@ls-stack/agent-eval 0.29.0 → 0.30.0

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.
Files changed (15) hide show
  1. package/dist/{app-D6-msfKP.mjs → app-CbOZBHju.mjs} +4 -4
  2. package/dist/apps/web/dist/assets/{index-BCr6J8Uj.js → index-DEikHy2a.js} +35 -35
  3. package/dist/apps/web/dist/index.html +1 -1
  4. package/dist/bin.mjs +1 -1
  5. package/dist/{cli-CIc_gBNM.mjs → cli-CiFOqMwS.mjs} +3 -3
  6. package/dist/index.d.mts +133 -109
  7. package/dist/index.mjs +3 -3
  8. package/dist/runChild.mjs +1 -1
  9. package/dist/{runOrchestration-CIARrLs6.mjs → runOrchestration-CO3Vf0cQ.mjs} +35 -9
  10. package/dist/{runner-Bq1f9B9d.mjs → runner-4pF_Qrc9.mjs} +1 -1
  11. package/dist/{runner-1F8MeY5V.mjs → runner-CXHkf7ih.mjs} +2 -2
  12. package/dist/src-BiPLv9ya.mjs +3 -0
  13. package/package.json +1 -1
  14. package/skills/agent-eval/SKILL.md +8 -5
  15. package/dist/src-CkWT1iSu.mjs +0 -3
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-BCr6J8Uj.js"></script>
28
+ <script type="module" crossorigin src="/assets/index-DEikHy2a.js"></script>
29
29
  <link rel="stylesheet" crossorigin href="/assets/index-DjUTm3M-.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-CIc_gBNM.mjs";
2
+ import { t as runCli } from "./cli-CiFOqMwS.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-CIc_gBNM.mjs → cli-CiFOqMwS.mjs} RENAMED
@@ -1,4 +1,4 @@
1
- import { B as getEvalDisplayStatus, C as loadConfig, D as createFsCacheStore, E as validateCharts, G as runSummarySchema, L as applyDerivedCallAttributes, S as resolveEvalDefaultConfig, T as normalizeScoreDef, V as deriveScopedSummaryFromCases, _ as buildManualInputDescriptor, _t as getCaseRowEvalKey, a as getLastRunStatuses, b as loadEvalModule, c as loadPersistedRunSnapshots, d as persistRunState, dt as resolveLlmCallsConfig, f as recomputeEvalStatusesInRuns, g as resolveArtifactPath, gt as getCaseRowCaseKey, h as resolveTracePresentation, ht as buildEvalKey, i as generateRunId, l as nextShortIdFromSnapshots, m as runTouchesEval, n as getTargetEvalKeys, o as getLatestRunInfos, p as recomputePersistedCaseStatus, s as loadPersistedRunSnapshot, u as persistCaseDetail, ur as getEvalRegistry, ut as resolveApiCallsConfig, v as parseManualInputValues, w as buildDeclaredColumnDefs, x as parseEvalDiscovery, y as deriveEvalFreshness, z as getEvalTitle } from "./runOrchestration-CIARrLs6.mjs";
1
+ import { B as getEvalDisplayStatus, C as loadConfig, D as createFsCacheStore, E as validateCharts, G as runSummarySchema, L as applyDerivedCallAttributes, S as resolveEvalDefaultConfig, T as normalizeScoreDef, V as deriveScopedSummaryFromCases, _ as buildManualInputDescriptor, _t as getCaseRowEvalKey, a as getLastRunStatuses, b as loadEvalModule, c as loadPersistedRunSnapshots, d as persistRunState, dt as resolveLlmCallsConfig, f as recomputeEvalStatusesInRuns, g as resolveArtifactPath, gt as getCaseRowCaseKey, h as resolveTracePresentation, ht as buildEvalKey, i as generateRunId, l as nextShortIdFromSnapshots, m as runTouchesEval, n as getTargetEvalKeys, o as getLatestRunInfos, p as recomputePersistedCaseStatus, s as loadPersistedRunSnapshot, u as persistCaseDetail, ur as getEvalRegistry, ut as resolveApiCallsConfig, v as parseManualInputValues, w as buildDeclaredColumnDefs, x as parseEvalDiscovery, y as deriveEvalFreshness, z as getEvalTitle } from "./runOrchestration-CO3Vf0cQ.mjs";
2
2
  import { createHash, randomUUID } from "node:crypto";
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";
@@ -1940,8 +1940,8 @@ async function commandApp(args) {
1940
1940
  const { serve } = await import("@hono/node-server");
1941
1941
  const bundledWebDist = resolve(currentDir, "apps/web/dist");
1942
1942
  if (existsSync(bundledWebDist)) process.env.AGENT_EVALS_WEB_DIST = bundledWebDist;
1943
- const appModule = await import("./app-D6-msfKP.mjs");
1944
- const runnerModule = await import("./runner-Bq1f9B9d.mjs");
1943
+ const appModule = await import("./app-CbOZBHju.mjs");
1944
+ const runnerModule = await import("./runner-4pF_Qrc9.mjs");
1945
1945
  if (!isHonoAppModule(appModule)) throw new Error("Server app module is invalid");
1946
1946
  if (!isServerRunnerModule(runnerModule)) throw new Error("Server runner module is invalid");
1947
1947
  await runnerModule.initRunner();