feat: eval package scaffold with CLI, schemas, types, task loader

New package @united-workforce/eval (uwf-eval CLI): - CLI skeleton: run/report/diff/list subcommands (stubs) - 5 OCAS schemas: eval-run, judge-frontmatter, judge-upstream, judge-hallucination, judge-token-stats - TaskManifest type + parser/validator for task.yaml - JudgeOutput/JudgeInput types for judge contract - EvalRunPayload/EvalRunConfig/EvalJudgeRecord storage types - 19 unit tests: task loader validation + schema definitions Refs #69
2026-06-04 23:42:16 +00:00
parent b94234652a
commit 99619d85db
21 changed files with 675 additions and 1 deletions
@@ -0,0 +1,63 @@
+import { describe, expect, test } from "vitest";
+import {
+  EVAL_JUDGE_FRONTMATTER_SCHEMA,
+  EVAL_JUDGE_HALLUCINATION_SCHEMA,
+  EVAL_JUDGE_TOKEN_STATS_SCHEMA,
+  EVAL_JUDGE_UPSTREAM_SCHEMA,
+  EVAL_RUN_SCHEMA,
+} from "../src/storage/index.js";
+
+describe("OCAS schema definitions", () => {
+  test("eval-run schema has correct title and required fields", () => {
+    expect(EVAL_RUN_SCHEMA.title).toBe("@uwf/eval-run");
+    const required = EVAL_RUN_SCHEMA.required as string[];
+    expect(required).toContain("task");
+    expect(required).toContain("config");
+    expect(required).toContain("threadId");
+    expect(required).toContain("judges");
+    expect(required).toContain("overall");
+    expect(required).toContain("timestamp");
+  });
+
+  test("frontmatter judge schema has correct title", () => {
+    expect(EVAL_JUDGE_FRONTMATTER_SCHEMA.title).toBe("@uwf/eval-judge-frontmatter");
+    const required = EVAL_JUDGE_FRONTMATTER_SCHEMA.required as string[];
+    expect(required).toContain("stepsTotal");
+    expect(required).toContain("stepsValid");
+    expect(required).toContain("invalidSteps");
+  });
+
+  test("upstream judge schema has correct title", () => {
+    expect(EVAL_JUDGE_UPSTREAM_SCHEMA.title).toBe("@uwf/eval-judge-upstream");
+    const required = EVAL_JUDGE_UPSTREAM_SCHEMA.required as string[];
+    expect(required).toContain("perStep");
+  });
+
+  test("hallucination judge schema has correct title", () => {
+    expect(EVAL_JUDGE_HALLUCINATION_SCHEMA.title).toBe("@uwf/eval-judge-hallucination");
+    const required = EVAL_JUDGE_HALLUCINATION_SCHEMA.required as string[];
+    expect(required).toContain("perStep");
+  });
+
+  test("token-stats judge schema has correct title", () => {
+    expect(EVAL_JUDGE_TOKEN_STATS_SCHEMA.title).toBe("@uwf/eval-judge-token-stats");
+    const required = EVAL_JUDGE_TOKEN_STATS_SCHEMA.required as string[];
+    expect(required).toContain("totalInput");
+    expect(required).toContain("totalOutput");
+    expect(required).toContain("totalTurns");
+    expect(required).toContain("perStep");
+  });
+
+  test("all schemas have type object at root", () => {
+    const schemas = [
+      EVAL_RUN_SCHEMA,
+      EVAL_JUDGE_FRONTMATTER_SCHEMA,
+      EVAL_JUDGE_UPSTREAM_SCHEMA,
+      EVAL_JUDGE_HALLUCINATION_SCHEMA,
+      EVAL_JUDGE_TOKEN_STATS_SCHEMA,
+    ];
+    for (const s of schemas) {
+      expect(s.type).toBe("object");
+    }
+  });
+});