vitest-evals

vitest-evals

Use the harness-backed API as the only authoring model.

First Steps

1. Read the package, app, or eval file being changed.
2. Identify the runtime target, then open only the needed reference.
3. Keep suites close to Vitest: one harness per describeEval(...), explicit run(...) inside each test, ordinary expect(...) assertions over the returned result.

Reference Router

Need	Open
Write or review a normal eval suite	references/suite-authoring.md
Build a custom app Harness without a first-party adapter	references/custom-harness.md
Integrate AI SDK generateText, generateObject, tools, or an AI SDK-style agent	references/harness-ai-sdk.md
Integrate a Pi AI or Pi Mono-style agent	references/harness-pi-ai.md
Add custom judges, suite judges, built-in judges, or toSatisfyJudge(...) assertions	references/judges-and-assertions.md
Assert on message, tool-call, or span history	references/utilities.md
Configure tool recording or replay	references/tool-replay.md
Diagnose failures, missing traces, odd output, or choose verification commands	references/troubleshooting.md

Runtime Defaults

• Import describeEval(...), judges, and helpers from vitest-evals.
• Bind exactly one harness to a suite.
• Call run(input) where the test should execute the system.
• Assert on result.output for app-facing behavior.
• Use toolCalls(result) and message helpers for normalized session assertions.
• Use spans(result), spansByKind(result, kind), and failedSpans(result) for span assertions.
• Keep HarnessRun, NormalizedSession, usage, artifacts, and tool records JSON-serializable.
• Report known tokens and USD cost in run.usage. Omit costUsd when unknown; use 0 only when the run is known to be free.
• Keep rubric text and parsing in the judge. Put provider calls in a judgeHarness, then call them through ctx.runJudge(...) so usage is recorded.
• Put scenario-owned criteria on the input value. Put direct-check expected values in Vitest case rows. Pass per-case judge criteria through explicit matcher options, and suite-wide criteria through judge config.
• Custom judges should use createJudge(...) for stable reporter labels.

Verification

Prefer the smallest command that covers the edited files:

Task	Command
Lint file	pnpm exec biome lint path/to/file.ts
Format file	pnpm exec biome format --write path/to/file.ts
Test file	pnpm exec vitest run path/to/file.test.ts -c vitest.config.ts
Eval file	pnpm exec vitest run path/to/file.eval.ts -c vitest.config.ts --reporter=./packages/vitest-evals/src/reporter.ts
Type surface	pnpm typecheck
Package build	pnpm build