@yansirplus/evals
v0.6.4
Published
<!-- generated by packages/cli/src/generate/generate-docs.mjs; edit docs/surface.json and docs/packages/evals.md -->
Readme
@yansirplus/evals
Purpose
Authoring DSL for generated app behavior evals. The package lets product
repos declare evals/**/*.eval.ts cases, symbolic provider needs, public
driver facades, and deterministic assertions over generated app behavior.
Public API Status
0.6.x active development. The package is an authoring surface for eval definitions; execution, app serving, provider resolution, and artifact persistence remain owned by the CLI/runtime/client seams.
Audience
| Entrypoint | Audience | Capability |
| ----------------- | ---------------- | -------------------------------- |
| @yansirplus/evals | default-direct | generated app eval authoring DSL |
Invariant
@yansirplus/evals is declaration-only. It owns eval definitions, datasets,
symbolic config, assertion declarations, and public facade types. It does not
assemble runtime graphs, construct model providers, serve generated apps,
import runtime internals, or read eval result artifacts as source facts.
Minimal Usage
import { defineEval, evalAssertion } from "@yansirplus/evals";
export default defineEval({
path: import.meta.url,
cases: [{ id: "turn", input: { sessionRef: "session-1", turnRef: "turn-1" } }],
assertions: [evalAssertion.completed()],
run: async ({ case: testCase, sessions }) => {
await sessions.submitTurn({
intent: "answer",
context: {},
sessionRef: testCase.input.sessionRef,
turnRef: testCase.input.turnRef,
});
},
});Verification
pnpm --filter @yansirplus/evals test
pnpm run agentos eval --cwd /path/to/generated-app --target local