@agentv/core 3.6.0 → 3.7.0
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/index.cjs +10 -11
- package/dist/index.cjs.map +1 -1
- package/dist/index.d.cts +1 -3
- package/dist/index.d.ts +1 -3
- package/dist/index.js +10 -11
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
package/dist/index.cjs
CHANGED
|
@@ -1786,12 +1786,10 @@ function computeTraceSummary(messages) {
|
|
|
1786
1786
|
}
|
|
1787
1787
|
}
|
|
1788
1788
|
}
|
|
1789
|
-
const toolNames = Object.keys(toolCallCounts).sort();
|
|
1790
1789
|
return {
|
|
1791
1790
|
trace: {
|
|
1792
1791
|
eventCount: totalToolCalls,
|
|
1793
|
-
|
|
1794
|
-
toolCallsByName: toolCallCounts,
|
|
1792
|
+
toolCalls: toolCallCounts,
|
|
1795
1793
|
errorCount: 0,
|
|
1796
1794
|
llmCallCount,
|
|
1797
1795
|
...hasAnyDuration ? { toolDurations } : {}
|
|
@@ -1815,7 +1813,7 @@ var DEFAULT_EXPLORATION_TOOLS = [
|
|
|
1815
1813
|
function explorationRatio(summary, explorationTools = DEFAULT_EXPLORATION_TOOLS) {
|
|
1816
1814
|
if (summary.eventCount === 0) return void 0;
|
|
1817
1815
|
const explorationCalls = explorationTools.reduce(
|
|
1818
|
-
(sum, tool2) => sum + (summary.
|
|
1816
|
+
(sum, tool2) => sum + (summary.toolCalls[tool2] ?? 0),
|
|
1819
1817
|
0
|
|
1820
1818
|
);
|
|
1821
1819
|
return explorationCalls / summary.eventCount;
|
|
@@ -15832,11 +15830,9 @@ var ToolTrajectoryEvaluator = class {
|
|
|
15832
15830
|
for (const call of toolCalls) {
|
|
15833
15831
|
toolCallsByName[call.name] = (toolCallsByName[call.name] ?? 0) + 1;
|
|
15834
15832
|
}
|
|
15835
|
-
const toolNames = Object.keys(toolCallsByName).sort();
|
|
15836
15833
|
return {
|
|
15837
15834
|
eventCount: toolCalls.length,
|
|
15838
|
-
|
|
15839
|
-
toolCallsByName,
|
|
15835
|
+
toolCalls: toolCallsByName,
|
|
15840
15836
|
errorCount: 0
|
|
15841
15837
|
};
|
|
15842
15838
|
}
|
|
@@ -15854,7 +15850,7 @@ var ToolTrajectoryEvaluator = class {
|
|
|
15854
15850
|
const assertions = [];
|
|
15855
15851
|
for (const toolName of toolNames) {
|
|
15856
15852
|
const required = minimums[toolName];
|
|
15857
|
-
const actual = summary.
|
|
15853
|
+
const actual = summary.toolCalls[toolName] ?? 0;
|
|
15858
15854
|
if (actual >= required) {
|
|
15859
15855
|
assertions.push({
|
|
15860
15856
|
text: `${toolName}: called ${actual} times (required >=${required})`,
|
|
@@ -18535,7 +18531,7 @@ async function runBatchEvaluation(options) {
|
|
|
18535
18531
|
const providerResponse = batchResponse[i];
|
|
18536
18532
|
const output = providerResponse.output;
|
|
18537
18533
|
const hasExecutionMetrics = providerResponse.tokenUsage !== void 0 || providerResponse.costUsd !== void 0 || providerResponse.durationMs !== void 0;
|
|
18538
|
-
const computed = output ? computeTraceSummary(output) : hasExecutionMetrics ? { trace: { eventCount: 0,
|
|
18534
|
+
const computed = output ? computeTraceSummary(output) : hasExecutionMetrics ? { trace: { eventCount: 0, toolCalls: {}, errorCount: 0 } } : void 0;
|
|
18539
18535
|
const merged = computed ? mergeExecutionMetrics(computed, {
|
|
18540
18536
|
tokenUsage: providerResponse.tokenUsage,
|
|
18541
18537
|
costUsd: providerResponse.costUsd,
|
|
@@ -18932,7 +18928,7 @@ async function runEvalCase(options) {
|
|
|
18932
18928
|
}
|
|
18933
18929
|
const output = providerResponse.output;
|
|
18934
18930
|
const hasExecutionMetrics = providerResponse.tokenUsage !== void 0 || providerResponse.costUsd !== void 0 || providerResponse.durationMs !== void 0;
|
|
18935
|
-
const computed = output ? computeTraceSummary(output) : hasExecutionMetrics ? { trace: { eventCount: 0,
|
|
18931
|
+
const computed = output ? computeTraceSummary(output) : hasExecutionMetrics ? { trace: { eventCount: 0, toolCalls: {}, errorCount: 0 } } : void 0;
|
|
18936
18932
|
const merged = computed ? mergeExecutionMetrics(computed, {
|
|
18937
18933
|
tokenUsage: providerResponse.tokenUsage,
|
|
18938
18934
|
costUsd: providerResponse.costUsd,
|
|
@@ -20312,7 +20308,10 @@ var OtelTraceExporter = class {
|
|
|
20312
20308
|
if (result.trace) {
|
|
20313
20309
|
const t = result.trace;
|
|
20314
20310
|
rootSpan.setAttribute("agentv.trace.event_count", t.eventCount);
|
|
20315
|
-
rootSpan.setAttribute(
|
|
20311
|
+
rootSpan.setAttribute(
|
|
20312
|
+
"agentv.trace.tool_names",
|
|
20313
|
+
Object.keys(t.toolCalls).sort().join(",")
|
|
20314
|
+
);
|
|
20316
20315
|
if (t.llmCallCount != null)
|
|
20317
20316
|
rootSpan.setAttribute("agentv.trace.llm_call_count", t.llmCallCount);
|
|
20318
20317
|
}
|