@open-insight/eval 0.0.0

package/dist/index.d.mts

@@ -0,0 +1,35 @@
+import { $ as fromStream, A as all, B as InitError$1, C as all$2, D as each$1, E as all$1, F as build$2, G as BenchmarkError, H as build, I as init$2, J as Loader, K as InitError, L as withAgentProvider, M as reduce, O as reduce$1, P as Harness, Q as fromIterable, R as withSandboxProvider, S as withTrajReduce, T as reduce$2, U as init, W as withTasks, X as fromArray, Y as Tasks, Z as fromAsyncIterable, _ as withTask, a as build$3, at as withContext, b as withTraj, c as withHarness, ct as withResources, d as index_d_exports$2, dt as Grader, et as withGitRepo, f as Metrics, ft as Map, g as withBenchmark, gt as bash, h as withBenchReduce, ht as index_d_exports, i as Executor, it as init$1, j as each, k as Metric, l as withMetrics, lt as withSnapshot, m as withBenchEach, mt as ResultSchema, n as run, nt as Task, o as init$4, ot as withGrader, p as init$3, pt as Result, q as index_d_exports$1, r as runPromise, rt as build$1, s as withBenchmark$1, st as withPrompt, tt as fromDir, u as withTrailCount, ut as Exec, v as withTaskEach, w as each$2, x as withTrajEach, y as withTaskReduce, z as HarnessError } from "./index-CjLj-Fvc.mjs";
+import { Effect } from "effect";
+import { Prompt } from "effect/unstable/ai";
+export * from "@open-insight/core";
+
+//#region src/benchmark/export.d.ts
+declare namespace export_d_exports {
+  export { BenchmarkError, InitError, build, init, withTasks };
+}
+declare namespace export_d_exports$9 {
+  export { Exec, Grader, index_d_exports as Internal, Map, Result, ResultSchema, bash };
+}
+declare namespace export_d_exports$5 {
+  export { export_d_exports$9 as Grade, Grader, index_d_exports$1 as Internal, Loader, Task, Tasks, build$1 as build, fromArray, fromAsyncIterable, fromDir, fromIterable, fromStream, init$1 as init, withContext, withGitRepo, withGrader, withPrompt, withResources, withSnapshot };
+}
+declare namespace export_d_exports$2 {
+  export { Harness, HarnessError, InitError$1 as InitError, build$2 as build, init$2 as init, withAgentProvider, withSandboxProvider };
+}
+declare namespace export_d_exports$8 {
+  export { Metric, all, each, reduce };
+}
+declare namespace export_d_exports$7 {
+  export { all$1 as all, each$1 as each, reduce$1 as reduce };
+}
+declare namespace export_d_exports$6 {
+  export { all$2 as all, each$2 as each, reduce$2 as reduce };
+}
+declare namespace export_d_exports$4 {
+  export { export_d_exports$6 as Bench, index_d_exports$2 as Internal, Metrics, export_d_exports$7 as Task, export_d_exports$8 as Traj, init$3 as init, withBenchEach, withBenchReduce, withBenchmark, withTask, withTaskEach, withTaskReduce, withTraj, withTrajEach, withTrajReduce };
+}
+declare namespace export_d_exports$1 {
+  export { Executor, build$3 as build, init$4 as init, run, runPromise, withBenchmark$1 as withBenchmark, withHarness, withMetrics, withTrailCount };
+}
+//#endregion
+export { export_d_exports as Benchmark, Effect, export_d_exports$1 as Exec, export_d_exports$2 as Harness, export_d_exports$3 as Matrix, export_d_exports$4 as Metric, Prompt, export_d_exports$5 as Task };

package/dist/index.mjs

@@ -0,0 +1,105 @@
+import { t as __exportAll } from "./rolldown-runtime-D7D4PA-g.mjs";
+import { $ as withSnapshot, B as fromIterable, C as reduce, D as all$2, E as reduce$1, F as HarnessError, G as grade_exports, H as withGitRepo, I as InitError$1, J as init$1, K as bash, L as task_exports, M as init$2, N as withAgentProvider, O as each$2, P as withSandboxProvider, Q as withResources, R as fromArray, S as each, T as each$1, U as fromDir, V as fromStream, W as ResultSchema, X as withGrader, Y as withContext, Z as withPrompt, _ as withTaskReduce, a as init$4, at as InitError, b as withTrajReduce, c as withMetrics, d as init$3, f as withBenchEach, g as withTaskEach, h as withTask, i as build$3, it as BenchmarkError, j as build$2, k as reduce$2, l as withTrailCount, m as withBenchmark, n as run, nt as init, o as withBenchmark$1, p as withBenchReduce, q as build$1, r as runPromise, rt as withTasks, s as withHarness, tt as build, u as metric_exports, v as withTraj, w as all$1, x as all, y as withTrajEach, z as fromAsyncIterable } from "./exec-w0BrRdRA.mjs";
+import { Effect } from "effect";
+import { Prompt } from "effect/unstable/ai";
+export * from "@open-insight/core";
+//#region src/benchmark/export.ts
+var export_exports = /* @__PURE__ */ __exportAll({
+	BenchmarkError: () => BenchmarkError,
+	InitError: () => InitError,
+	build: () => build,
+	init: () => init,
+	withTasks: () => withTasks
+});
+//#endregion
+//#region src/task/grade/export.ts
+var export_exports$9 = /* @__PURE__ */ __exportAll({
+	Internal: () => grade_exports,
+	ResultSchema: () => ResultSchema,
+	bash: () => bash
+});
+//#endregion
+//#region src/task/export.ts
+var export_exports$5 = /* @__PURE__ */ __exportAll({
+	Grade: () => export_exports$9,
+	Internal: () => task_exports,
+	build: () => build$1,
+	fromArray: () => fromArray,
+	fromAsyncIterable: () => fromAsyncIterable,
+	fromDir: () => fromDir,
+	fromIterable: () => fromIterable,
+	fromStream: () => fromStream,
+	init: () => init$1,
+	withContext: () => withContext,
+	withGitRepo: () => withGitRepo,
+	withGrader: () => withGrader,
+	withPrompt: () => withPrompt,
+	withResources: () => withResources,
+	withSnapshot: () => withSnapshot
+});
+//#endregion
+//#region src/harness/export.ts
+var export_exports$2 = /* @__PURE__ */ __exportAll({
+	HarnessError: () => HarnessError,
+	InitError: () => InitError$1,
+	build: () => build$2,
+	init: () => init$2,
+	withAgentProvider: () => withAgentProvider,
+	withSandboxProvider: () => withSandboxProvider
+});
+//#endregion
+//#region src/metric/traj/export.ts
+var export_exports$8 = /* @__PURE__ */ __exportAll({
+	all: () => all,
+	each: () => each,
+	reduce: () => reduce
+});
+//#endregion
+//#region src/metric/task/export.ts
+var export_exports$7 = /* @__PURE__ */ __exportAll({
+	all: () => all$1,
+	each: () => each$1,
+	reduce: () => reduce$1
+});
+//#endregion
+//#region src/metric/bench/export.ts
+var export_exports$6 = /* @__PURE__ */ __exportAll({
+	all: () => all$2,
+	each: () => each$2,
+	reduce: () => reduce$2
+});
+//#endregion
+//#region src/metric/export.ts
+var export_exports$4 = /* @__PURE__ */ __exportAll({
+	Bench: () => export_exports$6,
+	Internal: () => metric_exports,
+	Task: () => export_exports$7,
+	Traj: () => export_exports$8,
+	init: () => init$3,
+	withBenchEach: () => withBenchEach,
+	withBenchReduce: () => withBenchReduce,
+	withBenchmark: () => withBenchmark,
+	withTask: () => withTask,
+	withTaskEach: () => withTaskEach,
+	withTaskReduce: () => withTaskReduce,
+	withTraj: () => withTraj,
+	withTrajEach: () => withTrajEach,
+	withTrajReduce: () => withTrajReduce
+});
+//#endregion
+//#region src/exec/export.ts
+var export_exports$1 = /* @__PURE__ */ __exportAll({
+	build: () => build$3,
+	init: () => init$4,
+	run: () => run,
+	runPromise: () => runPromise,
+	withBenchmark: () => withBenchmark$1,
+	withHarness: () => withHarness,
+	withMetrics: () => withMetrics,
+	withTrailCount: () => withTrailCount
+});
+//#endregion
+//#region src/matrix/export.ts
+var export_exports$3 = /* @__PURE__ */ __exportAll({});
+//#endregion
+export { export_exports as Benchmark, Effect, export_exports$1 as Exec, export_exports$2 as Harness, export_exports$3 as Matrix, export_exports$4 as Metric, Prompt, export_exports$5 as Task };

package/dist/internal.d.mts

@@ -0,0 +1,6 @@
+import { N as index_d_exports$2, V as index_d_exports, d as index_d_exports$4, q as index_d_exports$5, t as index_d_exports$1 } from "./index-CjLj-Fvc.mjs";
+
+//#region src/matrix/index.d.ts
+
+//#endregion
+export { index_d_exports as Benchmark, index_d_exports$1 as Exec, index_d_exports$2 as Harness, index_d_exports$3 as Matrix, index_d_exports$4 as Metric, index_d_exports$5 as Task };

package/dist/internal.mjs

@@ -0,0 +1,6 @@
+import { t as __exportAll } from "./rolldown-runtime-D7D4PA-g.mjs";
+import { A as harness_exports, L as task_exports, et as benchmark_exports, t as exec_exports, u as metric_exports } from "./exec-w0BrRdRA.mjs";
+//#region src/matrix/index.ts
+var matrix_exports = /* @__PURE__ */ __exportAll({});
+//#endregion
+export { benchmark_exports as Benchmark, exec_exports as Exec, harness_exports as Harness, matrix_exports as Matrix, metric_exports as Metric, task_exports as Task };

package/dist/rolldown-runtime-D7D4PA-g.mjs

@@ -0,0 +1,13 @@
+//#region \0rolldown/runtime.js
+var __defProp = Object.defineProperty;
+var __exportAll = (all, no_symbols) => {
+	let target = {};
+	for (var name in all) __defProp(target, name, {
+		get: all[name],
+		enumerable: true
+	});
+	if (!no_symbols) __defProp(target, Symbol.toStringTag, { value: "Module" });
+	return target;
+};
+//#endregion
+export { __exportAll as t };

package/package.json

@@ -0,0 +1,52 @@
+{
+  "name": "@open-insight/eval",
+  "version": "0.0.0",
+  "description": "Evaluator framework for Open Insight — benchmarks, task grading, harnesses.",
+  "homepage": "https://github.com/open-insight/open-insight#readme",
+  "bugs": {
+    "url": "https://github.com/open-insight/open-insight/issues"
+  },
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/open-insight/open-insight.git",
+    "directory": "packages/eval"
+  },
+  "files": [
+    "dist"
+  ],
+  "type": "module",
+  "exports": {
+    ".": "./dist/index.mjs",
+    "./internal": "./dist/internal.mjs",
+    "./package.json": "./package.json"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "dependencies": {
+    "@effect/platform-node": "4.0.0-beta.91",
+    "@effect/vitest": "4.0.0-beta.91",
+    "ai": "^7.0.0",
+    "effect": "4.0.0-beta.90",
+    "immer": "^10.1.1",
+    "picomatch": "^4.0.4",
+    "@open-insight/core": "0.0.0"
+  },
+  "devDependencies": {
+    "@types/node": "^25.6.2",
+    "@types/picomatch": "^4.0.3",
+    "@typescript/native-preview": "7.0.0-dev.20260629.1",
+    "bumpp": "^11.1.0",
+    "typescript": "^5",
+    "vite-plus": "latest",
+    "vitest": "4.1.9",
+    "@open-insight/utils": "0.0.0"
+  },
+  "scripts": {
+    "build": "vp pack",
+    "dev": "vp pack --watch",
+    "test": "vp test",
+    "check": "vp check"
+  }
+}