Kill the mutants your tests miss.

/ Reference measurement

Reference measurement

faster Stryker run after one ignore pattern: 38.4 s → 12.5 s, same results

3x

of a small diff run happens before the first test runs

85%

score of the diff, against 87.0% package-wide in Stryker's own report

100%

Measured on one package of a pnpm + Turborepo monorepo (~55k files incl. build caches), StrykerJS 9.6.1 and Vitest 4, medians of 5 runs on a loaded 22-thread Linux machine. The 3x comes from adding .turbo to Stryker's ignorePatterns (Turborepo's cache was 97% of the files Stryker copied into its sandbox), not from Reapr. The 87.0% includes results that incremental: true carried over from outside the diff; the diff's own 3 mutants were all killed. Read the full bench report, or the guide How to speed up a slow Stryker run.

/ Why

Coverage measures execution, not verification.

Teams ship a lot of AI-generated code, and agents write tests that pass. The mutation score tells you whether those tests protect behavior. Time keeps it out of pull requests: a full run takes 40 to 90 minutes. Reapr scopes it to the diff.

More in the guides Mutation testing for AI-generated code and Mutation testing on pull requests.

Line coverage asks

Did a test run this line?yes, 100%

The mutation score asks

If this line changed, would a test fail?no, it survived

/ How it works

Your engine, your config, only your diff.

1. Read the diff

   git diff <base>...HEAD gives the changed line ranges.

   src/a.ts:8-12
   src/b.ts:5-6
   docs/usage.md:3-4

2. Run Stryker on it

   Your config's mutate globs pick the files. A wrapper imports your config and sets only mutate and a json reporter.

   src/a.ts:8-12
   src/b.ts:5-6
   docs/usage.md:3-4

3. Score the diff

   Reapr keeps the mutants inside the changed ranges, computes the score and writes the run record.

   score 80.00%
   2 survived

File names are an example. Your test runner, thresholds, timeouts, checkers and plugins stay as you wrote them. What Reapr changes in your config. Why the score covers only the diff: Stryker incremental mode and package-wide scores.

/ Output

A summary for people. JSON for agents.

Each survivor has the file, line, column, mutator and replacement: enough for an agent to write the missing test and run Reapr again. Reapr reports; your agent writes the missing test. The mutation-audit agent skill does that with fixed rules.

rpr · stryker · diff vs origin/main · 3 ranges in 2 files
score 80.00%  (8 killed, 0 timeout, 2 survived, 0 no coverage)  1m 23s
survivors:
  src/a.ts:10:3  ConditionalExpression  → false
  src/b.ts:6:2  ConditionalExpression  → false
run: /home/me/repo/.git/reapr/runs/1793640000-4242

"survivors": [
  { "file": "src/a.ts", "line": 10, "column": 3,
    "mutator": "ConditionalExpression",
    "replacement": "false", "status": "Survived" },
  { "file": "src/b.ts", "line": 6, "column": 2,
    "mutator": "ConditionalExpression",
    "replacement": "false", "status": "Survived" }
]

Samples from the output section of the README. Exit codes: 0 passed, 1 below --min-score or the engine's threshold failed, 2 error.

/ GitHub Action

Run it on every pull request.

reapr-corp/run@v1 writes a job summary with the score, counts, survivors and warnings, and exits with Reapr's exit code. Linux runners. No token needed.

on: pull_request

jobs:
  rpr:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5
        with:
          fetch-depth: 0 # full history, so origin/<base>...HEAD has a merge base
      - uses: actions/setup-node@v5
        with:
          node-version: 22
          cache: npm
      - run: npm ci # rpr runs your engine (StrykerJS) with your installed deps
      - name: rpr
        uses: reapr-corp/run@v1
        id: rpr
        with:
          min-score: 80

      - name: rpr
        uses: reapr-corp/run@v1
        id: rpr
        with:
          configs: |
            # one per package; blank lines and # comments are ignored
            stryker.api.config.mjs
            stryker.web.config.mjs
            stryker.worker.config.mjs

/ Principles

What Reapr does and does not do.

We invent nothing

Reapr runs your engine config as written and adds only the diff scope and a JSON reporter. It forces no flags. When a setting can miscount results, it warns and changes nothing.

Agents are first-class users

--json prints one document with every surviving mutant, and the mutation-audit agent skill turns it into tests. An MCP server is planned.

Every run leaves raw data

Run facts, both reports and the wrapper config go to .git/reapr/runs/<run_id>/ on your machine. Reapr never uploads or deletes them.

Anonymous, opt-out telemetry

One event per run: outcome, counts, durations, bucketed diff size. Never code, paths or names. DO_NOT_TRACK=1 turns it off. See every telemetry property.

/ Engines

StrykerJS and cargo-mutants today. More engines planned.

Reapr runs mutation engines; it does not replace them. Each engine plugs in through one adapter. Add an adapter.

/ Open core

Free to use, modify and self-host.

The CLI, the Action and the engine adapters are open source and work on their own. A hosted layer is planned as the paid product.

/ Reapr Cloud

Reapr Cloud is planned. Join the waitlist.

A hosted layer on top of the open source tool: faster runs on warm workers, a dashboard of mutation scores across pull requests, and a bot that proposes tests for surviving mutants. None of it is built yet. The CLI and the Action stay free and work without it.