@ls-stack/agent-eval 0.2.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/app-CKa9TjXw.mjs +244 -0
- package/dist/apps/web/dist/assets/index-BUz24J7O.css +1 -0
- package/dist/apps/web/dist/assets/index-Dm50Ynbs.js +109 -0
- package/dist/apps/web/dist/favicon.svg +20 -0
- package/dist/apps/web/dist/index.html +34 -0
- package/dist/bin.d.mts +1 -0
- package/dist/bin.mjs +41 -0
- package/dist/cli-CwEFLP0w.mjs +3422 -0
- package/dist/index.d.mts +2043 -0
- package/dist/index.mjs +3 -0
- package/dist/runner-CD5aDJ0C.mjs +15 -0
- package/dist/runner-Ck4X0H3p.mjs +2 -0
- package/dist/src-BDRmaWFu.mjs +2 -0
- package/package.json +71 -0
package/dist/index.mjs
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
1
|
+
import { $ as jsonCellSchema, A as scoreTraceSchema, B as traceAttributeDisplayFormatSchema, C as caseDetailSchema, D as evalStatItemSchema, E as evalStatAggregateSchema, F as evalChartConfigSchema, G as traceDisplayInputConfigSchema, H as traceAttributeDisplayPlacementSchema, I as evalChartMetricSchema, J as cellValueSchema, K as traceSpanKindSchema, L as evalChartTooltipExtraSchema, M as evalChartAxisSchema, N as evalChartBuiltinMetricSchema, O as evalStatsConfigSchema, P as evalChartColorSchema, Q as fileRefSchema, R as evalChartTypeSchema, S as assertionFailureSchema, T as evalFreshnessStatusSchema, U as traceAttributeDisplaySchema, V as traceAttributeDisplayInputSchema, W as traceDisplayConfigSchema, X as columnFormatSchema, Y as columnDefSchema, Z as columnKindSchema, _ as cacheModeSchema, _t as getEvalRegistry, a as sseEnvelopeSchema, at as evalTracer, b as serializedCacheSpanSchema, c as deriveScopedSummaryFromCases, ct as evalAssert, d as runManifestSchema, dt as isInEvalScope, et as numberDisplayOptionsSchema, f as runSummarySchema, ft as runInEvalScope, g as cacheListItemSchema, gt as defineEval, h as cacheEntrySchema, ht as repoFile, i as updateManualScoreRequestSchema, it as evalSpan, j as evalChartAggregateSchema, k as evalSummarySchema, l as deriveStatusFromCaseRows, lt as getCurrentScope, m as trialSelectionModeSchema, mt as setScopeCacheContext, n as createRunner, nt as runArtifactRefSchema, o as getEvalTitle, ot as hashCacheKey, p as agentEvalsConfigSchema, pt as setEvalOutput, q as traceSpanSchema, r as createRunRequestSchema, rt as buildTraceTree, s as getEvalDisplayStatus, st as EvalAssertionError, t as runCli, tt as repoFileRefSchema, u as deriveStatusFromChildStatuses, ut as incrementEvalOutput, v as cacheRecordingOpSchema, w as caseRowSchema, x as spanCacheOptionsSchema, y as cacheRecordingSchema, z as evalChartsConfigSchema } from "./cli-CwEFLP0w.mjs";
|
|
2
|
+
import "./src-BDRmaWFu.mjs";
|
|
3
|
+
export { EvalAssertionError, agentEvalsConfigSchema, assertionFailureSchema, buildTraceTree, cacheEntrySchema, cacheListItemSchema, cacheModeSchema, cacheRecordingOpSchema, cacheRecordingSchema, caseDetailSchema, caseRowSchema, cellValueSchema, columnDefSchema, columnFormatSchema, columnKindSchema, createRunRequestSchema, createRunner, defineEval, deriveScopedSummaryFromCases, deriveStatusFromCaseRows, deriveStatusFromChildStatuses, evalAssert, evalChartAggregateSchema, evalChartAxisSchema, evalChartBuiltinMetricSchema, evalChartColorSchema, evalChartConfigSchema, evalChartMetricSchema, evalChartTooltipExtraSchema, evalChartTypeSchema, evalChartsConfigSchema, evalFreshnessStatusSchema, evalSpan, evalStatAggregateSchema, evalStatItemSchema, evalStatsConfigSchema, evalSummarySchema, evalTracer, fileRefSchema, getCurrentScope, getEvalDisplayStatus, getEvalRegistry, getEvalTitle, hashCacheKey, incrementEvalOutput, isInEvalScope, jsonCellSchema, numberDisplayOptionsSchema, repoFile, repoFileRefSchema, runArtifactRefSchema, runCli, runInEvalScope, runManifestSchema, runSummarySchema, scoreTraceSchema, serializedCacheSpanSchema, setEvalOutput, setScopeCacheContext, spanCacheOptionsSchema, sseEnvelopeSchema, traceAttributeDisplayFormatSchema, traceAttributeDisplayInputSchema, traceAttributeDisplayPlacementSchema, traceAttributeDisplaySchema, traceDisplayConfigSchema, traceDisplayInputConfigSchema, traceSpanKindSchema, traceSpanSchema, trialSelectionModeSchema, updateManualScoreRequestSchema };
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
import { n as createRunner } from "./cli-CwEFLP0w.mjs";
|
|
2
|
+
import "./src-BDRmaWFu.mjs";
|
|
3
|
+
//#region ../../apps/server/src/runner.ts
|
|
4
|
+
let runnerInstance = null;
|
|
5
|
+
function getRunnerInstance() {
|
|
6
|
+
if (!runnerInstance) runnerInstance = createRunner();
|
|
7
|
+
return runnerInstance;
|
|
8
|
+
}
|
|
9
|
+
async function initRunner() {
|
|
10
|
+
const runner = getRunnerInstance();
|
|
11
|
+
await runner.init();
|
|
12
|
+
return runner;
|
|
13
|
+
}
|
|
14
|
+
//#endregion
|
|
15
|
+
export { initRunner as n, getRunnerInstance as t };
|
package/package.json
ADDED
|
@@ -0,0 +1,71 @@
|
|
|
1
|
+
{
|
|
2
|
+
"name": "@ls-stack/agent-eval",
|
|
3
|
+
"version": "0.2.0",
|
|
4
|
+
"type": "module",
|
|
5
|
+
"bin": {
|
|
6
|
+
"agent-evals": "./dist/bin.mjs"
|
|
7
|
+
},
|
|
8
|
+
"exports": {
|
|
9
|
+
".": {
|
|
10
|
+
"types": "./dist/index.d.mts",
|
|
11
|
+
"import": "./dist/index.mjs",
|
|
12
|
+
"default": "./dist/index.mjs"
|
|
13
|
+
}
|
|
14
|
+
},
|
|
15
|
+
"files": [
|
|
16
|
+
"dist"
|
|
17
|
+
],
|
|
18
|
+
"tsdown": {
|
|
19
|
+
"clean": true,
|
|
20
|
+
"copy": [
|
|
21
|
+
{
|
|
22
|
+
"from": "../../apps/web/dist/**/*",
|
|
23
|
+
"to": "dist/web",
|
|
24
|
+
"flatten": false
|
|
25
|
+
}
|
|
26
|
+
],
|
|
27
|
+
"deps": {
|
|
28
|
+
"alwaysBundle": [
|
|
29
|
+
"@agent-evals/runner",
|
|
30
|
+
"@agent-evals/sdk",
|
|
31
|
+
"@agent-evals/shared"
|
|
32
|
+
]
|
|
33
|
+
},
|
|
34
|
+
"dts": {
|
|
35
|
+
"eager": true
|
|
36
|
+
},
|
|
37
|
+
"entry": [
|
|
38
|
+
"src/index.ts",
|
|
39
|
+
"src/bin.ts"
|
|
40
|
+
],
|
|
41
|
+
"format": "esm",
|
|
42
|
+
"platform": "node"
|
|
43
|
+
},
|
|
44
|
+
"dependencies": {
|
|
45
|
+
"@hono/zod-validator": "^0.7.3",
|
|
46
|
+
"@hono/node-server": "^1.14.3",
|
|
47
|
+
"@ls-stack/utils": "^3.58.0",
|
|
48
|
+
"chokidar": "^4.0.3",
|
|
49
|
+
"glob": "^11.0.2",
|
|
50
|
+
"hono": "^4.9.9",
|
|
51
|
+
"launch-editor": "^2.13.2",
|
|
52
|
+
"t-result": "^0.6.1",
|
|
53
|
+
"vitest": "^3.2.4",
|
|
54
|
+
"zod": "^4.1.11"
|
|
55
|
+
},
|
|
56
|
+
"devDependencies": {
|
|
57
|
+
"@types/node": "^24.7.2",
|
|
58
|
+
"typescript": "^5.9.2",
|
|
59
|
+
"@agent-evals/runner": "0.0.1",
|
|
60
|
+
"@agent-evals/shared": "0.0.1",
|
|
61
|
+
"@agent-evals/sdk": "0.0.1"
|
|
62
|
+
},
|
|
63
|
+
"scripts": {
|
|
64
|
+
"build": "pnpm --filter @agent-evals/web build && tsdown",
|
|
65
|
+
"test": "vitest run",
|
|
66
|
+
"tsc": "tsgo --noEmit",
|
|
67
|
+
"lint": "tsgo --noEmit && eslint .",
|
|
68
|
+
"eslint": "eslint ."
|
|
69
|
+
},
|
|
70
|
+
"types": "./dist/index.d.mts"
|
|
71
|
+
}
|