@ls-stack/agent-eval 0.25.0 → 0.26.0
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-DS3j_AyX.mjs → app-BWjrZUSt.mjs} +2 -2
- package/dist/apps/web/dist/assets/{index-DqR1YaMG.js → index-3bLi3uQJ.js} +1 -1
- package/dist/apps/web/dist/index.html +1 -1
- package/dist/bin.mjs +1 -1
- package/dist/{cli-ETfZ15RB.mjs → cli-CmNNm6d0.mjs} +6 -5
- package/dist/index.mjs +2 -2
- package/dist/{runner-B2f2TEjp.mjs → runner-NPxmmHwa.mjs} +1 -1
- package/dist/{runner-cj1TkR-H.mjs → runner-dnkl4pWd.mjs} +2 -2
- package/dist/{src-CyNb2ycA.mjs → src-DIFRuMuH.mjs} +1 -1
- package/package.json +1 -1
|
@@ -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-3bLi3uQJ.js"></script>
|
|
29
29
|
<link rel="stylesheet" crossorigin href="/assets/index-DNsZjOms.css">
|
|
30
30
|
</head>
|
|
31
31
|
<body>
|
package/dist/bin.mjs
CHANGED
|
@@ -742,6 +742,9 @@ function createRunner({ watchForChanges = true } = {}) {
|
|
|
742
742
|
ignoreInitial: true,
|
|
743
743
|
persistent: true
|
|
744
744
|
});
|
|
745
|
+
const watcherReady = new Promise((ready) => {
|
|
746
|
+
watcher.once("ready", ready);
|
|
747
|
+
});
|
|
745
748
|
discoveryWatcher = watcher;
|
|
746
749
|
const scheduleRefresh = () => {
|
|
747
750
|
if (discoveryRefreshTimer !== void 0) clearTimeout(discoveryRefreshTimer);
|
|
@@ -756,9 +759,7 @@ function createRunner({ watchForChanges = true } = {}) {
|
|
|
756
759
|
watcher.on("addDir", scheduleRefresh);
|
|
757
760
|
watcher.on("unlinkDir", scheduleRefresh);
|
|
758
761
|
await setupRunHistoryWatcher();
|
|
759
|
-
await
|
|
760
|
-
watcher.once("ready", ready);
|
|
761
|
-
});
|
|
762
|
+
await watcherReady;
|
|
762
763
|
}
|
|
763
764
|
async function setupRunHistoryWatcher() {
|
|
764
765
|
const watcher = watch(join(localStateDir, "runs"), {
|
|
@@ -1040,8 +1041,8 @@ async function commandApp(args) {
|
|
|
1040
1041
|
const { serve } = await import("@hono/node-server");
|
|
1041
1042
|
const bundledWebDist = resolve(currentDir, "apps/web/dist");
|
|
1042
1043
|
if (existsSync(bundledWebDist)) process.env.AGENT_EVALS_WEB_DIST = bundledWebDist;
|
|
1043
|
-
const appModule = await import("./app-
|
|
1044
|
-
const runnerModule = await import("./runner-
|
|
1044
|
+
const appModule = await import("./app-BWjrZUSt.mjs");
|
|
1045
|
+
const runnerModule = await import("./runner-NPxmmHwa.mjs");
|
|
1045
1046
|
if (!isHonoAppModule(appModule)) throw new Error("Server app module is invalid");
|
|
1046
1047
|
if (!isServerRunnerModule(runnerModule)) throw new Error("Server runner module is invalid");
|
|
1047
1048
|
await runnerModule.initRunner();
|
package/dist/index.mjs
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
1
|
import { $ as llmCallsConfigSchema, $t as traceSpanKindSchema, A as extractApiCalls, An as getEvalStartTime, At as evalChartTypeSchema, B as runSummarySchema, Bn as startEvalBackgroundJob, Bt as cacheRecordingOpSchema, Cn as advanceEvalTime, Ct as evalChartAggregateSchema, D as sseEnvelopeSchema, Dn as evalLog, Dt as evalChartConfigSchema, E as updateManualScoreRequestSchema, En as evalAssert, Et as evalChartColorSchema, F as getEvalDisplayStatus, Fn as runInEvalRuntimeScope, Ft as cacheEntryWithDebugKeySchema, G as apiCallMetricPlacementSchema, Gt as traceCacheRefSchema, H as DEFAULT_LLM_CALLS_CONFIG, Hn as defineEval, Ht as cacheStatusSchema, I as deriveScopedSummaryFromCases, In as runInEvalScope, It as cacheFileSchema, J as defaultConfigKeySchema, Jt as traceAttributeDisplayPlacementSchema, K as apiCallMetricSchema, Kt as traceAttributeDisplayFormatSchema, L as deriveStatusFromCaseRows, Ln as runInExistingEvalScope, Lt as cacheListItemSchema, M as applyDerivedCallAttributes, Mn as isInEvalScope, Mt as cacheDebugKeyEntrySchema, N as getNestedAttribute, Nn as mergeEvalOutput, Nt as cacheDebugKeyFileSchema, O as extractCacheEntries, On as getCurrentScope, Ot as evalChartMetricSchema, P as getEvalTitle, Pn as nextEvalId, Pt as cacheEntrySchema, Q as llmCallPricingSchema, Qt as traceSpanErrorSchema, R as deriveStatusFromChildStatuses, Rn as setEvalOutput, Rt as cacheModeSchema, Sn as EvalAssertionError, St as scoreTraceSchema, T as createRunRequestSchema, Tt as evalChartBuiltinMetricSchema, U as agentEvalsConfigSchema, Un as getEvalRegistry, Ut as serializedCacheSpanSchema, V as DEFAULT_API_CALLS_CONFIG, Vn as repoFile, Vt as cacheRecordingSchema, W as apiCallMetricFormatSchema, Wt as spanCacheOptionsSchema, X as llmCallMetricPlacementSchema, Xt as traceDisplayConfigSchema, Y as llmCallMetricFormatSchema, Yt as traceAttributeDisplaySchema, Z as llmCallMetricSchema, Zt as traceDisplayInputConfigSchema, _n as hashCacheKeySync, _t as evalSummarySchema, an as columnKindSchema, at as buildCaseKey, bn as serializeCacheRecording, bt as runLogLocationSchema, cn as numberDisplayOptionsSchema, ct as getCaseRowEvalKey, dn as z, dt as caseRowSchema, en as traceSpanSchema, et as removeDefaultConfigSchema, fn as buildTraceTree, ft as discoveryIssueSchema, gn as hashCacheKey, gt as evalStatsConfigSchema, hn as evalTracer, ht as evalStatItemSchema, in as columnFormatSchema, it as trialSelectionModeSchema, j as extractLlmCalls, jn as incrementEvalOutput, jt as evalChartsConfigSchema, k as extractCacheHits, kn as getEvalCaseInput, kt as evalChartTooltipExtraSchema, ln as repoFileRefSchema, lt as assertionFailureSchema, mn as evalSpan, mt as evalStatAggregateSchema, nn as cellValueSchema, nt as resolveLlmCallsConfig, on as fileRefSchema, ot as buildEvalKey, pn as captureEvalSpanError, pt as evalFreshnessStatusSchema, q as apiCallsConfigSchema, qt as traceAttributeDisplayInputSchema, rn as columnDefSchema, rt as runLogsConfigSchema, sn as jsonCellSchema, st as getCaseRowCaseKey, tn as traceSpanWarningSchema, tt as resolveApiCallsConfig, un as runArtifactRefSchema, ut as caseDetailSchema, vn as deserializeCacheRecording, vt as runLogEntrySchema, wn as appendToEvalOutput, wt as evalChartAxisSchema, xn as serializeCacheValue, xt as runLogPhaseSchema, yn as deserializeCacheValue, yt as runLogLevelSchema, z as runManifestSchema, zn as setScopeCacheContext, zt as cacheOperationTypeSchema } from "./runOrchestration-B31SV_Bq.mjs";
|
|
2
|
-
import { n as createRunner, t as runCli } from "./cli-
|
|
3
|
-
import "./src-
|
|
2
|
+
import { n as createRunner, t as runCli } from "./cli-CmNNm6d0.mjs";
|
|
3
|
+
import "./src-DIFRuMuH.mjs";
|
|
4
4
|
export { DEFAULT_API_CALLS_CONFIG, DEFAULT_LLM_CALLS_CONFIG, EvalAssertionError, advanceEvalTime, agentEvalsConfigSchema, apiCallMetricFormatSchema, apiCallMetricPlacementSchema, apiCallMetricSchema, apiCallsConfigSchema, appendToEvalOutput, applyDerivedCallAttributes, assertionFailureSchema, buildCaseKey, buildEvalKey, buildTraceTree, cacheDebugKeyEntrySchema, cacheDebugKeyFileSchema, cacheEntrySchema, cacheEntryWithDebugKeySchema, cacheFileSchema, cacheListItemSchema, cacheModeSchema, cacheOperationTypeSchema, cacheRecordingOpSchema, cacheRecordingSchema, cacheStatusSchema, captureEvalSpanError, caseDetailSchema, caseRowSchema, cellValueSchema, columnDefSchema, columnFormatSchema, columnKindSchema, createRunRequestSchema, createRunner, defaultConfigKeySchema, defineEval, deriveScopedSummaryFromCases, deriveStatusFromCaseRows, deriveStatusFromChildStatuses, deserializeCacheRecording, deserializeCacheValue, discoveryIssueSchema, evalAssert, evalChartAggregateSchema, evalChartAxisSchema, evalChartBuiltinMetricSchema, evalChartColorSchema, evalChartConfigSchema, evalChartMetricSchema, evalChartTooltipExtraSchema, evalChartTypeSchema, evalChartsConfigSchema, evalFreshnessStatusSchema, evalLog, evalSpan, evalStatAggregateSchema, evalStatItemSchema, evalStatsConfigSchema, evalSummarySchema, evalTracer, extractApiCalls, extractCacheEntries, extractCacheHits, extractLlmCalls, fileRefSchema, getCaseRowCaseKey, getCaseRowEvalKey, getCurrentScope, getEvalCaseInput, getEvalDisplayStatus, getEvalRegistry, getEvalStartTime, getEvalTitle, getNestedAttribute, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, jsonCellSchema, llmCallMetricFormatSchema, llmCallMetricPlacementSchema, llmCallMetricSchema, llmCallPricingSchema, llmCallsConfigSchema, mergeEvalOutput, nextEvalId, numberDisplayOptionsSchema, removeDefaultConfigSchema, repoFile, repoFileRefSchema, resolveApiCallsConfig, resolveLlmCallsConfig, runArtifactRefSchema, runCli, runInEvalRuntimeScope, runInEvalScope, runInExistingEvalScope, runLogEntrySchema, runLogLevelSchema, runLogLocationSchema, runLogPhaseSchema, runLogsConfigSchema, runManifestSchema, runSummarySchema, scoreTraceSchema, serializeCacheRecording, serializeCacheValue, serializedCacheSpanSchema, setEvalOutput, setScopeCacheContext, spanCacheOptionsSchema, sseEnvelopeSchema, startEvalBackgroundJob, traceAttributeDisplayFormatSchema, traceAttributeDisplayInputSchema, traceAttributeDisplayPlacementSchema, traceAttributeDisplaySchema, traceCacheRefSchema, traceDisplayConfigSchema, traceDisplayInputConfigSchema, traceSpanErrorSchema, traceSpanKindSchema, traceSpanSchema, traceSpanWarningSchema, trialSelectionModeSchema, updateManualScoreRequestSchema, z };
|
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
import { n as initRunner, t as getRunnerInstance } from "./runner-
|
|
1
|
+
import { n as initRunner, t as getRunnerInstance } from "./runner-dnkl4pWd.mjs";
|
|
2
2
|
export { getRunnerInstance, initRunner };
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { n as createRunner } from "./cli-
|
|
2
|
-
import "./src-
|
|
1
|
+
import { n as createRunner } from "./cli-CmNNm6d0.mjs";
|
|
2
|
+
import "./src-DIFRuMuH.mjs";
|
|
3
3
|
//#region ../../apps/server/src/runner.ts
|
|
4
4
|
let runnerInstance = null;
|
|
5
5
|
function getRunnerInstance() {
|