vitest-evals

作者： sentry

在编写、审查或调试基于 harness 的 vitest-evals 套件、自定义 Harness 适配器、第一方 ai-sdk 或 pi-ai harness 集成时使用，…

npx skills add https://github.com/getsentry/vitest-evals --skill vitest-evals

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

First Steps

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

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`

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.
Keep judge model calls on judges. Use createJudge("Name", assess) for custom judges; use the provider-helper overload only when multiple judges reuse setup and need curried run options.
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.

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`