agentv 4.15.1 → 4.15.2-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.
- package/dist/{chunk-CA3SJJN7.js → chunk-6X2UFVGI.js} +119 -41
- package/dist/chunk-6X2UFVGI.js.map +1 -0
- package/dist/{chunk-VFJ3UPNX.js → chunk-O4RLSNIR.js} +34 -37
- package/dist/chunk-O4RLSNIR.js.map +1 -0
- package/dist/{chunk-BUQ2JYIB.js → chunk-ZO4VTVKQ.js} +26 -14
- package/dist/chunk-ZO4VTVKQ.js.map +1 -0
- package/dist/cli.js +3 -3
- package/dist/{dist-524K2DD2.js → dist-7ZYP7EBD.js} +6 -4
- package/dist/index.js +3 -3
- package/dist/{interactive-G4WWA32J.js → interactive-ULCAOSHJ.js} +3 -3
- package/package.json +1 -1
- package/dist/chunk-BUQ2JYIB.js.map +0 -1
- package/dist/chunk-CA3SJJN7.js.map +0 -1
- package/dist/chunk-VFJ3UPNX.js.map +0 -1
- /package/dist/{dist-524K2DD2.js.map → dist-7ZYP7EBD.js.map} +0 -0
- /package/dist/{interactive-G4WWA32J.js.map → interactive-ULCAOSHJ.js.map} +0 -0
|
@@ -40,13 +40,14 @@ import {
|
|
|
40
40
|
subscribeToCopilotSdkLogEntries,
|
|
41
41
|
subscribeToPiLogEntries,
|
|
42
42
|
syncResultsRepo,
|
|
43
|
-
toCamelCaseDeep
|
|
44
|
-
|
|
43
|
+
toCamelCaseDeep,
|
|
44
|
+
toTranscriptJsonLines
|
|
45
|
+
} from "./chunk-6X2UFVGI.js";
|
|
45
46
|
|
|
46
47
|
// package.json
|
|
47
48
|
var package_default = {
|
|
48
49
|
name: "agentv",
|
|
49
|
-
version: "4.15.1",
|
|
50
|
+
version: "4.15.2-next.1",
|
|
50
51
|
description: "CLI entry point for AgentV",
|
|
51
52
|
type: "module",
|
|
52
53
|
repository: {
|
|
@@ -1447,6 +1448,31 @@ async function writeJsonlFile(filePath, records) {
|
|
|
1447
1448
|
`;
|
|
1448
1449
|
await writeFile(filePath, content, "utf8");
|
|
1449
1450
|
}
|
|
1451
|
+
function buildTranscriptMessageLines(results) {
|
|
1452
|
+
const lines = [];
|
|
1453
|
+
for (const result of results) {
|
|
1454
|
+
const transcriptLines = toTranscriptJsonLines(
|
|
1455
|
+
{
|
|
1456
|
+
messages: [...result.input ?? [], ...result.output],
|
|
1457
|
+
source: {
|
|
1458
|
+
provider: result.target,
|
|
1459
|
+
sessionId: result.conversationId ?? result.testId,
|
|
1460
|
+
startedAt: result.timestamp
|
|
1461
|
+
},
|
|
1462
|
+
tokenUsage: result.tokenUsage,
|
|
1463
|
+
durationMs: result.durationMs,
|
|
1464
|
+
costUsd: result.costUsd
|
|
1465
|
+
},
|
|
1466
|
+
{
|
|
1467
|
+
testId: result.testId,
|
|
1468
|
+
target: result.target
|
|
1469
|
+
}
|
|
1470
|
+
);
|
|
1471
|
+
lines.push(...transcriptLines.map((line) => JSON.stringify(line)));
|
|
1472
|
+
}
|
|
1473
|
+
return lines.length > 0 ? `${lines.join("\n")}
|
|
1474
|
+
` : "";
|
|
1475
|
+
}
|
|
1450
1476
|
async function writeArtifactsFromResults(results, outputDir, options) {
|
|
1451
1477
|
const testArtifactDir = outputDir;
|
|
1452
1478
|
const timingPath = path7.join(outputDir, "timing.json");
|
|
@@ -1492,36 +1518,7 @@ async function writeArtifactsFromResults(results, outputDir, options) {
|
|
|
1492
1518
|
`, "utf8");
|
|
1493
1519
|
await writeJsonlFile(indexPath, indexRecords);
|
|
1494
1520
|
const transcriptPath = path7.join(outputDir, "transcript.jsonl");
|
|
1495
|
-
|
|
1496
|
-
let inputText = "";
|
|
1497
|
-
if (typeof result.input === "string") {
|
|
1498
|
-
inputText = result.input;
|
|
1499
|
-
} else if (Array.isArray(result.input)) {
|
|
1500
|
-
const firstUserMsg = result.input.find((m) => m.role === "user");
|
|
1501
|
-
inputText = typeof firstUserMsg?.content === "string" ? firstUserMsg.content : "";
|
|
1502
|
-
}
|
|
1503
|
-
return {
|
|
1504
|
-
input: inputText,
|
|
1505
|
-
output: result.output,
|
|
1506
|
-
token_usage: result.tokenUsage ? {
|
|
1507
|
-
input: result.tokenUsage.input,
|
|
1508
|
-
output: result.tokenUsage.output,
|
|
1509
|
-
cached: result.tokenUsage.cached
|
|
1510
|
-
} : void 0,
|
|
1511
|
-
duration_ms: result.durationMs,
|
|
1512
|
-
cost_usd: result.costUsd,
|
|
1513
|
-
source: {
|
|
1514
|
-
provider: result.target,
|
|
1515
|
-
session_id: result.conversationId ?? result.testId,
|
|
1516
|
-
timestamp: result.timestamp
|
|
1517
|
-
}
|
|
1518
|
-
};
|
|
1519
|
-
});
|
|
1520
|
-
await writeFile(
|
|
1521
|
-
transcriptPath,
|
|
1522
|
-
transcriptLines.map((line) => JSON.stringify(line)).join("\n") + (transcriptLines.length ? "\n" : ""),
|
|
1523
|
-
"utf8"
|
|
1524
|
-
);
|
|
1521
|
+
await writeFile(transcriptPath, buildTranscriptMessageLines(results), "utf8");
|
|
1525
1522
|
return { testArtifactDir, timingPath, benchmarkPath, indexPath };
|
|
1526
1523
|
}
|
|
1527
1524
|
|
|
@@ -5351,7 +5348,7 @@ async function runEvalCommand(input) {
|
|
|
5351
5348
|
const useFileExport = !!options.otelFile;
|
|
5352
5349
|
if (options.exportOtel || useFileExport) {
|
|
5353
5350
|
try {
|
|
5354
|
-
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-
|
|
5351
|
+
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-7ZYP7EBD.js");
|
|
5355
5352
|
let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
|
|
5356
5353
|
let headers = {};
|
|
5357
5354
|
if (options.otelBackend) {
|
|
@@ -5547,7 +5544,7 @@ async function runEvalCommand(input) {
|
|
|
5547
5544
|
const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
|
|
5548
5545
|
let transcriptProviderFactory;
|
|
5549
5546
|
if (options.transcript) {
|
|
5550
|
-
const { TranscriptProvider } = await import("./dist-
|
|
5547
|
+
const { TranscriptProvider } = await import("./dist-7ZYP7EBD.js");
|
|
5551
5548
|
const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
|
|
5552
5549
|
const totalTests = [...fileMetadata.values()].reduce(
|
|
5553
5550
|
(sum, meta) => sum + meta.testCases.length,
|
|
@@ -5555,7 +5552,7 @@ async function runEvalCommand(input) {
|
|
|
5555
5552
|
);
|
|
5556
5553
|
if (transcriptProvider.lineCount !== totalTests) {
|
|
5557
5554
|
throw new Error(
|
|
5558
|
-
`Transcript has ${transcriptProvider.lineCount}
|
|
5555
|
+
`Transcript has ${transcriptProvider.lineCount} entr${transcriptProvider.lineCount === 1 ? "y" : "ies"} but eval defines ${totalTests} test(s). Each transcript entry maps positionally to one test case.`
|
|
5559
5556
|
);
|
|
5560
5557
|
}
|
|
5561
5558
|
transcriptProviderFactory = () => transcriptProvider;
|
|
@@ -5869,4 +5866,4 @@ export {
|
|
|
5869
5866
|
getCategories,
|
|
5870
5867
|
filterByCategory
|
|
5871
5868
|
};
|
|
5872
|
-
//# sourceMappingURL=chunk-
|
|
5869
|
+
//# sourceMappingURL=chunk-O4RLSNIR.js.map
|