agentv 4.13.0 → 4.14.0
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.
- package/dist/{chunk-72AVLOSN.js → chunk-HP3ZUKYB.js} +846 -509
- package/dist/chunk-HP3ZUKYB.js.map +1 -0
- package/dist/{chunk-R747FXKW.js → chunk-IX3FYDXN.js} +5 -5
- package/dist/chunk-IX3FYDXN.js.map +1 -0
- package/dist/{chunk-I4BXYYI6.js → chunk-XDIQBOPH.js} +11 -26
- package/dist/chunk-XDIQBOPH.js.map +1 -0
- package/dist/cli.js +3 -3
- package/dist/{dist-SG5AIKQK.js → dist-Y3NRIBUT.js} +2 -2
- package/dist/index.js +3 -3
- package/dist/{interactive-ENAAA4RQ.js → interactive-TNCUWC46.js} +3 -3
- package/package.json +1 -1
- package/dist/chunk-72AVLOSN.js.map +0 -1
- package/dist/chunk-I4BXYYI6.js.map +0 -1
- package/dist/chunk-R747FXKW.js.map +0 -1
- /package/dist/{dist-SG5AIKQK.js.map → dist-Y3NRIBUT.js.map} +0 -0
- /package/dist/{interactive-ENAAA4RQ.js.map → interactive-TNCUWC46.js.map} +0 -0
|
@@ -41,12 +41,12 @@ import {
|
|
|
41
41
|
subscribeToPiLogEntries,
|
|
42
42
|
syncResultsRepo,
|
|
43
43
|
toCamelCaseDeep
|
|
44
|
-
} from "./chunk-
|
|
44
|
+
} from "./chunk-HP3ZUKYB.js";
|
|
45
45
|
|
|
46
46
|
// package.json
|
|
47
47
|
var package_default = {
|
|
48
48
|
name: "agentv",
|
|
49
|
-
version: "4.
|
|
49
|
+
version: "4.14.0",
|
|
50
50
|
description: "CLI entry point for AgentV",
|
|
51
51
|
type: "module",
|
|
52
52
|
repository: {
|
|
@@ -4997,18 +4997,6 @@ async function prepareFileMetadata(params) {
|
|
|
4997
4997
|
tags: suite.metadata?.tags
|
|
4998
4998
|
};
|
|
4999
4999
|
}
|
|
5000
|
-
async function runWithLimit(items, limit, task) {
|
|
5001
|
-
const safeLimit = Math.max(1, limit);
|
|
5002
|
-
let index = 0;
|
|
5003
|
-
const workers = Array.from({ length: safeLimit }, async () => {
|
|
5004
|
-
while (index < items.length) {
|
|
5005
|
-
const current = items[index];
|
|
5006
|
-
index += 1;
|
|
5007
|
-
await task(current);
|
|
5008
|
-
}
|
|
5009
|
-
});
|
|
5010
|
-
await Promise.all(workers);
|
|
5011
|
-
}
|
|
5012
5000
|
async function runSingleEvalFile(params) {
|
|
5013
5001
|
const {
|
|
5014
5002
|
testFilePath,
|
|
@@ -5254,7 +5242,7 @@ async function runEvalCommand(input) {
|
|
|
5254
5242
|
const useFileExport = !!options.otelFile;
|
|
5255
5243
|
if (options.exportOtel || useFileExport) {
|
|
5256
5244
|
try {
|
|
5257
|
-
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-
|
|
5245
|
+
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-Y3NRIBUT.js");
|
|
5258
5246
|
let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
|
|
5259
5247
|
let headers = {};
|
|
5260
5248
|
if (options.otelBackend) {
|
|
@@ -5312,12 +5300,7 @@ async function runEvalCommand(input) {
|
|
|
5312
5300
|
const remoteEvalSummaries = [];
|
|
5313
5301
|
const seenTestCases = /* @__PURE__ */ new Set();
|
|
5314
5302
|
const displayIdTracker = createDisplayIdTracker();
|
|
5315
|
-
const
|
|
5316
|
-
const fileConcurrency = Math.min(
|
|
5317
|
-
Math.max(1, totalWorkers),
|
|
5318
|
-
Math.max(1, resolvedTestFiles.length)
|
|
5319
|
-
);
|
|
5320
|
-
const perFileWorkers = options.workers ? Math.max(1, Math.floor(totalWorkers / fileConcurrency)) : void 0;
|
|
5303
|
+
const perFileWorkers = options.workers;
|
|
5321
5304
|
const fileMetadata = /* @__PURE__ */ new Map();
|
|
5322
5305
|
const tsFiles = [];
|
|
5323
5306
|
const yamlFiles = [];
|
|
@@ -5398,7 +5381,9 @@ async function runEvalCommand(input) {
|
|
|
5398
5381
|
}
|
|
5399
5382
|
throw new Error("No tests matched the provided filters.");
|
|
5400
5383
|
}
|
|
5401
|
-
const progressReporter = createProgressReporter(
|
|
5384
|
+
const progressReporter = createProgressReporter(options.workers ?? DEFAULT_WORKERS, {
|
|
5385
|
+
verbose: options.verbose
|
|
5386
|
+
});
|
|
5402
5387
|
progressReporter.start();
|
|
5403
5388
|
progressReporter.setTotal(totalEvalCount);
|
|
5404
5389
|
const seenCodexLogPaths = /* @__PURE__ */ new Set();
|
|
@@ -5453,7 +5438,7 @@ async function runEvalCommand(input) {
|
|
|
5453
5438
|
const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
|
|
5454
5439
|
let transcriptProviderFactory;
|
|
5455
5440
|
if (options.transcript) {
|
|
5456
|
-
const { TranscriptProvider } = await import("./dist-
|
|
5441
|
+
const { TranscriptProvider } = await import("./dist-Y3NRIBUT.js");
|
|
5457
5442
|
const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
|
|
5458
5443
|
const totalTests = [...fileMetadata.values()].reduce(
|
|
5459
5444
|
(sum, meta) => sum + meta.testCases.length,
|
|
@@ -5470,7 +5455,7 @@ async function runEvalCommand(input) {
|
|
|
5470
5455
|
);
|
|
5471
5456
|
}
|
|
5472
5457
|
try {
|
|
5473
|
-
|
|
5458
|
+
for (const testFilePath of activeTestFiles) {
|
|
5474
5459
|
const targetPrep = fileMetadata.get(testFilePath);
|
|
5475
5460
|
if (!targetPrep) {
|
|
5476
5461
|
throw new Error(`Missing metadata for ${testFilePath}`);
|
|
@@ -5555,7 +5540,7 @@ async function runEvalCommand(input) {
|
|
|
5555
5540
|
for (const results of targetResults) {
|
|
5556
5541
|
allResults.push(...results);
|
|
5557
5542
|
}
|
|
5558
|
-
}
|
|
5543
|
+
}
|
|
5559
5544
|
progressReporter.finish();
|
|
5560
5545
|
if (retryNonErrorResults && retryNonErrorResults.length > 0) {
|
|
5561
5546
|
for (const preserved of retryNonErrorResults) {
|
|
@@ -5775,4 +5760,4 @@ export {
|
|
|
5775
5760
|
getCategories,
|
|
5776
5761
|
filterByCategory
|
|
5777
5762
|
};
|
|
5778
|
-
//# sourceMappingURL=chunk-
|
|
5763
|
+
//# sourceMappingURL=chunk-XDIQBOPH.js.map
|