agentv 4.31.3 → 4.32.0-next.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 (32) hide show
  1. package/README.md +14 -0
  2. package/dist/{artifact-writer-FZ5RUHWC.js → artifact-writer-VDF7KRWL.js} +4 -4
  3. package/dist/{chunk-IJPWTVDU.js → chunk-2ZEY3WBH.js} +3 -3
  4. package/dist/{chunk-CO23UZAA.js → chunk-5JMFFG36.js} +13 -9
  5. package/dist/chunk-5JMFFG36.js.map +1 -0
  6. package/dist/{chunk-4POCDPC2.js → chunk-IGTRNQAM.js} +106 -63
  7. package/dist/chunk-IGTRNQAM.js.map +1 -0
  8. package/dist/{chunk-LOYPSIE7.js → chunk-LX5AK3P7.js} +22 -12
  9. package/dist/chunk-LX5AK3P7.js.map +1 -0
  10. package/dist/{chunk-377GONL7.js → chunk-TAZBCVEZ.js} +66 -62
  11. package/dist/chunk-TAZBCVEZ.js.map +1 -0
  12. package/dist/cli.js +5 -5
  13. package/dist/cli.js.map +1 -1
  14. package/dist/{studio/assets/index-sazn_aO8.js → dashboard/assets/index-BdoQWnyM.js} +1 -1
  15. package/dist/{studio/assets/index-CYUVGyMn.css → dashboard/assets/index-DcPH8PyS.css} +1 -1
  16. package/dist/dashboard/assets/index-EXkiwqam.js +116 -0
  17. package/dist/{studio → dashboard}/index.html +3 -3
  18. package/dist/{dist-CRYAFKLS.js → dist-GICSKMNP.js} +5 -3
  19. package/dist/index.js +9 -7
  20. package/dist/{interactive-QBLQNYDH.js → interactive-GIDBBDYZ.js} +5 -5
  21. package/dist/{ts-eval-loader-XR6DNOZ3-GA7B5N26.js → ts-eval-loader-Z6IUSDNA-YBOE4JIQ.js} +2 -2
  22. package/package.json +1 -1
  23. package/dist/chunk-377GONL7.js.map +0 -1
  24. package/dist/chunk-4POCDPC2.js.map +0 -1
  25. package/dist/chunk-CO23UZAA.js.map +0 -1
  26. package/dist/chunk-LOYPSIE7.js.map +0 -1
  27. package/dist/studio/assets/index-DBSSri9Q.js +0 -116
  28. /package/dist/{artifact-writer-FZ5RUHWC.js.map → artifact-writer-VDF7KRWL.js.map} +0 -0
  29. /package/dist/{chunk-IJPWTVDU.js.map → chunk-2ZEY3WBH.js.map} +0 -0
  30. /package/dist/{dist-CRYAFKLS.js.map → dist-GICSKMNP.js.map} +0 -0
  31. /package/dist/{interactive-QBLQNYDH.js.map → interactive-GIDBBDYZ.js.map} +0 -0
  32. /package/dist/{ts-eval-loader-XR6DNOZ3-GA7B5N26.js.map → ts-eval-loader-Z6IUSDNA-YBOE4JIQ.js.map} +0 -0
package/README.md CHANGED
@@ -124,6 +124,20 @@ bun test
124
124
 
125
125
  See [AGENTS.md](AGENTS.md) for development guidelines.
126
126
 
127
+ ## Docker Dashboard Deployment
128
+
129
+ To simulate a one-command production deployment of AgentV Dashboard with the
130
+ AgentV examples project and a remote results repository:
131
+
132
+ ```bash
133
+ AGENTV_RESULTS_REPO=EntityProcess/agentv-evalresults \
134
+ scripts/setup-dashboard-deployment.sh
135
+ ```
136
+
137
+ The script clones AgentV examples into `~/agentv-dashboard`, clones the results
138
+ repo, writes the Dashboard project registry, builds the Docker image, and starts
139
+ Dashboard at `http://localhost:3117`.
140
+
127
141
  ## License
128
142
 
129
143
  MIT
@@ -14,11 +14,11 @@ import {
14
14
  writeArtifactsFromResults,
15
15
  writeInitialBenchmarkArtifact,
16
16
  writePerTestArtifacts
17
- } from "./chunk-IJPWTVDU.js";
18
- import "./chunk-LOYPSIE7.js";
17
+ } from "./chunk-2ZEY3WBH.js";
18
+ import "./chunk-LX5AK3P7.js";
19
19
  import "./chunk-QOBQ5XYF.js";
20
20
  import "./chunk-BPGJ4HBU.js";
21
- import "./chunk-377GONL7.js";
21
+ import "./chunk-TAZBCVEZ.js";
22
22
  import "./chunk-NPVGBFF6.js";
23
23
  import "./chunk-M7BUKBAF.js";
24
24
  import "./chunk-5H446C7X.js";
@@ -38,4 +38,4 @@ export {
38
38
  writeInitialBenchmarkArtifact,
39
39
  writePerTestArtifacts
40
40
  };
41
- //# sourceMappingURL=artifact-writer-FZ5RUHWC.js.map
41
+ //# sourceMappingURL=artifact-writer-VDF7KRWL.js.map
@@ -1,10 +1,10 @@
1
1
  import { createRequire } from 'node:module'; const require = createRequire(import.meta.url);
2
2
  import {
3
3
  toTranscriptJsonLines
4
- } from "./chunk-LOYPSIE7.js";
4
+ } from "./chunk-LX5AK3P7.js";
5
5
  import {
6
6
  DEFAULT_THRESHOLD
7
- } from "./chunk-377GONL7.js";
7
+ } from "./chunk-TAZBCVEZ.js";
8
8
 
9
9
  // src/commands/eval/artifact-writer.ts
10
10
  import { mkdir, readFile, writeFile } from "node:fs/promises";
@@ -726,4 +726,4 @@ export {
726
726
  writePerTestArtifacts,
727
727
  writeArtifactsFromResults
728
728
  };
729
- //# sourceMappingURL=chunk-IJPWTVDU.js.map
729
+ //# sourceMappingURL=chunk-2ZEY3WBH.js.map
@@ -16,7 +16,7 @@ import {
16
16
  toSnakeCaseDeep,
17
17
  writeArtifactsFromResults,
18
18
  writeInitialBenchmarkArtifact
19
- } from "./chunk-IJPWTVDU.js";
19
+ } from "./chunk-2ZEY3WBH.js";
20
20
  import {
21
21
  ResponseCache,
22
22
  RunBudgetTracker,
@@ -32,7 +32,7 @@ import {
32
32
  shouldEnableCache,
33
33
  shouldSkipCacheForTemperature,
34
34
  syncResultsRepo
35
- } from "./chunk-LOYPSIE7.js";
35
+ } from "./chunk-LX5AK3P7.js";
36
36
  import {
37
37
  CLI_PLACEHOLDERS,
38
38
  COMMON_TARGET_SETTINGS,
@@ -64,7 +64,7 @@ import {
64
64
  subscribeToCopilotSdkLogEntries,
65
65
  subscribeToPiLogEntries,
66
66
  toCamelCaseDeep
67
- } from "./chunk-377GONL7.js";
67
+ } from "./chunk-TAZBCVEZ.js";
68
68
 
69
69
  // src/commands/eval/shared.ts
70
70
  import { constants } from "node:fs";
@@ -180,7 +180,7 @@ async function findRepoRoot(start) {
180
180
  // package.json
181
181
  var package_default = {
182
182
  name: "agentv",
183
- version: "4.31.3",
183
+ version: "4.32.0-next.1",
184
184
  description: "CLI entry point for AgentV",
185
185
  type: "module",
186
186
  repository: {
@@ -1248,7 +1248,10 @@ async function ensureRemoteRunAvailable(cwd, meta) {
1248
1248
  if (relativeManifestPath.length === 0 || relativeManifestPath === meta.path || relativeManifestPath.startsWith("../")) {
1249
1249
  throw new Error(`Remote manifest path is outside the results repo clone: ${meta.path}`);
1250
1250
  }
1251
- const relativeRunPath = path5.posix.relative("runs", path5.posix.dirname(relativeManifestPath));
1251
+ const relativeRunPath = path5.posix.relative(
1252
+ ".agentv/results/runs",
1253
+ path5.posix.dirname(relativeManifestPath)
1254
+ );
1252
1255
  await materializeGitRun(config.path, relativeRunPath);
1253
1256
  }
1254
1257
  async function maybeAutoExportRunArtifacts(payload) {
@@ -4332,6 +4335,7 @@ async function validateConfigFile(filePath) {
4332
4335
  "required_version",
4333
4336
  "execution",
4334
4337
  "results",
4338
+ "dashboard",
4335
4339
  "studio"
4336
4340
  ]);
4337
4341
  const unexpectedFields = Object.keys(config).filter((key) => !allowedFields.has(key));
@@ -5555,7 +5559,7 @@ async function runEvalCommand(input) {
5555
5559
  const useFileExport = !!options.otelFile;
5556
5560
  if (options.exportOtel || useFileExport) {
5557
5561
  try {
5558
- const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-CRYAFKLS.js");
5562
+ const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-GICSKMNP.js");
5559
5563
  let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
5560
5564
  let headers = {};
5561
5565
  if (options.otelBackend) {
@@ -5754,7 +5758,7 @@ async function runEvalCommand(input) {
5754
5758
  const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
5755
5759
  let transcriptProviderFactory;
5756
5760
  if (options.transcript) {
5757
- const { TranscriptProvider } = await import("./dist-CRYAFKLS.js");
5761
+ const { TranscriptProvider } = await import("./dist-GICSKMNP.js");
5758
5762
  const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
5759
5763
  const totalTests = [...fileMetadata.values()].reduce(
5760
5764
  (sum, meta) => sum + meta.testCases.length,
@@ -5931,7 +5935,7 @@ async function runEvalCommand(input) {
5931
5935
  if (usesDefaultArtifactWorkspace && allResults.length > 0) {
5932
5936
  const evalFile = activeTestFiles.length === 1 ? activeTestFiles[0] : "";
5933
5937
  if (isResumeAppend) {
5934
- const { writePerTestArtifacts } = await import("./artifact-writer-FZ5RUHWC.js");
5938
+ const { writePerTestArtifacts } = await import("./artifact-writer-VDF7KRWL.js");
5935
5939
  await writePerTestArtifacts(allResults, runDir, {
5936
5940
  experiment: normalizeExperimentName(options.experiment)
5937
5941
  });
@@ -6156,4 +6160,4 @@ export {
6156
6160
  getCategories,
6157
6161
  filterByCategory
6158
6162
  };
6159
- //# sourceMappingURL=chunk-CO23UZAA.js.map
6163
+ //# sourceMappingURL=chunk-5JMFFG36.js.map