@ls-stack/agent-eval 0.58.2 → 0.58.3
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-BxD6aHbp.mjs → app-ROCEce9X.mjs} +4 -4
- package/dist/apps/web/dist/assets/{index-BMWBZw_u.js → index-PTikBbhf.js} +65 -65
- package/dist/apps/web/dist/index.html +1 -1
- package/dist/bin.mjs +1 -1
- package/dist/caseChild.mjs +1 -1
- package/dist/{cli-HBwXIJsg.mjs → cli-SP4kEtYL.mjs} +4 -4
- package/dist/index.d.mts +116 -120
- package/dist/index.mjs +3 -3
- package/dist/runChild.mjs +2 -2
- package/dist/{runExecution-pHJ0_TzH.mjs → runExecution-CFw0MQFs.mjs} +5 -16
- package/dist/{runOrchestration-ngVXShH4.mjs → runOrchestration-CxjiQmof.mjs} +1 -1
- package/dist/{runner-BnZMGBla.mjs → runner-BlFQyvN2.mjs} +1 -1
- package/dist/{runner-D_pz2NON.mjs → runner-CY3bgsjU.mjs} +2 -2
- package/dist/{src-AeXGBJ26.mjs → src-7GbQj1sb.mjs} +2 -2
- package/package.json +1 -1
- package/skills/agent-eval/SKILL.md +5 -4
|
@@ -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-PTikBbhf.js"></script>
|
|
29
29
|
<link rel="stylesheet" crossorigin href="/assets/index-CHH7m5Cv.css">
|
|
30
30
|
</head>
|
|
31
31
|
<body>
|
package/dist/bin.mjs
CHANGED
package/dist/caseChild.mjs
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { Ct as resolveLlmCallsConfig, It as runWithEvalRegistry, J as runInEvalRuntimeScope, L as configureEvalRunLogs, St as resolveApiCallsConfig, _ as createBufferedCacheStore, a as isCaseChildParentMessage, d as loadEvalModule, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, r as runCase, v as createFsCacheStore } from "./runExecution-
|
|
1
|
+
import { Ct as resolveLlmCallsConfig, It as runWithEvalRegistry, J as runInEvalRuntimeScope, L as configureEvalRunLogs, St as resolveApiCallsConfig, _ as createBufferedCacheStore, a as isCaseChildParentMessage, d as loadEvalModule, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, r as runCase, v as createFsCacheStore } from "./runExecution-CFw0MQFs.mjs";
|
|
2
2
|
//#region ../runner/src/caseChild.ts
|
|
3
3
|
let fatalErrorReported = false;
|
|
4
4
|
let disconnectExpected = false;
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { Ct as resolveLlmCallsConfig, Et as getCaseRowCaseKey, Ft as getEvalRegistry, Ot as caseRowSchema, St as resolveApiCallsConfig, Tt as buildEvalKey, _t as matchesTagsFilter, c as resolveArtifactPath, dt as getEvalTitle, f as resolveEvalDefaultConfig, ft as getEvalDisplayStatus, h as normalizeScoreDef, lt as applyDerivedCallAttributes, m as buildDeclaredColumnDefs, o as stripTerminalControlCodes, p as loadConfig, pt as deriveScopedSummaryFromCases, s as resolveTracePresentation, v as createFsCacheStore, xt as runSummarySchema } from "./runExecution-
|
|
2
|
-
import { C as validateCharts, S as parseEvalDiscovery, _ as runTouchesEval, a as validateTagsFilters, b as deriveEvalFreshness, c as getLatestRunInfos, d as nextShortIdFromSnapshots, f as persistCaseDetail, g as recomputePersistedCaseStatus, h as recomputeEvalStatusesInRuns, i as resolveEvalTags, l as loadPersistedRunSnapshot, m as persistRunState, n as getTargetEvalKeys, o as generateRunId, p as deleteTemporaryRuns, s as getLastRunStatuses, u as loadPersistedRunSnapshots, v as buildManualInputDescriptor, x as loadIsolatedEvalRegistry, y as parseManualInputValues } from "./runOrchestration-
|
|
1
|
+
import { Ct as resolveLlmCallsConfig, Et as getCaseRowCaseKey, Ft as getEvalRegistry, Ot as caseRowSchema, St as resolveApiCallsConfig, Tt as buildEvalKey, _t as matchesTagsFilter, c as resolveArtifactPath, dt as getEvalTitle, f as resolveEvalDefaultConfig, ft as getEvalDisplayStatus, h as normalizeScoreDef, lt as applyDerivedCallAttributes, m as buildDeclaredColumnDefs, o as stripTerminalControlCodes, p as loadConfig, pt as deriveScopedSummaryFromCases, s as resolveTracePresentation, v as createFsCacheStore, xt as runSummarySchema } from "./runExecution-CFw0MQFs.mjs";
|
|
2
|
+
import { C as validateCharts, S as parseEvalDiscovery, _ as runTouchesEval, a as validateTagsFilters, b as deriveEvalFreshness, c as getLatestRunInfos, d as nextShortIdFromSnapshots, f as persistCaseDetail, g as recomputePersistedCaseStatus, h as recomputeEvalStatusesInRuns, i as resolveEvalTags, l as loadPersistedRunSnapshot, m as persistRunState, n as getTargetEvalKeys, o as generateRunId, p as deleteTemporaryRuns, s as getLastRunStatuses, u as loadPersistedRunSnapshots, v as buildManualInputDescriptor, x as loadIsolatedEvalRegistry, y as parseManualInputValues } from "./runOrchestration-CxjiQmof.mjs";
|
|
3
3
|
import { copyFile, mkdir, readFile, rm, writeFile } from "node:fs/promises";
|
|
4
4
|
import { basename, dirname, extname, isAbsolute, join, relative, resolve, sep } from "node:path";
|
|
5
5
|
import { createHash, randomUUID } from "node:crypto";
|
|
@@ -2198,8 +2198,8 @@ async function commandApp(args) {
|
|
|
2198
2198
|
const { serve } = await import("@hono/node-server");
|
|
2199
2199
|
const bundledWebDist = resolve(currentDir, "apps/web/dist");
|
|
2200
2200
|
if (existsSync(bundledWebDist)) process.env.AGENT_EVALS_WEB_DIST = bundledWebDist;
|
|
2201
|
-
const appModule = await import("./app-
|
|
2202
|
-
const runnerModule = await import("./runner-
|
|
2201
|
+
const appModule = await import("./app-ROCEce9X.mjs");
|
|
2202
|
+
const runnerModule = await import("./runner-BlFQyvN2.mjs");
|
|
2203
2203
|
if (!isHonoAppModule(appModule)) throw new Error("Server app module is invalid");
|
|
2204
2204
|
if (!isServerRunnerModule(runnerModule)) throw new Error("Server runner module is invalid");
|
|
2205
2205
|
await runnerModule.initRunner();
|