agentv 4.12.5 → 4.12.6
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-TQMECG6F.js → chunk-BOFV2GTT.js} +8 -5
- package/dist/chunk-BOFV2GTT.js.map +1 -0
- package/dist/{chunk-MY5UKOPW.js → chunk-QYXM32WE.js} +2 -2
- package/dist/{chunk-MY5UKOPW.js.map → chunk-QYXM32WE.js.map} +1 -1
- package/dist/cli.js +2 -2
- package/dist/index.js +2 -2
- package/dist/{interactive-33ACYB2B.js → interactive-MXG34JBR.js} +2 -2
- package/dist/studio/assets/{index-Al4o1T1V.js → index-CFlLDFI8.js} +9 -9
- package/dist/studio/assets/{index-Uu2TYsOb.js → index-D74g07Nj.js} +1 -1
- package/dist/studio/index.html +1 -1
- package/package.json +1 -1
- package/dist/chunk-TQMECG6F.js.map +0 -1
- /package/dist/{interactive-33ACYB2B.js.map → interactive-MXG34JBR.js.map} +0 -0
|
@@ -42,7 +42,7 @@ import {
|
|
|
42
42
|
validateFileReferences,
|
|
43
43
|
validateTargetsFile,
|
|
44
44
|
writeArtifactsFromResults
|
|
45
|
-
} from "./chunk-
|
|
45
|
+
} from "./chunk-QYXM32WE.js";
|
|
46
46
|
import {
|
|
47
47
|
DEFAULT_CATEGORY,
|
|
48
48
|
DEFAULT_THRESHOLD,
|
|
@@ -3912,7 +3912,7 @@ var evalRunCommand = command({
|
|
|
3912
3912
|
},
|
|
3913
3913
|
handler: async (args) => {
|
|
3914
3914
|
if (args.evalPaths.length === 0 && process.stdin.isTTY) {
|
|
3915
|
-
const { launchInteractiveWizard } = await import("./interactive-
|
|
3915
|
+
const { launchInteractiveWizard } = await import("./interactive-MXG34JBR.js");
|
|
3916
3916
|
await launchInteractiveWizard();
|
|
3917
3917
|
return;
|
|
3918
3918
|
}
|
|
@@ -7817,17 +7817,20 @@ function stripHeavyFields(results) {
|
|
|
7817
7817
|
};
|
|
7818
7818
|
});
|
|
7819
7819
|
}
|
|
7820
|
-
async function handleRuns(c4, { searchDir }) {
|
|
7820
|
+
async function handleRuns(c4, { searchDir, agentvDir }) {
|
|
7821
7821
|
const { runs: metas } = await listMergedResultFiles(searchDir);
|
|
7822
|
+
const { threshold: passThreshold } = loadStudioConfig(agentvDir);
|
|
7822
7823
|
return c4.json({
|
|
7823
7824
|
runs: metas.map((m) => {
|
|
7824
7825
|
let target;
|
|
7825
7826
|
let experiment;
|
|
7827
|
+
let passRate = m.passRate;
|
|
7826
7828
|
try {
|
|
7827
7829
|
const records = loadLightweightResults(m.path);
|
|
7828
7830
|
if (records.length > 0) {
|
|
7829
7831
|
target = records[0].target;
|
|
7830
7832
|
experiment = records[0].experiment;
|
|
7833
|
+
passRate = records.filter((r) => r.score >= passThreshold).length / records.length;
|
|
7831
7834
|
}
|
|
7832
7835
|
} catch {
|
|
7833
7836
|
}
|
|
@@ -7837,7 +7840,7 @@ async function handleRuns(c4, { searchDir }) {
|
|
|
7837
7840
|
path: m.path,
|
|
7838
7841
|
timestamp: m.timestamp,
|
|
7839
7842
|
test_count: m.testCount,
|
|
7840
|
-
pass_rate:
|
|
7843
|
+
pass_rate: passRate,
|
|
7841
7844
|
avg_score: m.avgScore,
|
|
7842
7845
|
size_bytes: m.sizeBytes,
|
|
7843
7846
|
source: m.source,
|
|
@@ -9853,4 +9856,4 @@ export {
|
|
|
9853
9856
|
preprocessArgv,
|
|
9854
9857
|
runCli
|
|
9855
9858
|
};
|
|
9856
|
-
//# sourceMappingURL=chunk-
|
|
9859
|
+
//# sourceMappingURL=chunk-BOFV2GTT.js.map
|