@ls-stack/agent-eval 0.5.0 → 0.6.0
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-C5CJ1sX6.mjs → app-7qDBq_ub.mjs} +3 -3
- package/dist/apps/web/dist/assets/index-CdxG9-O-.css +1 -0
- package/dist/apps/web/dist/assets/index-J1yKYGfN.js +112 -0
- package/dist/apps/web/dist/index.html +2 -2
- package/dist/bin.mjs +1 -1
- package/dist/{cli-C5FL7C4G.mjs → cli-C-n-Fd4o.mjs} +746 -269
- package/dist/index.d.mts +1070 -826
- package/dist/index.mjs +3 -3
- package/dist/{runner-K2bN8KRS.mjs → runner-CwEtnUFf.mjs} +2 -2
- package/dist/{runner-Cdlvk56X.mjs → runner-uzzY8kk1.mjs} +1 -1
- package/dist/src-Dy31CPXH.mjs +2 -0
- package/package.json +3 -3
- package/dist/apps/web/dist/assets/index-CBvHVkE7.js +0 -109
- package/dist/apps/web/dist/assets/index-Dd7I28ts.css +0 -1
- package/dist/src-gqm1z1Nu.mjs +0 -2
package/dist/index.mjs
CHANGED
|
@@ -1,3 +1,3 @@
|
|
|
1
|
-
import { $ as
|
|
2
|
-
import "./src-
|
|
3
|
-
export { EvalAssertionError, agentEvalsConfigSchema, assertionFailureSchema, buildTraceTree, cacheEntrySchema, cacheFileSchema, cacheListItemSchema, cacheModeSchema, cacheRecordingOpSchema, cacheRecordingSchema, 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, fileRefSchema, getCurrentScope, getEvalDisplayStatus, getEvalRegistry, getEvalTitle, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, jsonCellSchema, numberDisplayOptionsSchema, repoFile, repoFileRefSchema, runArtifactRefSchema, runCli, runInEvalScope, runManifestSchema, runSummarySchema, scoreTraceSchema, serializedCacheSpanSchema, setEvalOutput, setScopeCacheContext, spanCacheOptionsSchema, sseEnvelopeSchema, traceAttributeDisplayFormatSchema, traceAttributeDisplayInputSchema, traceAttributeDisplayPlacementSchema, traceAttributeDisplaySchema, traceDisplayConfigSchema, traceDisplayInputConfigSchema, traceSpanErrorSchema, traceSpanKindSchema, traceSpanSchema, trialSelectionModeSchema, updateManualScoreRequestSchema };
|
|
1
|
+
import { $ as columnDefSchema, A as evalStatsConfigSchema, B as evalChartTypeSchema, C as spanCacheOptionsSchema, Ct as setEvalOutput, D as evalFreshnessStatusSchema, Dt as getEvalRegistry, E as caseRowSchema, Et as defineEval, F as evalChartBuiltinMetricSchema, G as traceAttributeDisplaySchema, H as traceAttributeDisplayFormatSchema, I as evalChartColorSchema, J as traceSpanErrorSchema, K as traceDisplayConfigSchema, L as evalChartConfigSchema, M as scoreTraceSchema, N as evalChartAggregateSchema, O as evalStatAggregateSchema, P as evalChartAxisSchema, Q as cellValueSchema, R as evalChartMetricSchema, S as serializedCacheSpanSchema, St as runInEvalScope, T as caseDetailSchema, Tt as repoFile, U as traceAttributeDisplayInputSchema, V as evalChartsConfigSchema, W as traceAttributeDisplayPlacementSchema, X as traceSpanSchema, Y as traceSpanKindSchema, Z as traceSpanWarningSchema, _ as cacheListItemSchema, _t as getCurrentScope, a as sseEnvelopeSchema, at as repoFileRefSchema, b as cacheRecordingOpSchema, bt as isInEvalScope, c as deriveScopedSummaryFromCases, ct as buildTraceTree, d as runManifestSchema, dt as evalTracer, et as columnFormatSchema, f as runSummarySchema, ft as hashCacheKey, g as cacheFileSchema, gt as evalAssert, h as cacheEntrySchema, ht as appendToEvalOutput, i as updateManualScoreRequestSchema, it as numberDisplayOptionsSchema, j as evalSummarySchema, k as evalStatItemSchema, l as deriveStatusFromCaseRows, lt as captureEvalSpanError, m as trialSelectionModeSchema, mt as EvalAssertionError, n as createRunner, nt as fileRefSchema, o as getEvalTitle, ot as runArtifactRefSchema, p as agentEvalsConfigSchema, pt as hashCacheKeySync, q as traceDisplayInputConfigSchema, r as createRunRequestSchema, rt as jsonCellSchema, s as getEvalDisplayStatus, st as z, t as runCli, tt as columnKindSchema, u as deriveStatusFromChildStatuses, ut as evalSpan, v as cacheModeSchema, vt as getEvalCaseInput, w as assertionFailureSchema, wt as setScopeCacheContext, x as cacheRecordingSchema, xt as mergeEvalOutput, y as cacheOperationTypeSchema, yt as incrementEvalOutput, z as evalChartTooltipExtraSchema } from "./cli-C-n-Fd4o.mjs";
|
|
2
|
+
import "./src-Dy31CPXH.mjs";
|
|
3
|
+
export { EvalAssertionError, agentEvalsConfigSchema, appendToEvalOutput, assertionFailureSchema, buildTraceTree, cacheEntrySchema, cacheFileSchema, cacheListItemSchema, cacheModeSchema, cacheOperationTypeSchema, cacheRecordingOpSchema, cacheRecordingSchema, 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, fileRefSchema, getCurrentScope, getEvalCaseInput, getEvalDisplayStatus, getEvalRegistry, getEvalTitle, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, jsonCellSchema, mergeEvalOutput, numberDisplayOptionsSchema, repoFile, repoFileRefSchema, runArtifactRefSchema, runCli, runInEvalScope, runManifestSchema, runSummarySchema, scoreTraceSchema, serializedCacheSpanSchema, setEvalOutput, setScopeCacheContext, spanCacheOptionsSchema, sseEnvelopeSchema, traceAttributeDisplayFormatSchema, traceAttributeDisplayInputSchema, traceAttributeDisplayPlacementSchema, traceAttributeDisplaySchema, 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-C-n-Fd4o.mjs";
|
|
2
|
+
import "./src-Dy31CPXH.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-CwEtnUFf.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.
|
|
3
|
+
"version": "0.6.0",
|
|
4
4
|
"type": "module",
|
|
5
5
|
"bin": {
|
|
6
6
|
"agent-evals": "./dist/bin.mjs"
|
|
@@ -56,9 +56,9 @@
|
|
|
56
56
|
"devDependencies": {
|
|
57
57
|
"@types/node": "^24.7.2",
|
|
58
58
|
"typescript": "^5.9.2",
|
|
59
|
-
"@agent-evals/runner": "0.0.1",
|
|
60
59
|
"@agent-evals/sdk": "0.0.1",
|
|
61
|
-
"@agent-evals/shared": "0.0.1"
|
|
60
|
+
"@agent-evals/shared": "0.0.1",
|
|
61
|
+
"@agent-evals/runner": "0.0.1"
|
|
62
62
|
},
|
|
63
63
|
"scripts": {
|
|
64
64
|
"build": "pnpm --filter @agent-evals/web build && tsdown",
|