@ls-stack/agent-eval 0.12.0 → 0.12.1
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-DI3IUGb_.mjs → app-BZsYXjOt.mjs} +3 -3
- package/dist/apps/web/dist/assets/{index-BMe8ZR3n.js → index-DJf7Af49.js} +34 -32
- package/dist/apps/web/dist/index.html +1 -1
- package/dist/bin.mjs +1 -1
- package/dist/{cli-COzPxKg2.mjs → cli-DolSsprq.mjs} +2 -2
- package/dist/index.mjs +2 -2
- package/dist/{runner-nQjuRZGC.mjs → runner-CSydBBpq.mjs} +2 -2
- package/dist/{runner-sMZXoDp3.mjs → runner-CsZEhUEd.mjs} +1 -1
- package/dist/{src-OZSs693X.mjs → src-CzAf_rfb.mjs} +1 -1
- package/package.json +3 -3
|
@@ -25,7 +25,7 @@
|
|
|
25
25
|
href="https://fonts.googleapis.com/css2?family=Geist:wght@400;500;600&family=JetBrains+Mono:wght@400;500&display=swap"
|
|
26
26
|
rel="stylesheet"
|
|
27
27
|
/>
|
|
28
|
-
<script type="module" crossorigin src="/assets/index-
|
|
28
|
+
<script type="module" crossorigin src="/assets/index-DJf7Af49.js"></script>
|
|
29
29
|
<link rel="stylesheet" crossorigin href="/assets/index-BVnLr79e.css">
|
|
30
30
|
</head>
|
|
31
31
|
<body>
|
package/dist/bin.mjs
CHANGED
|
@@ -1050,8 +1050,8 @@ async function commandApp(args) {
|
|
|
1050
1050
|
const { serve } = await import("@hono/node-server");
|
|
1051
1051
|
const bundledWebDist = resolve(currentDir, "apps/web/dist");
|
|
1052
1052
|
if (existsSync(bundledWebDist)) process.env.AGENT_EVALS_WEB_DIST = bundledWebDist;
|
|
1053
|
-
const appModule = await import("./app-
|
|
1054
|
-
const runnerModule = await import("./runner-
|
|
1053
|
+
const appModule = await import("./app-BZsYXjOt.mjs");
|
|
1054
|
+
const runnerModule = await import("./runner-CsZEhUEd.mjs");
|
|
1055
1055
|
if (!isHonoAppModule(appModule)) throw new Error("Server app module is invalid");
|
|
1056
1056
|
if (!isServerRunnerModule(runnerModule)) throw new Error("Server runner module is invalid");
|
|
1057
1057
|
await runnerModule.initRunner();
|
package/dist/index.mjs
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
1
|
import { $ as evalFreshnessStatusSchema, $t as evalAssert, A as getEvalDisplayStatus, At as traceDisplayInputConfigSchema, B as apiCallMetricPlacementSchema, Bt as jsonCellSchema, C as updateManualScoreRequestSchema, Ct as spanCacheOptionsSchema, D as extractLlmCalls, Dt as traceAttributeDisplayPlacementSchema, E as extractApiCalls, Et as traceAttributeDisplayInputSchema, F as runSummarySchema, Ft as cellValueSchema, G as llmCallMetricSchema, Gt as buildTraceTree, H as apiCallsConfigSchema, Ht as repoFileRefSchema, I as DEFAULT_API_CALLS_CONFIG, It as columnDefSchema, J as resolveLlmCallsConfig, Jt as evalTracer, K as llmCallsConfigSchema, Kt as captureEvalSpanError, L as DEFAULT_LLM_CALLS_CONFIG, Lt as columnFormatSchema, M as deriveStatusFromCaseRows, Mt as traceSpanKindSchema, N as deriveStatusFromChildStatuses, Nt as traceSpanSchema, O as getNestedAttribute, Ot as traceAttributeDisplaySchema, P as runManifestSchema, Pt as traceSpanWarningSchema, Q as caseRowSchema, Qt as appendToEvalOutput, R as agentEvalsConfigSchema, Rt as columnKindSchema, S as createRunRequestSchema, St as serializedCacheSpanSchema, T as extractCacheHits, Tt as traceAttributeDisplayFormatSchema, U as llmCallMetricFormatSchema, Ut as runArtifactRefSchema, V as apiCallMetricSchema, Vt as numberDisplayOptionsSchema, W as llmCallMetricPlacementSchema, Wt as z, X as assertionFailureSchema, Xt as hashCacheKeySync, Y as trialSelectionModeSchema, Yt as hashCacheKey, Z as caseDetailSchema, Zt as EvalAssertionError, _t as cacheModeSchema, an as nextEvalId, at as evalChartAggregateSchema, bt as cacheRecordingSchema, cn as setScopeCacheContext, ct as evalChartColorSchema, dn as getEvalRegistry, dt as evalChartTooltipExtraSchema, en as getCurrentScope, et as evalStatAggregateSchema, ft as evalChartTypeSchema, gt as cacheListItemSchema, ht as cacheFileSchema, in as mergeEvalOutput, it as scoreTraceSchema, j as deriveScopedSummaryFromCases, jt as traceSpanErrorSchema, k as getEvalTitle, kt as traceDisplayConfigSchema, ln as repoFile, lt as evalChartConfigSchema, mt as cacheEntrySchema, nn as incrementEvalOutput, nt as evalStatsConfigSchema, on as runInEvalScope, ot as evalChartAxisSchema, pt as evalChartsConfigSchema, q as resolveApiCallsConfig, qt as evalSpan, rn as isInEvalScope, rt as evalSummarySchema, sn as setEvalOutput, st as evalChartBuiltinMetricSchema, tn as getEvalCaseInput, tt as evalStatItemSchema, un as defineEval, ut as evalChartMetricSchema, vt as cacheOperationTypeSchema, w as sseEnvelopeSchema, wt as traceCacheRefSchema, xt as cacheStatusSchema, yt as cacheRecordingOpSchema, z as apiCallMetricFormatSchema, zt as fileRefSchema } from "./runOrchestration-COFhQvTJ.mjs";
|
|
2
|
-
import { n as createRunner, t as runCli } from "./cli-
|
|
3
|
-
import "./src-
|
|
2
|
+
import { n as createRunner, t as runCli } from "./cli-DolSsprq.mjs";
|
|
3
|
+
import "./src-CzAf_rfb.mjs";
|
|
4
4
|
export { DEFAULT_API_CALLS_CONFIG, DEFAULT_LLM_CALLS_CONFIG, EvalAssertionError, agentEvalsConfigSchema, apiCallMetricFormatSchema, apiCallMetricPlacementSchema, apiCallMetricSchema, apiCallsConfigSchema, appendToEvalOutput, assertionFailureSchema, buildTraceTree, cacheEntrySchema, cacheFileSchema, cacheListItemSchema, cacheModeSchema, cacheOperationTypeSchema, cacheRecordingOpSchema, cacheRecordingSchema, cacheStatusSchema, captureEvalSpanError, 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, extractApiCalls, extractCacheHits, extractLlmCalls, fileRefSchema, getCurrentScope, getEvalCaseInput, getEvalDisplayStatus, getEvalRegistry, getEvalTitle, getNestedAttribute, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, jsonCellSchema, llmCallMetricFormatSchema, llmCallMetricPlacementSchema, llmCallMetricSchema, llmCallsConfigSchema, mergeEvalOutput, nextEvalId, numberDisplayOptionsSchema, repoFile, repoFileRefSchema, resolveApiCallsConfig, resolveLlmCallsConfig, runArtifactRefSchema, runCli, runInEvalScope, runManifestSchema, runSummarySchema, scoreTraceSchema, serializedCacheSpanSchema, setEvalOutput, setScopeCacheContext, spanCacheOptionsSchema, sseEnvelopeSchema, traceAttributeDisplayFormatSchema, traceAttributeDisplayInputSchema, traceAttributeDisplayPlacementSchema, traceAttributeDisplaySchema, traceCacheRefSchema, traceDisplayConfigSchema, traceDisplayInputConfigSchema, traceSpanErrorSchema, traceSpanKindSchema, traceSpanSchema, traceSpanWarningSchema, trialSelectionModeSchema, updateManualScoreRequestSchema, z };
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { n as createRunner } from "./cli-
|
|
2
|
-
import "./src-
|
|
1
|
+
import { n as createRunner } from "./cli-DolSsprq.mjs";
|
|
2
|
+
import "./src-CzAf_rfb.mjs";
|
|
3
3
|
//#region ../../apps/server/src/runner.ts
|
|
4
4
|
let runnerInstance = null;
|
|
5
5
|
function getRunnerInstance() {
|
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
import { n as initRunner, t as getRunnerInstance } from "./runner-
|
|
1
|
+
import { n as initRunner, t as getRunnerInstance } from "./runner-CSydBBpq.mjs";
|
|
2
2
|
export { getRunnerInstance, initRunner };
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@ls-stack/agent-eval",
|
|
3
|
-
"version": "0.12.
|
|
3
|
+
"version": "0.12.1",
|
|
4
4
|
"type": "module",
|
|
5
5
|
"bin": {
|
|
6
6
|
"agent-evals": "./dist/bin.mjs"
|
|
@@ -57,9 +57,9 @@
|
|
|
57
57
|
"devDependencies": {
|
|
58
58
|
"@types/node": "^24.7.2",
|
|
59
59
|
"typescript": "^5.9.2",
|
|
60
|
+
"@agent-evals/shared": "0.0.1",
|
|
60
61
|
"@agent-evals/runner": "0.0.1",
|
|
61
|
-
"@agent-evals/sdk": "0.0.1"
|
|
62
|
-
"@agent-evals/shared": "0.0.1"
|
|
62
|
+
"@agent-evals/sdk": "0.0.1"
|
|
63
63
|
},
|
|
64
64
|
"scripts": {
|
|
65
65
|
"build": "pnpm --filter @agent-evals/web build && tsdown",
|