@ls-stack/agent-eval 0.55.2 → 0.56.1

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 (18) hide show
  1. package/dist/{app-NI4to6lp.mjs → app-Db_x-Rit.mjs} +4 -4
  2. package/dist/apps/web/dist/assets/{index-2I-eWzVL.css → index-Xa_7PteQ.css} +1 -1
  3. package/dist/apps/web/dist/assets/index-bB8IBDp1.js +377 -0
  4. package/dist/apps/web/dist/index.html +2 -2
  5. package/dist/bin.mjs +1 -1
  6. package/dist/caseChild.mjs +2 -1
  7. package/dist/{cli-Bu9347r1.mjs → cli-Ck0mqxd-.mjs} +4 -4
  8. package/dist/index.d.mts +593 -58
  9. package/dist/index.mjs +3 -3
  10. package/dist/runChild.mjs +2 -2
  11. package/dist/{runExecution-C31dpemR.mjs → runExecution-BH7DlMXl.mjs} +100 -8
  12. package/dist/{runOrchestration-3RoHLW4U.mjs → runOrchestration-C1Ex9QI-.mjs} +1 -1
  13. package/dist/{runner-B4EfMn1d.mjs → runner-B3hEOT_I.mjs} +2 -2
  14. package/dist/{runner-CTp9zHbM.mjs → runner-DbVYcapC.mjs} +1 -1
  15. package/dist/{src-FR60ZR_4.mjs → src-B3iq-tuv.mjs} +2 -2
  16. package/package.json +3 -3
  17. package/skills/agent-eval/SKILL.md +13 -3
  18. package/dist/apps/web/dist/assets/index-C7QjETk8.js +0 -377
package/dist/apps/web/dist/index.html CHANGED
@@ -25,8 +25,8 @@
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-C7QjETk8.js"></script>
29
- <link rel="stylesheet" crossorigin href="/assets/index-2I-eWzVL.css">
28
+ <script type="module" crossorigin src="/assets/index-bB8IBDp1.js"></script>
29
+ <link rel="stylesheet" crossorigin href="/assets/index-Xa_7PteQ.css">
30
30
  </head>
31
31
  <body>
32
32
  <div id="root"></div>
package/dist/bin.mjs CHANGED
@@ -1,5 +1,5 @@
1
1
  #!/usr/bin/env node
2
- import { t as runCli } from "./cli-Bu9347r1.mjs";
2
+ import { t as runCli } from "./cli-Ck0mqxd-.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/caseChild.mjs CHANGED
@@ -1,4 +1,4 @@
1
- import { I as configureEvalRunLogs, Pt as runWithEvalRegistry, St as resolveLlmCallsConfig, _ as createBufferedCacheStore, a as isCaseChildParentMessage, d as loadEvalModule, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, q as runInEvalRuntimeScope, r as runCase, v as createFsCacheStore, xt as resolveApiCallsConfig } from "./runExecution-C31dpemR.mjs";
1
+ import { I as configureEvalRunLogs, Pt as runWithEvalRegistry, St as resolveLlmCallsConfig, _ as createBufferedCacheStore, a as isCaseChildParentMessage, d as loadEvalModule, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, q as runInEvalRuntimeScope, r as runCase, v as createFsCacheStore, xt as resolveApiCallsConfig } from "./runExecution-BH7DlMXl.mjs";
2
2
  //#region ../runner/src/caseChild.ts
3
3
  let fatalErrorReported = false;
4
4
  let disconnectExpected = false;
@@ -101,6 +101,7 @@ async function executeCaseChild(context) {
101
101
  cacheHits: caseRowUpdate.cacheHits ?? 0,
102
102
  cacheOperations: caseRowUpdate.cacheOperations ?? 0,
103
103
  columns: caseRowUpdate.columns ?? {},
104
+ ...caseRowUpdate.outputColumnDefs !== void 0 ? { outputColumnDefs: caseRowUpdate.outputColumnDefs } : {},
104
105
  trial: context.trial
105
106
  },
106
107
  pendingCacheWrites: bufferedCacheStore?.getPendingWrites() ?? []
package/dist/{cli-Bu9347r1.mjs → cli-Ck0mqxd-.mjs} RENAMED
@@ -1,5 +1,5 @@
1
- import { Ct as buildEvalKey, Nt as getEvalRegistry, St as resolveLlmCallsConfig, bt as runSummarySchema, c as resolveArtifactPath, ct as applyDerivedCallAttributes, dt as getEvalDisplayStatus, f as resolveEvalDefaultConfig, ft as deriveScopedSummaryFromCases, gt as matchesTagsFilter, h as normalizeScoreDef, m as buildDeclaredColumnDefs, o as stripTerminalControlCodes, p as loadConfig, s as resolveTracePresentation, ut as getEvalTitle, v as createFsCacheStore, wt as getCaseRowCaseKey, xt as resolveApiCallsConfig } from "./runExecution-C31dpemR.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-3RoHLW4U.mjs";
1
+ import { Ct as buildEvalKey, Nt as getEvalRegistry, St as resolveLlmCallsConfig, bt as runSummarySchema, c as resolveArtifactPath, ct as applyDerivedCallAttributes, dt as getEvalDisplayStatus, f as resolveEvalDefaultConfig, ft as deriveScopedSummaryFromCases, gt as matchesTagsFilter, h as normalizeScoreDef, m as buildDeclaredColumnDefs, o as stripTerminalControlCodes, p as loadConfig, s as resolveTracePresentation, ut as getEvalTitle, v as createFsCacheStore, wt as getCaseRowCaseKey, xt as resolveApiCallsConfig } from "./runExecution-BH7DlMXl.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-C1Ex9QI-.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";
@@ -2171,8 +2171,8 @@ async function commandApp(args) {
2171
2171
  const { serve } = await import("@hono/node-server");
2172
2172
  const bundledWebDist = resolve(currentDir, "apps/web/dist");
2173
2173
  if (existsSync(bundledWebDist)) process.env.AGENT_EVALS_WEB_DIST = bundledWebDist;
2174
- const appModule = await import("./app-NI4to6lp.mjs");
2175
- const runnerModule = await import("./runner-CTp9zHbM.mjs");
2174
+ const appModule = await import("./app-Db_x-Rit.mjs");
2175
+ const runnerModule = await import("./runner-DbVYcapC.mjs");
2176
2176
  if (!isHonoAppModule(appModule)) throw new Error("Server app module is invalid");
2177
2177
  if (!isServerRunnerModule(runnerModule)) throw new Error("Server runner module is invalid");
2178
2178
  await runnerModule.initRunner();