stryker-mutation

Overview

Per stryker-intro:

"Stryker started as a pure JavaScript mutation testing framework" and now "supports most JavaScript projects, including TypeScript, React, Angular, VueJS, Svelte, and NodeJS."

Mutation testing introduces small bugs (mutants) into the production code; if tests still pass, the test suite is too weak. A green suite with high mutation-survival rate is the bug coverage hides.

When to use

• A JS/TS suite has ≥80% line/branch coverage and the team wants to verify assertions actually catch bugs.
• Code review keeps surfacing weak assertions in tests (.toBeTruthy() etc); mutation testing makes the cost visible.
• A critical module needs a higher quality bar than coverage alone proves.

Step 1 - Install

npm install --save-dev \
  @stryker-mutator/core \
  @stryker-mutator/jest-runner   # or mocha-runner / vitest-runner / etc.

Per stryker-intro, supported runners include Jest, Mocha, Karma, Vitest, Jasmine, Cucumber, Tap.

Step 2 - Initialize

npx stryker init

Generates stryker.conf.json:

{
  "$schema": "./node_modules/@stryker-mutator/core/schema/stryker-schema.json",
  "packageManager": "npm",
  "reporters": ["progress", "clear-text", "html"],
  "testRunner": "jest",
  "coverageAnalysis": "perTest",
  "mutate": ["src/**/*.ts", "!src/**/*.test.ts"],
  "thresholds": { "high": 80, "low": 60, "break": 50 }
}

Key fields:

Field	Use
mutate	Glob patterns: which files to mutate; exclude tests + types.
testRunner	jest / mocha / vitest / karma / etc.
coverageAnalysis	perTest (fastest - only re-run tests that touched the mutated line).
thresholds.break	Mutation score below this → npx stryker run exits non-zero (CI gate).
incremental	true to only mutate files changed since last run.
concurrency	Worker count (default = CPU count).

Step 3 - Run

npx stryker run

Per-mutant output:

[Survived] Conditional Boundary
src/cart.ts:42:5
- if (item.qty < 0) throw new Error(...);
+ if (item.qty <= 0) throw new Error(...);
Tests run: 12 (all passed - mutant survived).

Survived mutant = tests passed despite the introduced bug = the tests don't actually catch this regression. Add a test for qty = 0.

Step 4 - Report + threshold

Ran 142 tests.
Mutants killed:    198
Mutants survived:   24
Mutants timed out:   3
Mutants no coverage: 12

Mutation score: 84.7%

Per stryker-intro: thresholds via the thresholds object gate the build:

• >= high (80) → green.
• >= low (60) → yellow.
• < break (50) → red, exit code != 0.

The HTML report shows per-file mutation breakdown - drill in to see which mutants survived and where to add tests.

Step 5 - Incremental mode for PR runs

Full mutation runs are slow (5-30 min on medium codebases). For PRs, incremental mode only mutates changed files:

npx stryker run --incremental

This stores state in .stryker-tmp/ (commit to git so subsequent runs benefit). Per-PR runs typically complete in <2 min.

Step 6 - CI integration

- name: Mutation testing (changed files only)
  if: github.event_name == 'pull_request'
  run: npx stryker run --incremental

- name: Full mutation run
  if: github.ref == 'refs/heads/main' && github.event.schedule == '0 2 * * 0'
  run: npx stryker run

- uses: actions/upload-artifact@v4
  if: always()
  with:
    name: stryker-report
    path: reports/mutation/

Pattern: incremental on PRs (fast feedback) + full weekly run on main (catches drift in unchanged code).

Anti-patterns

Anti-pattern	Why it fails	Fix
Setting break: 100	Impossible bar; team disables.	Start at 50-60; ratchet up over time.
Running on every PR without --incremental	30-min PR feedback loop; team disables.	--incremental for PRs (Step 5).
Mutating test files	Mutants in test code don't reflect production quality.	Exclude in mutate glob (Step 2).
Ignoring "no coverage" mutants	Files in mutate glob without any test coverage; mutation testing wastes effort.	Either add tests OR exclude those files from mutate.
Using coverageAnalysis: off	Re-runs entire suite per mutant; very slow.	perTest (the default; Step 2).
Hard-coding break threshold without team agreement	Fails surprise PRs; team disables.	Land threshold via PR with team review.

Limitations

• Slow for large codebases. Even with incremental + perTest, mutation testing scales worse than test execution.
• Surviving mutants don't always indicate weak tests - some are equivalent mutants (semantically identical to the original); flag and exclude these.
• TypeScript checking adds time. Disable type-check (tsconfig.tsbuildinfo) for mutation runs if not needed.
• Per-runner integration quality varies. Jest is the most mature; Vitest / Karma have known edges.

References

• si - StrykerJS overview, supported test runners, supported frameworks (TS / React / Angular / Vue / Svelte / Node).
• stryker-net-mutation - .NET sibling.
• pitest-mutation - Java sibling.
• mutmut-mutation - Python sibling.
• mull-mutation - C/C++ sibling.