testpick

Run only the tests your diff can actually break.

Website: devtoolshq.net/testpick — overview, feature tour, and FAQ for coverage-based test selection with Jest, Vitest & monorepos.

testpick is a test-selection CLI for JavaScript/TypeScript. It looks at what you changed (git diff) and runs just the tests affected by those changes — turning multi-minute CI runs into seconds.

npx testpick map     # one-time: learn which tests touch which code
npx testpick run     # from now on: run only what your changes affect

$ testpick map
Mapping 23 test file(s) with vitest in 8 single-pass shard(s).
✔ Map saved to .testpick/map.json — 24 source files tracked.

$ vim src/util.ts        # change one file...

$ testpick run
testpick: 1/23 test file(s) affected by 1 change(s).

 ✓ src/greet.test.ts (1 test) 4ms

$ testpick explain       # ...and see *why*
Changed files (1):
  • src/util.ts
Decisions:
  ✓ src/util.ts  [coverage map → 1 test(s)]
      → src/greet.test.ts
Result: run 1 of 23 test file(s).

Want a GIF for your README/socials? A ready-to-run vhs script lives at assets/demo.tape: brew install vhs && vhs assets/demo.tape.

Those are great — until your code has couplings their static import graph can't see:

• a module loaded via a runtime-computed path — a plugin registry, DI container, or import(/* @vite-ignore */ pathFromConfig). Vite can glob a literal import(\./${name}.js`)`, but it cannot analyze a path that's data.
• code reached only at runtime that the static graph over- or under-counts

testpick builds its map from runtime coverage — what each test actually executed — so it captures those edges. Verified example (see testpick-real fixture):

const REGISTRY = { feat: "../features/feat.ts" };
export const load = (n) => import(/* @vite-ignore */ REGISTRY[n]); // Vite can't see this

Note: testpick and Vite's module graph are complementary, not strictly better. A coverage map is more precise for runtime-computed couplings and for trimming imported-but-unexecuted modules; the static graph can flag a yet-to-run branch a coverage map hasn't seen. That's why testpick always errs toward running more (see below) — and why it works the same for Jest, where there's no Vite graph.

A test selector is only useful if you can trust it not to skip something important. testpick's rule: when in doubt, run more — never less.

• Changed a file the map doesn't know about (new file, config)? → it runs all tests by default.
• Pass --ai and it asks an LLM to narrow those unmapped changes to likely tests — but if the model is unsure, it still falls back to running everything. The AI can never cause a skip.
• testpick explain shows exactly why each test was selected or skipped.

testpick map [--base <ref>]      # build/refresh the coverage map
testpick run [--base <ref>]      # run only affected tests
testpick explain [--base <ref>]  # dry-run: print the selection + reasoning

testpick map is built for speed three ways:

• Single-pass (Vitest & Jest, default): instead of starting the runner once per test file, testpick shards the files across your cores and runs each shard as one serial runner process (vitest run / jest --runInBand), attributing V8 precise-coverage deltas to each file. Far fewer startups — measurably faster wall-clock and less total CPU than one-process-per-file. Use --per-file to opt out.
• Incremental: each test file is hashed; only changed/new files are re-measured. A no-op refresh is instant; editing one test re-maps just that file.
• Parallel: the per-file path (and the shards) run up to one lane per CPU (-j to tune).

Use --full to force a clean rebuild. If single-pass can't account for a file (e.g. a project config it couldn't merge), that file falls back to isolated per-file measurement automatically — the map is never silently incomplete.

testpick detects workspaces (npm/yarn/pnpm workspaces, or pnpm-workspace.yaml) and treats each package as its own unit — its own runner, its own map. A package using Vitest and another using Jest in the same repo both work; testpick map / run iterate them automatically.

$ testpick run
Monorepo: 2 package(s).
[packages/api]  3/41 test file(s) affected by 2 change(s).
[packages/web]  1/120 test file(s) affected by 1 change(s).

Selection is per package, so changing one package doesn't run another's tests. A change outside every package (root config, lockfile, shared tooling) is treated as potentially affecting everything, so testpick safely runs all packages.

- run: npm ci
- run: npx testpick run --base origin/${{ github.base_ref }}

Commit .testpick/map.json to share the map across CI runs, or rebuild it on a schedule.

v0.1 — supports Vitest and Jest, single-file repos and monorepos (per-package, mixed runners). Roadmap: dependency-aware cross-package selection, more runners/languages.

MIT licensed.