agentv 3.7.0 → 3.8.0
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-7YS6YNJZ.js → chunk-ASYRKFAI.js} +125 -13
- package/dist/chunk-ASYRKFAI.js.map +1 -0
- package/dist/{chunk-XGG64VIY.js → chunk-F4UDJ7LG.js} +549 -545
- package/dist/chunk-F4UDJ7LG.js.map +1 -0
- package/dist/{chunk-TR6H437M.js → chunk-YZRGQ6ZS.js} +13 -9
- package/dist/chunk-YZRGQ6ZS.js.map +1 -0
- package/dist/cli.js +3 -3
- package/dist/{dist-VP6AXX6B.js → dist-4AQUJJAP.js} +2 -2
- package/dist/index.js +3 -3
- package/dist/{interactive-F6XECJ33.js → interactive-OPQGDF77.js} +3 -3
- package/package.json +1 -1
- package/dist/chunk-7YS6YNJZ.js.map +0 -1
- package/dist/chunk-TR6H437M.js.map +0 -1
- package/dist/chunk-XGG64VIY.js.map +0 -1
- /package/dist/{dist-VP6AXX6B.js.map → dist-4AQUJJAP.js.map} +0 -0
- /package/dist/{interactive-F6XECJ33.js.map → interactive-OPQGDF77.js.map} +0 -0
|
@@ -16,7 +16,7 @@ import {
|
|
|
16
16
|
validateEvalFile,
|
|
17
17
|
validateFileReferences,
|
|
18
18
|
validateTargetsFile
|
|
19
|
-
} from "./chunk-
|
|
19
|
+
} from "./chunk-ASYRKFAI.js";
|
|
20
20
|
import {
|
|
21
21
|
createBuiltinRegistry,
|
|
22
22
|
createProvider,
|
|
@@ -34,7 +34,7 @@ import {
|
|
|
34
34
|
toSnakeCaseDeep as toSnakeCaseDeep2,
|
|
35
35
|
transpileEvalYamlFile,
|
|
36
36
|
trimBaselineResult
|
|
37
|
-
} from "./chunk-
|
|
37
|
+
} from "./chunk-F4UDJ7LG.js";
|
|
38
38
|
import {
|
|
39
39
|
__commonJS,
|
|
40
40
|
__esm,
|
|
@@ -4180,11 +4180,16 @@ var evalRunCommand = command({
|
|
|
4180
4180
|
type: optional(string),
|
|
4181
4181
|
long: "model",
|
|
4182
4182
|
description: 'Override model for the grader target (e.g., "openai:gpt-5-mini")'
|
|
4183
|
+
}),
|
|
4184
|
+
outputMessages: option({
|
|
4185
|
+
type: optional(string),
|
|
4186
|
+
long: "output-messages",
|
|
4187
|
+
description: 'Number of trailing messages to include in results output (default: 1, or "all")'
|
|
4183
4188
|
})
|
|
4184
4189
|
},
|
|
4185
4190
|
handler: async (args) => {
|
|
4186
4191
|
if (args.evalPaths.length === 0 && process.stdin.isTTY) {
|
|
4187
|
-
const { launchInteractiveWizard } = await import("./interactive-
|
|
4192
|
+
const { launchInteractiveWizard } = await import("./interactive-OPQGDF77.js");
|
|
4188
4193
|
await launchInteractiveWizard();
|
|
4189
4194
|
return;
|
|
4190
4195
|
}
|
|
@@ -4220,7 +4225,8 @@ var evalRunCommand = command({
|
|
|
4220
4225
|
benchmarkJson: args.benchmarkJson,
|
|
4221
4226
|
artifacts: args.artifacts,
|
|
4222
4227
|
graderTarget: args.graderTarget,
|
|
4223
|
-
model: args.model
|
|
4228
|
+
model: args.model,
|
|
4229
|
+
outputMessages: args.outputMessages
|
|
4224
4230
|
};
|
|
4225
4231
|
await runEvalCommand({ testFiles: resolvedPaths, rawOptions });
|
|
4226
4232
|
}
|
|
@@ -4706,10 +4712,9 @@ function exportResults(sourceFile, content, outputDir) {
|
|
|
4706
4712
|
const outputsDir = path8.join(outputDir, "outputs");
|
|
4707
4713
|
mkdirSync2(outputsDir, { recursive: true });
|
|
4708
4714
|
for (const result of patched) {
|
|
4709
|
-
|
|
4710
|
-
if (outputText) {
|
|
4715
|
+
if (result.output && result.output.length > 0) {
|
|
4711
4716
|
const id = safeTestId(result);
|
|
4712
|
-
writeFileSync3(path8.join(outputsDir, `${id}.txt`),
|
|
4717
|
+
writeFileSync3(path8.join(outputsDir, `${id}.txt`), JSON.stringify(result.output, null, 2));
|
|
4713
4718
|
}
|
|
4714
4719
|
}
|
|
4715
4720
|
}
|
|
@@ -5021,7 +5026,6 @@ function toTraceSummary(raw) {
|
|
|
5021
5026
|
return toCamelCaseDeep(raw.trace);
|
|
5022
5027
|
}
|
|
5023
5028
|
function extractCandidate(raw) {
|
|
5024
|
-
if (raw.output_text !== void 0) return raw.output_text;
|
|
5025
5029
|
if (raw.output !== void 0)
|
|
5026
5030
|
return typeof raw.output === "string" ? raw.output : JSON.stringify(raw.output);
|
|
5027
5031
|
return "";
|
|
@@ -6276,4 +6280,4 @@ export {
|
|
|
6276
6280
|
preprocessArgv,
|
|
6277
6281
|
runCli
|
|
6278
6282
|
};
|
|
6279
|
-
//# sourceMappingURL=chunk-
|
|
6283
|
+
//# sourceMappingURL=chunk-YZRGQ6ZS.js.map
|