agentv 4.37.0-next.1 → 4.38.0
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/{artifact-writer-GFNKYREE.js → artifact-writer-MK5X5MSO.js} +4 -4
- package/dist/{chunk-P4LSNFZR.js → chunk-3G4BK6Z5.js} +21 -20
- package/dist/{chunk-P4LSNFZR.js.map → chunk-3G4BK6Z5.js.map} +1 -1
- package/dist/{chunk-N6E5XFOM.js → chunk-DKUAETXE.js} +3 -3
- package/dist/{chunk-M7AMFWBZ.js → chunk-EKMMIULD.js} +42 -37
- package/dist/chunk-EKMMIULD.js.map +1 -0
- package/dist/{chunk-OYI35QFW.js → chunk-NLTIK3LV.js} +32 -250
- package/dist/chunk-NLTIK3LV.js.map +1 -0
- package/dist/{chunk-RL4S2FBZ.js → chunk-VBHHZQS6.js} +902 -488
- package/dist/chunk-VBHHZQS6.js.map +1 -0
- package/dist/cli.js +5 -5
- package/dist/dashboard/assets/index-BpnllKET.css +1 -0
- package/dist/dashboard/assets/index-Cm9SUopp.js +118 -0
- package/dist/dashboard/assets/{index-BDRYJsGF.js → index-SIl6NbIJ.js} +1 -1
- package/dist/dashboard/index.html +2 -2
- package/dist/{dist-OY3JSP6Z.js → dist-HVLBDG5F.js} +17 -13
- package/dist/index.js +5 -5
- package/dist/{interactive-CQELHITQ.js → interactive-QFAAM4SI.js} +5 -5
- package/dist/skills/agentv-eval-writer/SKILL.md +28 -36
- package/dist/skills/agentv-eval-writer/references/eval-schema.json +57 -210
- package/dist/{ts-eval-loader-RBTB2HG2-H5TRXZLO.js → ts-eval-loader-TJT6BGFF-DI7XNSO4.js} +2 -2
- package/package.json +1 -1
- package/dist/chunk-M7AMFWBZ.js.map +0 -1
- package/dist/chunk-OYI35QFW.js.map +0 -1
- package/dist/chunk-RL4S2FBZ.js.map +0 -1
- package/dist/dashboard/assets/index-9tV-u4HJ.css +0 -1
- package/dist/dashboard/assets/index-DuESU7zZ.js +0 -118
- /package/dist/{artifact-writer-GFNKYREE.js.map → artifact-writer-MK5X5MSO.js.map} +0 -0
- /package/dist/{chunk-N6E5XFOM.js.map → chunk-DKUAETXE.js.map} +0 -0
- /package/dist/{dist-OY3JSP6Z.js.map → dist-HVLBDG5F.js.map} +0 -0
- /package/dist/{interactive-CQELHITQ.js.map → interactive-QFAAM4SI.js.map} +0 -0
- /package/dist/{ts-eval-loader-RBTB2HG2-H5TRXZLO.js.map → ts-eval-loader-TJT6BGFF-DI7XNSO4.js.map} +0 -0
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { createRequire } from 'node:module'; const require = createRequire(import.meta.url);
|
|
2
2
|
import {
|
|
3
3
|
traceToTranscriptJsonLines
|
|
4
|
-
} from "./chunk-
|
|
4
|
+
} from "./chunk-NLTIK3LV.js";
|
|
5
5
|
import {
|
|
6
6
|
DEFAULT_THRESHOLD,
|
|
7
7
|
buildTraceEnvelopeFromEvaluationResult,
|
|
@@ -9,7 +9,7 @@ import {
|
|
|
9
9
|
extractLastAssistantContent,
|
|
10
10
|
parseYamlValue,
|
|
11
11
|
toTraceEnvelopeWire
|
|
12
|
-
} from "./chunk-
|
|
12
|
+
} from "./chunk-VBHHZQS6.js";
|
|
13
13
|
|
|
14
14
|
// src/commands/eval/artifact-writer.ts
|
|
15
15
|
import { mkdir as mkdir2, readFile as readFile2, writeFile as writeFile2 } from "node:fs/promises";
|
|
@@ -1359,4 +1359,4 @@ export {
|
|
|
1359
1359
|
writePerTestArtifacts,
|
|
1360
1360
|
writeArtifactsFromResults
|
|
1361
1361
|
};
|
|
1362
|
-
//# sourceMappingURL=chunk-
|
|
1362
|
+
//# sourceMappingURL=chunk-DKUAETXE.js.map
|
|
@@ -16,7 +16,7 @@ import {
|
|
|
16
16
|
toSnakeCaseDeep as toSnakeCaseDeep2,
|
|
17
17
|
writeArtifactsFromResults,
|
|
18
18
|
writeInitialBenchmarkArtifact
|
|
19
|
-
} from "./chunk-
|
|
19
|
+
} from "./chunk-DKUAETXE.js";
|
|
20
20
|
import {
|
|
21
21
|
RunBudgetTracker,
|
|
22
22
|
buildWipBranchName,
|
|
@@ -24,8 +24,6 @@ import {
|
|
|
24
24
|
deriveCategory,
|
|
25
25
|
directPushResults,
|
|
26
26
|
directorySizeBytes,
|
|
27
|
-
getProject,
|
|
28
|
-
getProjectForPath,
|
|
29
27
|
getResultsRepoSyncStatus,
|
|
30
28
|
listGitRuns,
|
|
31
29
|
loadTsConfig,
|
|
@@ -36,7 +34,7 @@ import {
|
|
|
36
34
|
setupWipWorktree,
|
|
37
35
|
syncResultsRepoForProject,
|
|
38
36
|
traceFromTranscriptJsonLines
|
|
39
|
-
} from "./chunk-
|
|
37
|
+
} from "./chunk-NLTIK3LV.js";
|
|
40
38
|
import {
|
|
41
39
|
CLI_PLACEHOLDERS,
|
|
42
40
|
COMMON_TARGET_SETTINGS,
|
|
@@ -52,6 +50,8 @@ import {
|
|
|
52
50
|
findDeprecatedCamelCaseTargetWarnings,
|
|
53
51
|
findGitRoot,
|
|
54
52
|
getAgentvConfigDir,
|
|
53
|
+
getProject,
|
|
54
|
+
getProjectForPath,
|
|
55
55
|
interpolateEnv,
|
|
56
56
|
isGraderKind,
|
|
57
57
|
listTargetNames,
|
|
@@ -74,7 +74,7 @@ import {
|
|
|
74
74
|
subscribeToPiLogEntries,
|
|
75
75
|
toCamelCaseDeep,
|
|
76
76
|
toSnakeCaseDeep
|
|
77
|
-
} from "./chunk-
|
|
77
|
+
} from "./chunk-VBHHZQS6.js";
|
|
78
78
|
|
|
79
79
|
// src/commands/eval/shared.ts
|
|
80
80
|
import { constants } from "node:fs";
|
|
@@ -190,7 +190,7 @@ async function findRepoRoot(start) {
|
|
|
190
190
|
// package.json
|
|
191
191
|
var package_default = {
|
|
192
192
|
name: "agentv",
|
|
193
|
-
version: "4.
|
|
193
|
+
version: "4.38.0",
|
|
194
194
|
description: "CLI entry point for AgentV",
|
|
195
195
|
type: "module",
|
|
196
196
|
repository: {
|
|
@@ -2900,40 +2900,45 @@ function validateWorkspaceRepoConfig(workspace, filePath, errors) {
|
|
|
2900
2900
|
if (Array.isArray(repos)) {
|
|
2901
2901
|
for (const repo of repos) {
|
|
2902
2902
|
if (!isObject(repo)) continue;
|
|
2903
|
-
|
|
2904
|
-
const checkout = repo.checkout;
|
|
2905
|
-
const clone = repo.clone;
|
|
2906
|
-
if (!isObject(source) && !isObject(docker)) {
|
|
2903
|
+
if ("source" in repo) {
|
|
2907
2904
|
errors.push({
|
|
2908
2905
|
severity: "error",
|
|
2909
2906
|
filePath,
|
|
2910
2907
|
location: `workspace.repos[path=${repo.path ?? "(none)"}]`,
|
|
2911
|
-
message: "repos[].source
|
|
2908
|
+
message: "workspace.repos[].source has been removed. Use workspace.repos[].repo."
|
|
2912
2909
|
});
|
|
2913
2910
|
}
|
|
2914
|
-
if (
|
|
2915
|
-
|
|
2916
|
-
|
|
2917
|
-
|
|
2918
|
-
|
|
2919
|
-
|
|
2920
|
-
|
|
2921
|
-
location: `workspace.repos[path=${repo.path ?? "(none)"}]`,
|
|
2922
|
-
message: "checkout.resolve has no effect for a local source. Use source.type to choose where the repo comes from; keep checkout.ref, checkout.base_commit, or checkout.ancestor only when pinning a local source."
|
|
2923
|
-
});
|
|
2924
|
-
}
|
|
2911
|
+
if ("checkout" in repo) {
|
|
2912
|
+
errors.push({
|
|
2913
|
+
severity: "error",
|
|
2914
|
+
filePath,
|
|
2915
|
+
location: `workspace.repos[path=${repo.path ?? "(none)"}]`,
|
|
2916
|
+
message: "workspace.repos[].checkout has been removed. Use top-level commit, base_commit, and ancestor."
|
|
2917
|
+
});
|
|
2925
2918
|
}
|
|
2926
|
-
if (
|
|
2927
|
-
|
|
2928
|
-
|
|
2929
|
-
|
|
2930
|
-
|
|
2931
|
-
|
|
2932
|
-
|
|
2933
|
-
|
|
2934
|
-
|
|
2935
|
-
|
|
2936
|
-
|
|
2919
|
+
if ("clone" in repo) {
|
|
2920
|
+
errors.push({
|
|
2921
|
+
severity: "error",
|
|
2922
|
+
filePath,
|
|
2923
|
+
location: `workspace.repos[path=${repo.path ?? "(none)"}]`,
|
|
2924
|
+
message: "workspace.repos[].clone has been removed. Use top-level sparse if needed."
|
|
2925
|
+
});
|
|
2926
|
+
}
|
|
2927
|
+
if (!repo.repo && !isObject(docker)) {
|
|
2928
|
+
errors.push({
|
|
2929
|
+
severity: "error",
|
|
2930
|
+
filePath,
|
|
2931
|
+
location: `workspace.repos[path=${repo.path ?? "(none)"}]`,
|
|
2932
|
+
message: "repos[].repo is required for non-Docker workspaces. Repo-less entries are only valid when workspace.docker is configured."
|
|
2933
|
+
});
|
|
2934
|
+
}
|
|
2935
|
+
if (typeof repo.commit === "string" && typeof repo.base_commit === "string" && repo.commit !== repo.base_commit) {
|
|
2936
|
+
errors.push({
|
|
2937
|
+
severity: "error",
|
|
2938
|
+
filePath,
|
|
2939
|
+
location: `workspace.repos[path=${repo.path ?? "(none)"}]`,
|
|
2940
|
+
message: "repos[].commit and repos[].base_commit must match when both are set."
|
|
2941
|
+
});
|
|
2937
2942
|
}
|
|
2938
2943
|
}
|
|
2939
2944
|
}
|
|
@@ -5556,7 +5561,7 @@ async function runEvalCommand(input) {
|
|
|
5556
5561
|
const useFileExport = !!options.otelFile;
|
|
5557
5562
|
if (options.exportOtel || useFileExport) {
|
|
5558
5563
|
try {
|
|
5559
|
-
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-
|
|
5564
|
+
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-HVLBDG5F.js");
|
|
5560
5565
|
let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
|
|
5561
5566
|
let headers = {};
|
|
5562
5567
|
if (options.otelBackend) {
|
|
@@ -5752,7 +5757,7 @@ async function runEvalCommand(input) {
|
|
|
5752
5757
|
const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
|
|
5753
5758
|
let transcriptProviderFactory;
|
|
5754
5759
|
if (options.transcript) {
|
|
5755
|
-
const { TranscriptProvider } = await import("./dist-
|
|
5760
|
+
const { TranscriptProvider } = await import("./dist-HVLBDG5F.js");
|
|
5756
5761
|
const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
|
|
5757
5762
|
const totalTests = [...fileMetadata.values()].reduce(
|
|
5758
5763
|
(sum, meta) => sum + meta.testCases.length,
|
|
@@ -5968,7 +5973,7 @@ async function runEvalCommand(input) {
|
|
|
5968
5973
|
);
|
|
5969
5974
|
const taskBundleTargets = buildTaskBundleTargetSelections(activeTestFiles, fileMetadata);
|
|
5970
5975
|
if (isResumeAppend) {
|
|
5971
|
-
const { writePerTestArtifacts } = await import("./artifact-writer-
|
|
5976
|
+
const { writePerTestArtifacts } = await import("./artifact-writer-MK5X5MSO.js");
|
|
5972
5977
|
await writePerTestArtifacts(allResults, runDir, {
|
|
5973
5978
|
experiment: normalizeExperimentName(options.experiment),
|
|
5974
5979
|
cwd,
|
|
@@ -6203,4 +6208,4 @@ export {
|
|
|
6203
6208
|
getCategories,
|
|
6204
6209
|
filterByCategory
|
|
6205
6210
|
};
|
|
6206
|
-
//# sourceMappingURL=chunk-
|
|
6211
|
+
//# sourceMappingURL=chunk-EKMMIULD.js.map
|