@ls-stack/agent-eval 0.42.0 → 0.42.1

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.
Files changed (14) hide show
  1. package/dist/{app-1vE5Ryry.mjs → app-mOYjX9zq.mjs} +4 -4
  2. package/dist/apps/web/dist/assets/{index-5CB9eJZy.js → index-CANDLTsq.js} +2 -2
  3. package/dist/apps/web/dist/index.html +1 -1
  4. package/dist/bin.mjs +1 -1
  5. package/dist/{cli-Bk5g-bat.mjs → cli-DbVfkr9T.mjs} +3 -3
  6. package/dist/index.d.mts +108 -107
  7. package/dist/index.mjs +3 -3
  8. package/dist/runChild.mjs +1 -1
  9. package/dist/{runOrchestration-DhTiT4V0.mjs → runOrchestration-SPaHx-SC.mjs} +6 -7
  10. package/dist/{runner-B1Cyevvr.mjs → runner-BYOdLBle.mjs} +1 -1
  11. package/dist/{runner-BG0L4yId.mjs → runner-DiCQ57JQ.mjs} +2 -2
  12. package/dist/{src-t6OVp1li.mjs → src-CANi3gpd.mjs} +2 -2
  13. package/package.json +1 -1
  14. package/skills/agent-eval/SKILL.md +1 -1
package/dist/apps/web/dist/index.html CHANGED
@@ -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-5CB9eJZy.js"></script>
28
+ <script type="module" crossorigin src="/assets/index-CANDLTsq.js"></script>
29
29
  <link rel="stylesheet" crossorigin href="/assets/index-eFM9VIsz.css">
30
30
  </head>
31
31
  <body>
package/dist/bin.mjs CHANGED
@@ -1,5 +1,5 @@
1
1
  #!/usr/bin/env node
2
- import { t as runCli } from "./cli-Bk5g-bat.mjs";
2
+ import { t as runCli } from "./cli-DbVfkr9T.mjs";
3
3
  import { spawn } from "node:child_process";
4
4
  //#region src/bin.ts
5
5
  const moduleMocksFlag = "--experimental-test-module-mocks";
package/dist/{cli-Bk5g-bat.mjs → cli-DbVfkr9T.mjs} RENAMED
@@ -1,4 +1,4 @@
1
- import { A as validateCharts, At as buildEvalKey, C as deriveEvalFreshness, Ct as getEvalDisplayStatus, D as loadConfig, Dt as runSummarySchema, E as resolveEvalDefaultConfig, Mt as getCaseRowEvalKey, O as buildDeclaredColumnDefs, Ot as resolveApiCallsConfig, Rt as getEvalRegistry, S as parseManualInputValues, St as getEvalTitle, T as parseEvalDiscovery, Tt as matchesTagsFilter, _ as recomputePersistedCaseStatus, a as validateTagsFilters, b as resolveArtifactPath, bt as applyDerivedCallAttributes, c as getLastRunStatuses, d as loadPersistedRunSnapshots, f as nextShortIdFromSnapshots, g as recomputeEvalStatusesInRuns, h as persistRunState, i as resolveEvalTags, j as createFsCacheStore, jt as getCaseRowCaseKey, k as normalizeScoreDef, kt as resolveLlmCallsConfig, l as getLatestRunInfos, m as deleteTemporaryRuns, n as getTargetEvalKeys, o as stripTerminalControlCodes, p as persistCaseDetail, s as generateRunId, u as loadPersistedRunSnapshot, v as runTouchesEval, w as loadEvalModule, wt as deriveScopedSummaryFromCases, x as buildManualInputDescriptor, y as resolveTracePresentation } from "./runOrchestration-DhTiT4V0.mjs";
1
+ import { A as validateCharts, At as buildEvalKey, C as deriveEvalFreshness, Ct as getEvalDisplayStatus, D as loadConfig, Dt as runSummarySchema, E as resolveEvalDefaultConfig, Mt as getCaseRowEvalKey, O as buildDeclaredColumnDefs, Ot as resolveApiCallsConfig, Rt as getEvalRegistry, S as parseManualInputValues, St as getEvalTitle, T as parseEvalDiscovery, Tt as matchesTagsFilter, _ as recomputePersistedCaseStatus, a as validateTagsFilters, b as resolveArtifactPath, bt as applyDerivedCallAttributes, c as getLastRunStatuses, d as loadPersistedRunSnapshots, f as nextShortIdFromSnapshots, g as recomputeEvalStatusesInRuns, h as persistRunState, i as resolveEvalTags, j as createFsCacheStore, jt as getCaseRowCaseKey, k as normalizeScoreDef, kt as resolveLlmCallsConfig, l as getLatestRunInfos, m as deleteTemporaryRuns, n as getTargetEvalKeys, o as stripTerminalControlCodes, p as persistCaseDetail, s as generateRunId, u as loadPersistedRunSnapshot, v as runTouchesEval, w as loadEvalModule, wt as deriveScopedSummaryFromCases, x as buildManualInputDescriptor, y as resolveTracePresentation } from "./runOrchestration-SPaHx-SC.mjs";
2
2
  import { copyFile, mkdir, readFile, rm, writeFile } from "node:fs/promises";
3
3
  import { basename, dirname, extname, isAbsolute, join, relative, resolve, sep } from "node:path";
4
4
  import { createHash, randomUUID } from "node:crypto";
@@ -2046,8 +2046,8 @@ async function commandApp(args) {
2046
2046
  const { serve } = await import("@hono/node-server");
2047
2047
  const bundledWebDist = resolve(currentDir, "apps/web/dist");
2048
2048
  if (existsSync(bundledWebDist)) process.env.AGENT_EVALS_WEB_DIST = bundledWebDist;
2049
- const appModule = await import("./app-1vE5Ryry.mjs");
2050
- const runnerModule = await import("./runner-B1Cyevvr.mjs");
2049
+ const appModule = await import("./app-mOYjX9zq.mjs");
2050
+ const runnerModule = await import("./runner-BYOdLBle.mjs");
2051
2051
  if (!isHonoAppModule(appModule)) throw new Error("Server app module is invalid");
2052
2052
  if (!isServerRunnerModule(runnerModule)) throw new Error("Server runner module is invalid");
2053
2053
  await runnerModule.initRunner();