age > 18`	`$age >= 18` → `$age > 18`	`age >= 18` → `age > 18`	`age >= 18` → `age > 18`
wrong operator	`and` → `or`	`\|\|` → `&&`	`\|\|` → `&&`	`&&` → `\|\|`	`&&` → `\|\|`	`&&` → `\|\|`	`&&` → `\|\|`	`$a && $b` → `$a \|\| $b`	`&&` → `\|\|`	`&&` → `\|\|`
inverted condition	`if valid:` → `if not valid:`	`if (ok)` → `if (!ok)`	`if (ok)` → `if (!ok)`	`if (ok)` → `if (!ok)`	`if (ok)` → `if (!ok)`	`if ok` → `if !ok`	`if ok` → `unless ok`	`if ($ok)` → `if (!$ok)`	`if (ok)` → `if (!ok)`	`if (ok)` → `if (!ok)`
wrong default	`n=10` → `n=0`	`n = 10` → `n = 0`	`n = 10` → `n = 0`	`n := 10` → `n := 0`	`n = 10` → `n = 0`	`let n = 10` → `let n = 0`	`n = 10` → `n = 0`	`$n = 10` → `$n = 0`	`n = 10` → `n = 0`	`n = 10` → `n = 0`
swapped arguments	`f(a, b)` → `f(b, a)`	`f(a, b)` → `f(b, a)`	`f(a, b)` → `f(b, a)`	`f(a, b)` → `f(b, a)`	`f(a, b)` → `f(b, a)`	`f(a, b)` → `f(b, a)`	`f(a, b)` → `f(b, a)`	`f($a, $b)` → `f($b, $a)`	`f(a, b)` → `f(b, a)`	`f(a, b)` → `f(b, a)`
missing case	a guard clause removed
wrong value	`timeout=30` → `timeout=300`	`timeout: 30` → `timeout: 300`	`timeout: 30` → `timeout: 300`	`timeout := 30` → `timeout := 300`	`timeout = 30` → `timeout = 300`	`timeout = 30` → `timeout = 300`	`timeout = 30` → `timeout = 300`	`$timeout = 30` → `$timeout = 300`	`timeout = 30` → `timeout = 300`	`timeout = 30` → `timeout = 300`
error handling	`except ValueError` → `except Exception`	`throw …` → `return null`	`throw …` → `return null`	`if err != nil` check dropped	`catch (IOException)` → `catch (Exception)`	a `Result` check dropped	`rescue ArgumentError` → `rescue StandardError`	`catch (ValueError)` → `catch (Exception)`	`catch (IOException)` → `catch (Exception)`	`catch (IOException)` → `catch (Exception)`

Speaks your language

Two ways to run it: the dashboard (point-and-click, pick repos after connecting GitHub) or the command line (scriptable, with merge gates and a pre-commit hook).

Running in two minutes

Clone it and install the four dependencies:

git clone https://github.com/AVI2907/bugbait.git
cd bugbait && pip install -r requirements.txt

Add your model key to .env:
```
META_API_KEY=your_key_here
```
Run it on your own project:
```
python3 main.py ~/code/my-project
```

Then open the dashboard and test anything:

python3 app.py   # opens http://localhost:5050

Your code stays on your machine. Nothing is pushed anywhere unless you click “Open pull request”.

Break your tests before prod does.

Three steps to trustworthy tests

The Saboteur plants bugs

The sandbox scores your tests

The Fixer writes the missing tests

Meet the bugs

Speaks your language

Running in two minutes