agentv 4.15.4 → 4.15.5-next.1
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-CK47W2B5.js → chunk-KNKQS6OW.js} +3 -3
- package/dist/{chunk-TT36K77B.js → chunk-UM6OWWWZ.js} +37 -30
- package/dist/chunk-UM6OWWWZ.js.map +1 -0
- package/dist/cli.js +2 -2
- package/dist/index.js +2 -2
- package/dist/{interactive-RXDJ5OXY.js → interactive-5PDSLRVC.js} +2 -2
- package/package.json +1 -1
- package/dist/chunk-TT36K77B.js.map +0 -1
- /package/dist/{chunk-CK47W2B5.js.map → chunk-KNKQS6OW.js.map} +0 -0
- /package/dist/{interactive-RXDJ5OXY.js.map → interactive-5PDSLRVC.js.map} +0 -0
|
@@ -42,7 +42,7 @@ import {
|
|
|
42
42
|
validateFileReferences,
|
|
43
43
|
validateTargetsFile,
|
|
44
44
|
writeArtifactsFromResults
|
|
45
|
-
} from "./chunk-
|
|
45
|
+
} from "./chunk-UM6OWWWZ.js";
|
|
46
46
|
import {
|
|
47
47
|
DEFAULT_CATEGORY,
|
|
48
48
|
DEFAULT_THRESHOLD,
|
|
@@ -3916,7 +3916,7 @@ var evalRunCommand = command({
|
|
|
3916
3916
|
},
|
|
3917
3917
|
handler: async (args) => {
|
|
3918
3918
|
if (args.evalPaths.length === 0 && process.stdin.isTTY) {
|
|
3919
|
-
const { launchInteractiveWizard } = await import("./interactive-
|
|
3919
|
+
const { launchInteractiveWizard } = await import("./interactive-5PDSLRVC.js");
|
|
3920
3920
|
await launchInteractiveWizard();
|
|
3921
3921
|
return;
|
|
3922
3922
|
}
|
|
@@ -10066,4 +10066,4 @@ export {
|
|
|
10066
10066
|
preprocessArgv,
|
|
10067
10067
|
runCli
|
|
10068
10068
|
};
|
|
10069
|
-
//# sourceMappingURL=chunk-
|
|
10069
|
+
//# sourceMappingURL=chunk-KNKQS6OW.js.map
|
|
@@ -47,7 +47,7 @@ import {
|
|
|
47
47
|
// package.json
|
|
48
48
|
var package_default = {
|
|
49
49
|
name: "agentv",
|
|
50
|
-
version: "4.15.
|
|
50
|
+
version: "4.15.5-next.1",
|
|
51
51
|
description: "CLI entry point for AgentV",
|
|
52
52
|
type: "module",
|
|
53
53
|
repository: {
|
|
@@ -2654,7 +2654,7 @@ function useColors() {
|
|
|
2654
2654
|
function formatVerdict(score, verdict) {
|
|
2655
2655
|
if (verdict === void 0) return "";
|
|
2656
2656
|
const colors2 = useColors();
|
|
2657
|
-
const scoreStr = score !== void 0 ?
|
|
2657
|
+
const scoreStr = score !== void 0 ? `${Math.round(score * 100)}%` : "";
|
|
2658
2658
|
const verdictLabel = verdict === "ERROR" ? "ERROR" : `${scoreStr} ${verdict}`;
|
|
2659
2659
|
if (!colors2) return ` | ${verdictLabel}`;
|
|
2660
2660
|
const color = verdict === "PASS" ? ANSI_GREEN : verdict === "FAIL" ? ANSI_RED2 : ANSI_YELLOW2;
|
|
@@ -2666,7 +2666,6 @@ var ProgressDisplay = class {
|
|
|
2666
2666
|
completedTests = 0;
|
|
2667
2667
|
logPaths = [];
|
|
2668
2668
|
logPathSet = /* @__PURE__ */ new Set();
|
|
2669
|
-
hasPrintedLogHeader = false;
|
|
2670
2669
|
started = false;
|
|
2671
2670
|
finished = false;
|
|
2672
2671
|
verbose;
|
|
@@ -2702,19 +2701,23 @@ var ProgressDisplay = class {
|
|
|
2702
2701
|
console.log(`${countPrefix} \u{1F504} ${progress.testId}${targetSuffix}`);
|
|
2703
2702
|
}
|
|
2704
2703
|
break;
|
|
2705
|
-
case "completed":
|
|
2704
|
+
case "completed": {
|
|
2705
|
+
const icon = progress.verdict === "FAIL" ? "\u26A0\uFE0F" : progress.verdict === "ERROR" ? "\u274C" : "\u2705";
|
|
2706
2706
|
console.log(
|
|
2707
|
-
`${countPrefix}
|
|
2707
|
+
`${countPrefix} ${icon} ${progress.testId}${targetSuffix}${formatVerdict(progress.score, progress.verdict)}`
|
|
2708
2708
|
);
|
|
2709
2709
|
break;
|
|
2710
|
-
|
|
2710
|
+
}
|
|
2711
|
+
case "failed": {
|
|
2712
|
+
const failIcon = progress.verdict === "ERROR" ? "\u274C" : "\u26A0\uFE0F";
|
|
2711
2713
|
console.log(
|
|
2712
|
-
`${countPrefix}
|
|
2714
|
+
`${countPrefix} ${failIcon} ${progress.testId}${targetSuffix}${formatVerdict(progress.score, progress.verdict)}${progress.error ? `: ${progress.error}` : ""}`
|
|
2713
2715
|
);
|
|
2714
2716
|
break;
|
|
2717
|
+
}
|
|
2715
2718
|
}
|
|
2716
2719
|
}
|
|
2717
|
-
addLogPaths(paths
|
|
2720
|
+
addLogPaths(paths) {
|
|
2718
2721
|
const newPaths = [];
|
|
2719
2722
|
for (const path19 of paths) {
|
|
2720
2723
|
if (this.logPathSet.has(path19)) {
|
|
@@ -2727,16 +2730,9 @@ var ProgressDisplay = class {
|
|
|
2727
2730
|
return;
|
|
2728
2731
|
}
|
|
2729
2732
|
this.logPaths.push(...newPaths);
|
|
2730
|
-
|
|
2731
|
-
console.log(
|
|
2732
|
-
|
|
2733
|
-
console.log(`${label} logs:`);
|
|
2734
|
-
this.hasPrintedLogHeader = true;
|
|
2735
|
-
}
|
|
2736
|
-
const startIndex = this.logPaths.length - newPaths.length;
|
|
2737
|
-
newPaths.forEach((path19, offset) => {
|
|
2738
|
-
console.log(`${startIndex + offset + 1}. ${path19}`);
|
|
2739
|
-
});
|
|
2733
|
+
for (const p of newPaths) {
|
|
2734
|
+
console.log(`Provider log: ${p}`);
|
|
2735
|
+
}
|
|
2740
2736
|
}
|
|
2741
2737
|
finish() {
|
|
2742
2738
|
this.finished = true;
|
|
@@ -4957,12 +4953,18 @@ function createProgressReporter(maxWorkers, options) {
|
|
|
4957
4953
|
setTotal: (total) => display.setTotalTests(total),
|
|
4958
4954
|
update: (workerId, progress) => display.updateWorker({ ...progress, workerId }),
|
|
4959
4955
|
finish: () => display.finish(),
|
|
4960
|
-
addLogPaths: (paths
|
|
4956
|
+
addLogPaths: (paths) => display.addLogPaths(paths)
|
|
4961
4957
|
};
|
|
4962
4958
|
}
|
|
4963
4959
|
function makeTestCaseKey(testFilePath, testId) {
|
|
4964
4960
|
return `${path17.resolve(testFilePath)}::${testId}`;
|
|
4965
4961
|
}
|
|
4962
|
+
function resolveTargetLabel(requestedName, resolvedName) {
|
|
4963
|
+
if (resolvedName !== requestedName) {
|
|
4964
|
+
return `${requestedName} \u2192 ${resolvedName}`;
|
|
4965
|
+
}
|
|
4966
|
+
return requestedName;
|
|
4967
|
+
}
|
|
4966
4968
|
function createDisplayIdTracker() {
|
|
4967
4969
|
const map = /* @__PURE__ */ new Map();
|
|
4968
4970
|
let nextId = 1;
|
|
@@ -5071,7 +5073,7 @@ async function prepareFileMetadata(params) {
|
|
|
5071
5073
|
});
|
|
5072
5074
|
selections = multiSelections.map((sel) => ({
|
|
5073
5075
|
selection: sel,
|
|
5074
|
-
inlineTargetLabel: sel.targetName
|
|
5076
|
+
inlineTargetLabel: resolveTargetLabel(sel.targetName, sel.resolvedTarget.name)
|
|
5075
5077
|
}));
|
|
5076
5078
|
} else {
|
|
5077
5079
|
const selection = await selectTarget({
|
|
@@ -5089,7 +5091,10 @@ async function prepareFileMetadata(params) {
|
|
|
5089
5091
|
selections = [
|
|
5090
5092
|
{
|
|
5091
5093
|
selection,
|
|
5092
|
-
inlineTargetLabel:
|
|
5094
|
+
inlineTargetLabel: resolveTargetLabel(
|
|
5095
|
+
selection.targetName,
|
|
5096
|
+
selection.resolvedTarget.name
|
|
5097
|
+
)
|
|
5093
5098
|
}
|
|
5094
5099
|
];
|
|
5095
5100
|
}
|
|
@@ -5139,7 +5144,7 @@ async function runSingleEvalFile(params) {
|
|
|
5139
5144
|
const providerLabel = options.dryRun ? `${resolvedTargetSelection.resolvedTarget.kind} (dry-run)` : resolvedTargetSelection.resolvedTarget.kind;
|
|
5140
5145
|
const targetMessage = options.verbose ? `Using target (${resolvedTargetSelection.targetSource}): ${resolvedTargetSelection.targetName} ${buildTargetLabelSuffix(providerLabel, resolvedTargetSelection.resolvedTarget)} via ${resolvedTargetSelection.targetsFilePath}` : `Using target: ${inlineTargetLabel}`;
|
|
5141
5146
|
if (!progressReporter.isInteractive || options.verbose) {
|
|
5142
|
-
console.log(targetMessage);
|
|
5147
|
+
console.log(`${targetMessage}`);
|
|
5143
5148
|
}
|
|
5144
5149
|
const agentTimeoutMs = options.agentTimeoutSeconds != null ? Math.max(0, options.agentTimeoutSeconds) * 1e3 : void 0;
|
|
5145
5150
|
const workerPreference = workersOverride ?? options.workers;
|
|
@@ -5505,7 +5510,7 @@ async function runEvalCommand(input) {
|
|
|
5505
5510
|
return;
|
|
5506
5511
|
}
|
|
5507
5512
|
seenCodexLogPaths.add(entry.filePath);
|
|
5508
|
-
progressReporter.addLogPaths([entry.filePath]
|
|
5513
|
+
progressReporter.addLogPaths([entry.filePath]);
|
|
5509
5514
|
});
|
|
5510
5515
|
const seenPiLogPaths = /* @__PURE__ */ new Set();
|
|
5511
5516
|
const unsubscribePiLogs = subscribeToPiLogEntries((entry) => {
|
|
@@ -5513,7 +5518,7 @@ async function runEvalCommand(input) {
|
|
|
5513
5518
|
return;
|
|
5514
5519
|
}
|
|
5515
5520
|
seenPiLogPaths.add(entry.filePath);
|
|
5516
|
-
progressReporter.addLogPaths([entry.filePath]
|
|
5521
|
+
progressReporter.addLogPaths([entry.filePath]);
|
|
5517
5522
|
});
|
|
5518
5523
|
const seenCopilotLogPaths = /* @__PURE__ */ new Set();
|
|
5519
5524
|
const unsubscribeCopilotSdkLogs = subscribeToCopilotSdkLogEntries((entry) => {
|
|
@@ -5521,14 +5526,14 @@ async function runEvalCommand(input) {
|
|
|
5521
5526
|
return;
|
|
5522
5527
|
}
|
|
5523
5528
|
seenCopilotLogPaths.add(entry.filePath);
|
|
5524
|
-
progressReporter.addLogPaths([entry.filePath]
|
|
5529
|
+
progressReporter.addLogPaths([entry.filePath]);
|
|
5525
5530
|
});
|
|
5526
5531
|
const unsubscribeCopilotCliLogs = subscribeToCopilotCliLogEntries((entry) => {
|
|
5527
5532
|
if (!entry.filePath || seenCopilotLogPaths.has(entry.filePath)) {
|
|
5528
5533
|
return;
|
|
5529
5534
|
}
|
|
5530
5535
|
seenCopilotLogPaths.add(entry.filePath);
|
|
5531
|
-
progressReporter.addLogPaths([entry.filePath]
|
|
5536
|
+
progressReporter.addLogPaths([entry.filePath]);
|
|
5532
5537
|
});
|
|
5533
5538
|
for (const [testFilePath, meta] of fileMetadata.entries()) {
|
|
5534
5539
|
for (const { selection, inlineTargetLabel } of meta.selections) {
|
|
@@ -5625,9 +5630,11 @@ async function runEvalCommand(input) {
|
|
|
5625
5630
|
return result.results;
|
|
5626
5631
|
} catch (fileError) {
|
|
5627
5632
|
const message = fileError instanceof Error ? fileError.message : String(fileError);
|
|
5628
|
-
console.error(
|
|
5629
|
-
|
|
5630
|
-
|
|
5633
|
+
console.error(
|
|
5634
|
+
`
|
|
5635
|
+
[ERROR] \u26A0 Eval file failed: ${path17.basename(testFilePath)} \u2014 ${message}
|
|
5636
|
+
`
|
|
5637
|
+
);
|
|
5631
5638
|
const errorResults = applicableTestCases.map((testCase) => ({
|
|
5632
5639
|
timestamp: (/* @__PURE__ */ new Date()).toISOString(),
|
|
5633
5640
|
testId: testCase.id,
|
|
@@ -5876,4 +5883,4 @@ export {
|
|
|
5876
5883
|
getCategories,
|
|
5877
5884
|
filterByCategory
|
|
5878
5885
|
};
|
|
5879
|
-
//# sourceMappingURL=chunk-
|
|
5886
|
+
//# sourceMappingURL=chunk-UM6OWWWZ.js.map
|