agentv 4.15.9 → 4.16.0
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-3GULESPF.js → chunk-IRU2UOWN.js} +321 -241
- package/dist/chunk-IRU2UOWN.js.map +1 -0
- package/dist/{chunk-BMPIWUEE.js → chunk-NENUAMR5.js} +52 -24
- package/dist/chunk-NENUAMR5.js.map +1 -0
- package/dist/{chunk-NWHF3BEO.js → chunk-TA2KF32T.js} +8 -8
- package/dist/{chunk-NWHF3BEO.js.map → chunk-TA2KF32T.js.map} +1 -1
- package/dist/cli.js +3 -3
- package/dist/{dist-D45MEPD7.js → dist-U5EXNMON.js} +6 -2
- package/dist/index.js +3 -3
- package/dist/{interactive-AF3EC5PG.js → interactive-IWGFP7T2.js} +7 -7
- package/dist/{interactive-AF3EC5PG.js.map → interactive-IWGFP7T2.js.map} +1 -1
- package/package.json +1 -1
- package/dist/chunk-3GULESPF.js.map +0 -1
- package/dist/chunk-BMPIWUEE.js.map +0 -1
- /package/dist/{dist-D45MEPD7.js.map → dist-U5EXNMON.js.map} +0 -0
|
@@ -42,12 +42,12 @@ import {
|
|
|
42
42
|
syncResultsRepo,
|
|
43
43
|
toCamelCaseDeep,
|
|
44
44
|
toTranscriptJsonLines
|
|
45
|
-
} from "./chunk-
|
|
45
|
+
} from "./chunk-IRU2UOWN.js";
|
|
46
46
|
|
|
47
47
|
// package.json
|
|
48
48
|
var package_default = {
|
|
49
49
|
name: "agentv",
|
|
50
|
-
version: "4.
|
|
50
|
+
version: "4.16.0",
|
|
51
51
|
description: "CLI entry point for AgentV",
|
|
52
52
|
type: "module",
|
|
53
53
|
repository: {
|
|
@@ -4043,8 +4043,7 @@ var CODEX_SETTINGS = /* @__PURE__ */ new Set([
|
|
|
4043
4043
|
"log_directory",
|
|
4044
4044
|
"log_format",
|
|
4045
4045
|
"log_output_format",
|
|
4046
|
-
"system_prompt"
|
|
4047
|
-
"workspace_template"
|
|
4046
|
+
"system_prompt"
|
|
4048
4047
|
]);
|
|
4049
4048
|
var COPILOT_SDK_SETTINGS = /* @__PURE__ */ new Set([
|
|
4050
4049
|
...COMMON_SETTINGS,
|
|
@@ -4057,7 +4056,6 @@ var COPILOT_SDK_SETTINGS = /* @__PURE__ */ new Set([
|
|
|
4057
4056
|
"log_dir",
|
|
4058
4057
|
"log_format",
|
|
4059
4058
|
"system_prompt",
|
|
4060
|
-
"workspace_template",
|
|
4061
4059
|
"byok"
|
|
4062
4060
|
]);
|
|
4063
4061
|
var COPILOT_CLI_SETTINGS = /* @__PURE__ */ new Set([
|
|
@@ -4072,13 +4070,11 @@ var COPILOT_CLI_SETTINGS = /* @__PURE__ */ new Set([
|
|
|
4072
4070
|
"timeout_seconds",
|
|
4073
4071
|
"log_dir",
|
|
4074
4072
|
"log_format",
|
|
4075
|
-
"system_prompt"
|
|
4076
|
-
"workspace_template"
|
|
4073
|
+
"system_prompt"
|
|
4077
4074
|
]);
|
|
4078
4075
|
var VSCODE_SETTINGS = /* @__PURE__ */ new Set([
|
|
4079
4076
|
...COMMON_SETTINGS,
|
|
4080
4077
|
"executable",
|
|
4081
|
-
"workspace_template",
|
|
4082
4078
|
"wait",
|
|
4083
4079
|
"dry_run",
|
|
4084
4080
|
"subagent_root",
|
|
@@ -4092,6 +4088,9 @@ var MOCK_SETTINGS = /* @__PURE__ */ new Set([
|
|
|
4092
4088
|
]);
|
|
4093
4089
|
var CLAUDE_SETTINGS = /* @__PURE__ */ new Set([
|
|
4094
4090
|
...COMMON_SETTINGS,
|
|
4091
|
+
"executable",
|
|
4092
|
+
"command",
|
|
4093
|
+
"binary",
|
|
4095
4094
|
"model",
|
|
4096
4095
|
"cwd",
|
|
4097
4096
|
"timeout_seconds",
|
|
@@ -4100,10 +4099,10 @@ var CLAUDE_SETTINGS = /* @__PURE__ */ new Set([
|
|
|
4100
4099
|
"log_format",
|
|
4101
4100
|
"log_output_format",
|
|
4102
4101
|
"system_prompt",
|
|
4103
|
-
"workspace_template",
|
|
4104
4102
|
"max_turns",
|
|
4105
4103
|
"max_budget_usd"
|
|
4106
4104
|
]);
|
|
4105
|
+
var CC_MIRROR_SETTINGS = /* @__PURE__ */ new Set([...CLAUDE_SETTINGS, "variant"]);
|
|
4107
4106
|
function getKnownSettings(provider) {
|
|
4108
4107
|
const normalizedProvider = provider.toLowerCase();
|
|
4109
4108
|
switch (normalizedProvider) {
|
|
@@ -4129,6 +4128,8 @@ function getKnownSettings(provider) {
|
|
|
4129
4128
|
case "copilot":
|
|
4130
4129
|
case "copilot-cli":
|
|
4131
4130
|
return COPILOT_CLI_SETTINGS;
|
|
4131
|
+
case "cc-mirror":
|
|
4132
|
+
return CC_MIRROR_SETTINGS;
|
|
4132
4133
|
case "claude":
|
|
4133
4134
|
case "claude-code":
|
|
4134
4135
|
case "claude-cli":
|
|
@@ -4146,7 +4147,6 @@ function getKnownSettings(provider) {
|
|
|
4146
4147
|
}
|
|
4147
4148
|
}
|
|
4148
4149
|
function validateUnknownSettings(target, provider, absolutePath, location, errors) {
|
|
4149
|
-
const removedTargetFields = /* @__PURE__ */ new Set(["workspace_template", "workspaceTemplate"]);
|
|
4150
4150
|
const knownSettings = getKnownSettings(provider);
|
|
4151
4151
|
if (!knownSettings) {
|
|
4152
4152
|
return;
|
|
@@ -4160,13 +4160,14 @@ function validateUnknownSettings(target, provider, absolutePath, location, error
|
|
|
4160
4160
|
"$schema",
|
|
4161
4161
|
"targets"
|
|
4162
4162
|
]);
|
|
4163
|
+
const removedFields = /* @__PURE__ */ new Set(["workspace_template", "workspaceTemplate"]);
|
|
4163
4164
|
for (const key of Object.keys(target)) {
|
|
4164
|
-
if (
|
|
4165
|
+
if (removedFields.has(key)) {
|
|
4165
4166
|
errors.push({
|
|
4166
|
-
severity: "
|
|
4167
|
+
severity: "warning",
|
|
4167
4168
|
filePath: absolutePath,
|
|
4168
4169
|
location: `${location}.${key}`,
|
|
4169
|
-
message: "
|
|
4170
|
+
message: "workspace_template has been removed from targets. Use eval-level workspace.template instead."
|
|
4170
4171
|
});
|
|
4171
4172
|
continue;
|
|
4172
4173
|
}
|
|
@@ -4904,8 +4905,17 @@ async function selectMultipleTargets(options) {
|
|
|
4904
4905
|
dryRunDelayMin,
|
|
4905
4906
|
dryRunDelayMax,
|
|
4906
4907
|
env,
|
|
4907
|
-
targetNames
|
|
4908
|
+
targetNames,
|
|
4909
|
+
targetRefs
|
|
4908
4910
|
} = options;
|
|
4911
|
+
const hooksMap = /* @__PURE__ */ new Map();
|
|
4912
|
+
if (targetRefs) {
|
|
4913
|
+
for (const ref of targetRefs) {
|
|
4914
|
+
if (ref.hooks) {
|
|
4915
|
+
hooksMap.set(ref.name, ref.hooks);
|
|
4916
|
+
}
|
|
4917
|
+
}
|
|
4918
|
+
}
|
|
4909
4919
|
const targetsFilePath = await discoverTargetsFile({
|
|
4910
4920
|
explicitPath: explicitTargetsPath,
|
|
4911
4921
|
testFilePath,
|
|
@@ -4938,10 +4948,19 @@ Errors in ${targetsFilePath}:`);
|
|
|
4938
4948
|
}
|
|
4939
4949
|
throw new Error(`Targets file validation failed with ${errors.length} error(s)`);
|
|
4940
4950
|
}
|
|
4941
|
-
const
|
|
4951
|
+
const fileDefinitions = await readTargetDefinitions(targetsFilePath);
|
|
4952
|
+
const definitions = [...fileDefinitions];
|
|
4953
|
+
if (targetRefs) {
|
|
4954
|
+
for (const ref of targetRefs) {
|
|
4955
|
+
if (ref.use_target && !fileDefinitions.some((d) => d.name === ref.name)) {
|
|
4956
|
+
definitions.push({ name: ref.name, use_target: ref.use_target });
|
|
4957
|
+
}
|
|
4958
|
+
}
|
|
4959
|
+
}
|
|
4942
4960
|
const results = [];
|
|
4943
4961
|
for (const name of targetNames) {
|
|
4944
4962
|
const targetDefinition = resolveUseTarget(name, definitions, env, targetsFilePath);
|
|
4963
|
+
const hooks = hooksMap.get(name);
|
|
4945
4964
|
if (dryRun) {
|
|
4946
4965
|
const mockTarget = {
|
|
4947
4966
|
kind: "mock",
|
|
@@ -4959,7 +4978,8 @@ Errors in ${targetsFilePath}:`);
|
|
|
4959
4978
|
resolvedTarget: mockTarget,
|
|
4960
4979
|
targetName: name,
|
|
4961
4980
|
targetSource: "cli",
|
|
4962
|
-
targetsFilePath
|
|
4981
|
+
targetsFilePath,
|
|
4982
|
+
...hooks && { targetHooks: hooks }
|
|
4963
4983
|
});
|
|
4964
4984
|
} else {
|
|
4965
4985
|
try {
|
|
@@ -4971,7 +4991,8 @@ Errors in ${targetsFilePath}:`);
|
|
|
4971
4991
|
resolvedTarget,
|
|
4972
4992
|
targetName: name,
|
|
4973
4993
|
targetSource: "cli",
|
|
4974
|
-
targetsFilePath
|
|
4994
|
+
targetsFilePath,
|
|
4995
|
+
...hooks && { targetHooks: hooks }
|
|
4975
4996
|
});
|
|
4976
4997
|
} catch (error) {
|
|
4977
4998
|
const message = error instanceof Error ? error.message : String(error);
|
|
@@ -5287,6 +5308,7 @@ async function prepareFileMetadata(params) {
|
|
|
5287
5308
|
} else {
|
|
5288
5309
|
const cliTargets = options.cliTargets;
|
|
5289
5310
|
const suiteTargets2 = suite.targets;
|
|
5311
|
+
const suiteTargetRefs = suite.targetRefs;
|
|
5290
5312
|
let targetNames;
|
|
5291
5313
|
if (cliTargets.length > 0) {
|
|
5292
5314
|
targetNames = cliTargets;
|
|
@@ -5306,7 +5328,8 @@ async function prepareFileMetadata(params) {
|
|
|
5306
5328
|
dryRunDelayMin: options.dryRunDelayMin,
|
|
5307
5329
|
dryRunDelayMax: options.dryRunDelayMax,
|
|
5308
5330
|
env: process.env,
|
|
5309
|
-
targetNames
|
|
5331
|
+
targetNames,
|
|
5332
|
+
targetRefs: suiteTargetRefs
|
|
5310
5333
|
});
|
|
5311
5334
|
selections = multiSelections.map((sel) => ({
|
|
5312
5335
|
selection: sel,
|
|
@@ -5325,12 +5348,16 @@ async function prepareFileMetadata(params) {
|
|
|
5325
5348
|
dryRunDelayMax: options.dryRunDelayMax,
|
|
5326
5349
|
env: process.env
|
|
5327
5350
|
});
|
|
5351
|
+
const singleTargetHooks = suiteTargetRefs?.find(
|
|
5352
|
+
(ref) => ref.name === selection.targetName
|
|
5353
|
+
)?.hooks;
|
|
5354
|
+
const augmentedSelection = singleTargetHooks ? { ...selection, targetHooks: singleTargetHooks } : selection;
|
|
5328
5355
|
selections = [
|
|
5329
5356
|
{
|
|
5330
|
-
selection,
|
|
5357
|
+
selection: augmentedSelection,
|
|
5331
5358
|
inlineTargetLabel: resolveTargetLabel(
|
|
5332
|
-
|
|
5333
|
-
|
|
5359
|
+
augmentedSelection.targetName,
|
|
5360
|
+
augmentedSelection.resolvedTarget.name
|
|
5334
5361
|
)
|
|
5335
5362
|
}
|
|
5336
5363
|
];
|
|
@@ -5442,6 +5469,7 @@ async function runSingleEvalFile(params) {
|
|
|
5442
5469
|
graderTarget: options.graderTarget,
|
|
5443
5470
|
model: options.model,
|
|
5444
5471
|
threshold: options.threshold,
|
|
5472
|
+
targetHooks: resolvedTargetSelection.targetHooks,
|
|
5445
5473
|
providerFactory,
|
|
5446
5474
|
streamCallbacks: streamingObserver?.getStreamCallbacks(),
|
|
5447
5475
|
onResult: async (result) => {
|
|
@@ -5597,7 +5625,7 @@ async function runEvalCommand(input) {
|
|
|
5597
5625
|
const useFileExport = !!options.otelFile;
|
|
5598
5626
|
if (options.exportOtel || useFileExport) {
|
|
5599
5627
|
try {
|
|
5600
|
-
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-
|
|
5628
|
+
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-U5EXNMON.js");
|
|
5601
5629
|
let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
|
|
5602
5630
|
let headers = {};
|
|
5603
5631
|
if (options.otelBackend) {
|
|
@@ -5793,7 +5821,7 @@ async function runEvalCommand(input) {
|
|
|
5793
5821
|
const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
|
|
5794
5822
|
let transcriptProviderFactory;
|
|
5795
5823
|
if (options.transcript) {
|
|
5796
|
-
const { TranscriptProvider } = await import("./dist-
|
|
5824
|
+
const { TranscriptProvider } = await import("./dist-U5EXNMON.js");
|
|
5797
5825
|
const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
|
|
5798
5826
|
const totalTests = [...fileMetadata.values()].reduce(
|
|
5799
5827
|
(sum, meta) => sum + meta.testCases.length,
|
|
@@ -6122,4 +6150,4 @@ export {
|
|
|
6122
6150
|
getCategories,
|
|
6123
6151
|
filterByCategory
|
|
6124
6152
|
};
|
|
6125
|
-
//# sourceMappingURL=chunk-
|
|
6153
|
+
//# sourceMappingURL=chunk-NENUAMR5.js.map
|