agentv 4.23.0-next.1 → 4.24.0-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.
- package/dist/{artifact-writer-KBYDJNKS.js → artifact-writer-7CKE5ILM.js} +4 -4
- package/dist/{chunk-4ISLX6RP.js → chunk-AYXP3MTM.js} +6 -6
- package/dist/{chunk-SBI4Z7YY.js → chunk-D5HCHBPT.js} +2 -2
- package/dist/{chunk-KX2MJDMY.js → chunk-HL3MTSIW.js} +14 -9
- package/dist/{chunk-KX2MJDMY.js.map → chunk-HL3MTSIW.js.map} +1 -1
- package/dist/{chunk-ZVEESBXF.js → chunk-KTHVVL7Q.js} +3 -3
- package/dist/{chunk-32VOBCSX.js → chunk-TRA6SVUM.js} +24 -11
- package/dist/chunk-TRA6SVUM.js.map +1 -0
- package/dist/cli.js +5 -5
- package/dist/{dist-A7D52JMJ.js → dist-IA3D556U.js} +3 -3
- package/dist/index.js +5 -5
- package/dist/{interactive-66Z7DNMJ.js → interactive-YXXNDPKI.js} +5 -5
- package/dist/studio/assets/index-DY-K3PKr.css +1 -0
- package/dist/studio/assets/{index-SZVrc1UE.js → index-DZpM1lzZ.js} +25 -25
- package/dist/studio/assets/{index-BVSHI8Eo.js → index-jBBmFdWd.js} +1 -1
- package/dist/studio/index.html +2 -2
- package/dist/{ts-eval-loader-HT7NCO4M-V63KENQZ.js → ts-eval-loader-SYQYQPMC-FKU23VX3.js} +2 -2
- package/package.json +1 -1
- package/dist/chunk-32VOBCSX.js.map +0 -1
- package/dist/studio/assets/index-DcpbFwI5.css +0 -1
- /package/dist/{artifact-writer-KBYDJNKS.js.map → artifact-writer-7CKE5ILM.js.map} +0 -0
- /package/dist/{chunk-4ISLX6RP.js.map → chunk-AYXP3MTM.js.map} +0 -0
- /package/dist/{chunk-SBI4Z7YY.js.map → chunk-D5HCHBPT.js.map} +0 -0
- /package/dist/{chunk-ZVEESBXF.js.map → chunk-KTHVVL7Q.js.map} +0 -0
- /package/dist/{dist-A7D52JMJ.js.map → dist-IA3D556U.js.map} +0 -0
- /package/dist/{interactive-66Z7DNMJ.js.map → interactive-YXXNDPKI.js.map} +0 -0
- /package/dist/{ts-eval-loader-HT7NCO4M-V63KENQZ.js.map → ts-eval-loader-SYQYQPMC-FKU23VX3.js.map} +0 -0
|
@@ -13,11 +13,11 @@ import {
|
|
|
13
13
|
writeArtifacts,
|
|
14
14
|
writeArtifactsFromResults,
|
|
15
15
|
writePerTestArtifacts
|
|
16
|
-
} from "./chunk-
|
|
17
|
-
import "./chunk-
|
|
16
|
+
} from "./chunk-KTHVVL7Q.js";
|
|
17
|
+
import "./chunk-D5HCHBPT.js";
|
|
18
18
|
import "./chunk-QOBQ5XYF.js";
|
|
19
19
|
import "./chunk-BPGJ4HBU.js";
|
|
20
|
-
import "./chunk-
|
|
20
|
+
import "./chunk-HL3MTSIW.js";
|
|
21
21
|
import "./chunk-ZKO2LGRR.js";
|
|
22
22
|
import "./chunk-NPVGBFF6.js";
|
|
23
23
|
import "./chunk-HQDCIXVH.js";
|
|
@@ -38,4 +38,4 @@ export {
|
|
|
38
38
|
writeArtifactsFromResults,
|
|
39
39
|
writePerTestArtifacts
|
|
40
40
|
};
|
|
41
|
-
//# sourceMappingURL=artifact-writer-
|
|
41
|
+
//# sourceMappingURL=artifact-writer-7CKE5ILM.js.map
|
|
@@ -44,7 +44,7 @@ import {
|
|
|
44
44
|
validateFileReferences,
|
|
45
45
|
validateTargetsFile,
|
|
46
46
|
validateWorkspacePaths
|
|
47
|
-
} from "./chunk-
|
|
47
|
+
} from "./chunk-TRA6SVUM.js";
|
|
48
48
|
import {
|
|
49
49
|
RESULT_INDEX_FILENAME,
|
|
50
50
|
aggregateRunDir,
|
|
@@ -52,7 +52,7 @@ import {
|
|
|
52
52
|
resolveRunManifestPath,
|
|
53
53
|
toSnakeCaseDeep as toSnakeCaseDeep2,
|
|
54
54
|
writeArtifactsFromResults
|
|
55
|
-
} from "./chunk-
|
|
55
|
+
} from "./chunk-KTHVVL7Q.js";
|
|
56
56
|
import {
|
|
57
57
|
DEFAULT_CATEGORY,
|
|
58
58
|
addBenchmark,
|
|
@@ -71,7 +71,7 @@ import {
|
|
|
71
71
|
toTranscriptJsonLines,
|
|
72
72
|
transpileEvalYamlFile,
|
|
73
73
|
trimBaselineResult
|
|
74
|
-
} from "./chunk-
|
|
74
|
+
} from "./chunk-D5HCHBPT.js";
|
|
75
75
|
import {
|
|
76
76
|
DEFAULT_THRESHOLD,
|
|
77
77
|
createBuiltinRegistry,
|
|
@@ -101,7 +101,7 @@ import {
|
|
|
101
101
|
runStartsWithAssertion,
|
|
102
102
|
toCamelCaseDeep,
|
|
103
103
|
toSnakeCaseDeep
|
|
104
|
-
} from "./chunk-
|
|
104
|
+
} from "./chunk-HL3MTSIW.js";
|
|
105
105
|
import {
|
|
106
106
|
__commonJS,
|
|
107
107
|
__require,
|
|
@@ -3996,7 +3996,7 @@ var evalRunCommand = command({
|
|
|
3996
3996
|
},
|
|
3997
3997
|
handler: async (args) => {
|
|
3998
3998
|
if (args.evalPaths.length === 0 && process.stdin.isTTY) {
|
|
3999
|
-
const { launchInteractiveWizard } = await import("./interactive-
|
|
3999
|
+
const { launchInteractiveWizard } = await import("./interactive-YXXNDPKI.js");
|
|
4000
4000
|
await launchInteractiveWizard();
|
|
4001
4001
|
return;
|
|
4002
4002
|
}
|
|
@@ -11974,4 +11974,4 @@ export {
|
|
|
11974
11974
|
preprocessArgv,
|
|
11975
11975
|
runCli
|
|
11976
11976
|
};
|
|
11977
|
-
//# sourceMappingURL=chunk-
|
|
11977
|
+
//# sourceMappingURL=chunk-AYXP3MTM.js.map
|
|
@@ -12,7 +12,7 @@ import {
|
|
|
12
12
|
parseYamlValue,
|
|
13
13
|
toCamelCaseDeep,
|
|
14
14
|
toSnakeCaseDeep
|
|
15
|
-
} from "./chunk-
|
|
15
|
+
} from "./chunk-HL3MTSIW.js";
|
|
16
16
|
|
|
17
17
|
// ../../packages/core/dist/index.js
|
|
18
18
|
import { readFileSync } from "node:fs";
|
|
@@ -2269,4 +2269,4 @@ export {
|
|
|
2269
2269
|
TranscriptProvider,
|
|
2270
2270
|
createAgentKernel
|
|
2271
2271
|
};
|
|
2272
|
-
//# sourceMappingURL=chunk-
|
|
2272
|
+
//# sourceMappingURL=chunk-D5HCHBPT.js.map
|
|
@@ -8286,7 +8286,7 @@ var _a20;
|
|
|
8286
8286
|
_a20 = symbol20;
|
|
8287
8287
|
var defaultDownload2 = createDownload();
|
|
8288
8288
|
|
|
8289
|
-
// ../../packages/core/dist/chunk-
|
|
8289
|
+
// ../../packages/core/dist/chunk-CUVG5O5P.js
|
|
8290
8290
|
import path46 from "node:path";
|
|
8291
8291
|
import { pathToFileURL as pathToFileURL2 } from "node:url";
|
|
8292
8292
|
import { existsSync as existsSync6 } from "node:fs";
|
|
@@ -12976,7 +12976,7 @@ var openrouter = createOpenRouter({
|
|
|
12976
12976
|
// strict for OpenRouter API
|
|
12977
12977
|
});
|
|
12978
12978
|
|
|
12979
|
-
// ../../packages/core/dist/chunk-
|
|
12979
|
+
// ../../packages/core/dist/chunk-CUVG5O5P.js
|
|
12980
12980
|
import { spawn } from "node:child_process";
|
|
12981
12981
|
import { randomUUID } from "node:crypto";
|
|
12982
12982
|
import { createWriteStream } from "node:fs";
|
|
@@ -14480,7 +14480,7 @@ var RequestError = class _RequestError extends Error {
|
|
|
14480
14480
|
}
|
|
14481
14481
|
};
|
|
14482
14482
|
|
|
14483
|
-
// ../../packages/core/dist/chunk-
|
|
14483
|
+
// ../../packages/core/dist/chunk-CUVG5O5P.js
|
|
14484
14484
|
import { exec as execCallback } from "node:child_process";
|
|
14485
14485
|
import { readdirSync, statSync } from "node:fs";
|
|
14486
14486
|
import { readFile as readFile22, readdir as readdir2, stat as stat2 } from "node:fs/promises";
|
|
@@ -30268,7 +30268,7 @@ async function loadTestSuite(evalFilePath, repoRoot, options) {
|
|
|
30268
30268
|
return { tests: await loadTestsFromAgentSkills(evalFilePath) };
|
|
30269
30269
|
}
|
|
30270
30270
|
if (format === "typescript") {
|
|
30271
|
-
const { loadTsEvalSuite: loadTsEvalSuite2 } = await import("./ts-eval-loader-
|
|
30271
|
+
const { loadTsEvalSuite: loadTsEvalSuite2 } = await import("./ts-eval-loader-SYQYQPMC-FKU23VX3.js");
|
|
30272
30272
|
return loadTsEvalSuite2(evalFilePath, resolveToAbsolutePath(repoRoot), options);
|
|
30273
30273
|
}
|
|
30274
30274
|
const { tests, parsed, suiteWorkspacePath } = await loadTestsFromYaml(
|
|
@@ -30303,7 +30303,7 @@ async function loadTests(evalFilePath, repoRoot, options) {
|
|
|
30303
30303
|
return loadTestsFromAgentSkills(evalFilePath);
|
|
30304
30304
|
}
|
|
30305
30305
|
if (format === "typescript") {
|
|
30306
|
-
const { loadTsEvalSuite: loadTsEvalSuite2 } = await import("./ts-eval-loader-
|
|
30306
|
+
const { loadTsEvalSuite: loadTsEvalSuite2 } = await import("./ts-eval-loader-SYQYQPMC-FKU23VX3.js");
|
|
30307
30307
|
const suite = await loadTsEvalSuite2(evalFilePath, resolveToAbsolutePath(repoRoot), options);
|
|
30308
30308
|
return suite.tests;
|
|
30309
30309
|
}
|
|
@@ -31676,7 +31676,9 @@ async function runEvaluation(options) {
|
|
|
31676
31676
|
completedAt: Date.now(),
|
|
31677
31677
|
error: result.error,
|
|
31678
31678
|
score: result.score,
|
|
31679
|
-
executionStatus: result.executionStatus
|
|
31679
|
+
executionStatus: result.executionStatus,
|
|
31680
|
+
durationMs: result.durationMs,
|
|
31681
|
+
evalRunDurationMs: result.evalRun?.durationMs
|
|
31680
31682
|
});
|
|
31681
31683
|
}
|
|
31682
31684
|
if (onResult) {
|
|
@@ -31990,7 +31992,8 @@ async function runBatchEvaluation(options) {
|
|
|
31990
31992
|
completedAt: Date.now(),
|
|
31991
31993
|
error: error instanceof Error ? error.message : String(error),
|
|
31992
31994
|
score: errorResult.score,
|
|
31993
|
-
executionStatus: errorResult.executionStatus
|
|
31995
|
+
executionStatus: errorResult.executionStatus,
|
|
31996
|
+
evalRunDurationMs: errorResult.evalRun?.durationMs
|
|
31994
31997
|
});
|
|
31995
31998
|
}
|
|
31996
31999
|
continue;
|
|
@@ -32008,7 +32011,9 @@ async function runBatchEvaluation(options) {
|
|
|
32008
32011
|
completedAt: Date.now(),
|
|
32009
32012
|
error: result.error,
|
|
32010
32013
|
score: result.score,
|
|
32011
|
-
executionStatus: result.executionStatus
|
|
32014
|
+
executionStatus: result.executionStatus,
|
|
32015
|
+
durationMs: result.durationMs,
|
|
32016
|
+
evalRunDurationMs: result.evalRun?.durationMs
|
|
32012
32017
|
});
|
|
32013
32018
|
}
|
|
32014
32019
|
}
|
|
@@ -34135,4 +34140,4 @@ export {
|
|
|
34135
34140
|
loadTsEvalFile,
|
|
34136
34141
|
loadTsEvalSuite
|
|
34137
34142
|
};
|
|
34138
|
-
//# sourceMappingURL=chunk-
|
|
34143
|
+
//# sourceMappingURL=chunk-HL3MTSIW.js.map
|