agentv 3.8.0 → 3.9.0
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/README.md +1 -1
- package/dist/{chunk-ASYRKFAI.js → chunk-GC5P5HHZ.js} +6 -37
- package/dist/chunk-GC5P5HHZ.js.map +1 -0
- package/dist/{chunk-YZRGQ6ZS.js → chunk-Q2YWV4QM.js} +12 -16
- package/dist/chunk-Q2YWV4QM.js.map +1 -0
- package/dist/{chunk-F4UDJ7LG.js → chunk-TXDPYXHY.js} +356 -616
- package/dist/chunk-TXDPYXHY.js.map +1 -0
- package/dist/cli.js +3 -3
- package/dist/{dist-4AQUJJAP.js → dist-PIOSPBKX.js} +2 -4
- package/dist/index.js +3 -3
- package/dist/{interactive-OPQGDF77.js → interactive-3VTDK5NX.js} +3 -3
- package/package.json +1 -1
- package/dist/chunk-ASYRKFAI.js.map +0 -1
- package/dist/chunk-F4UDJ7LG.js.map +0 -1
- package/dist/chunk-YZRGQ6ZS.js.map +0 -1
- /package/dist/{dist-4AQUJJAP.js.map → dist-PIOSPBKX.js.map} +0 -0
- /package/dist/{interactive-OPQGDF77.js.map → interactive-3VTDK5NX.js.map} +0 -0
package/README.md
CHANGED
|
@@ -164,7 +164,7 @@ For large-scale evaluations, AgentV supports JSONL (JSON Lines) format as an alt
|
|
|
164
164
|
Optional sidecar YAML metadata file (`dataset.eval.yaml` alongside `dataset.jsonl`):
|
|
165
165
|
```yaml
|
|
166
166
|
description: Math evaluation dataset
|
|
167
|
-
|
|
167
|
+
name: math-tests
|
|
168
168
|
execution:
|
|
169
169
|
target: azure-llm
|
|
170
170
|
assertions:
|
|
@@ -27,12 +27,12 @@ import {
|
|
|
27
27
|
subscribeToCopilotCliLogEntries,
|
|
28
28
|
subscribeToCopilotSdkLogEntries,
|
|
29
29
|
subscribeToPiLogEntries
|
|
30
|
-
} from "./chunk-
|
|
30
|
+
} from "./chunk-TXDPYXHY.js";
|
|
31
31
|
|
|
32
32
|
// package.json
|
|
33
33
|
var package_default = {
|
|
34
34
|
name: "agentv",
|
|
35
|
-
version: "3.
|
|
35
|
+
version: "3.9.0",
|
|
36
36
|
description: "CLI entry point for AgentV",
|
|
37
37
|
type: "module",
|
|
38
38
|
repository: {
|
|
@@ -1524,7 +1524,7 @@ var JunitWriter = class _JunitWriter {
|
|
|
1524
1524
|
this.closed = true;
|
|
1525
1525
|
const grouped = /* @__PURE__ */ new Map();
|
|
1526
1526
|
for (const result of this.results) {
|
|
1527
|
-
const suite = result.
|
|
1527
|
+
const suite = result.eval_set ?? "default";
|
|
1528
1528
|
const existing = grouped.get(suite);
|
|
1529
1529
|
if (existing) {
|
|
1530
1530
|
existing.push(result);
|
|
@@ -3220,31 +3220,6 @@ async function validateConfigFile(filePath) {
|
|
|
3220
3220
|
return { valid: false, filePath, fileType: "config", errors };
|
|
3221
3221
|
}
|
|
3222
3222
|
const config = parsed;
|
|
3223
|
-
const guidelinePatterns = config.guideline_patterns;
|
|
3224
|
-
if (guidelinePatterns !== void 0) {
|
|
3225
|
-
if (!Array.isArray(guidelinePatterns)) {
|
|
3226
|
-
errors.push({
|
|
3227
|
-
severity: "error",
|
|
3228
|
-
filePath,
|
|
3229
|
-
location: "guideline_patterns",
|
|
3230
|
-
message: "Field 'guideline_patterns' must be an array"
|
|
3231
|
-
});
|
|
3232
|
-
} else if (!guidelinePatterns.every((p) => typeof p === "string")) {
|
|
3233
|
-
errors.push({
|
|
3234
|
-
severity: "error",
|
|
3235
|
-
filePath,
|
|
3236
|
-
location: "guideline_patterns",
|
|
3237
|
-
message: "All entries in 'guideline_patterns' must be strings"
|
|
3238
|
-
});
|
|
3239
|
-
} else if (guidelinePatterns.length === 0) {
|
|
3240
|
-
errors.push({
|
|
3241
|
-
severity: "warning",
|
|
3242
|
-
filePath,
|
|
3243
|
-
location: "guideline_patterns",
|
|
3244
|
-
message: "Field 'guideline_patterns' is empty. Consider removing it or adding patterns."
|
|
3245
|
-
});
|
|
3246
|
-
}
|
|
3247
|
-
}
|
|
3248
3223
|
const evalPatterns = config.eval_patterns;
|
|
3249
3224
|
if (evalPatterns !== void 0) {
|
|
3250
3225
|
if (!Array.isArray(evalPatterns)) {
|
|
@@ -3281,13 +3256,7 @@ async function validateConfigFile(filePath) {
|
|
|
3281
3256
|
});
|
|
3282
3257
|
}
|
|
3283
3258
|
}
|
|
3284
|
-
const allowedFields = /* @__PURE__ */ new Set([
|
|
3285
|
-
"$schema",
|
|
3286
|
-
"guideline_patterns",
|
|
3287
|
-
"eval_patterns",
|
|
3288
|
-
"required_version",
|
|
3289
|
-
"execution"
|
|
3290
|
-
]);
|
|
3259
|
+
const allowedFields = /* @__PURE__ */ new Set(["$schema", "eval_patterns", "required_version", "execution"]);
|
|
3291
3260
|
const unexpectedFields = Object.keys(config).filter((key) => !allowedFields.has(key));
|
|
3292
3261
|
if (unexpectedFields.length > 0) {
|
|
3293
3262
|
errors.push({
|
|
@@ -4156,7 +4125,7 @@ async function runEvalCommand(input) {
|
|
|
4156
4125
|
const useFileExport = !!(options.otelFile || options.traceFile);
|
|
4157
4126
|
if (options.exportOtel || useFileExport) {
|
|
4158
4127
|
try {
|
|
4159
|
-
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-
|
|
4128
|
+
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-PIOSPBKX.js");
|
|
4160
4129
|
let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
|
|
4161
4130
|
let headers = {};
|
|
4162
4131
|
if (options.otelBackend) {
|
|
@@ -4498,4 +4467,4 @@ export {
|
|
|
4498
4467
|
selectTarget,
|
|
4499
4468
|
runEvalCommand
|
|
4500
4469
|
};
|
|
4501
|
-
//# sourceMappingURL=chunk-
|
|
4470
|
+
//# sourceMappingURL=chunk-GC5P5HHZ.js.map
|