@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.d.cts
CHANGED
|
@@ -251,10 +251,8 @@ interface TokenUsage {
|
|
|
251
251
|
interface TraceSummary {
|
|
252
252
|
/** Total number of events in trace */
|
|
253
253
|
readonly eventCount: number;
|
|
254
|
-
/** Unique tool names, sorted alphabetically */
|
|
255
|
-
readonly toolNames: readonly string[];
|
|
256
254
|
/** Map of tool name to call count */
|
|
257
|
-
readonly
|
|
255
|
+
readonly toolCalls: Readonly<Record<string, number>>;
|
|
258
256
|
/** Number of error events */
|
|
259
257
|
readonly errorCount: number;
|
|
260
258
|
/** Per-tool duration arrays in milliseconds (optional) */
|
package/dist/index.d.ts
CHANGED
|
@@ -251,10 +251,8 @@ interface TokenUsage {
|
|
|
251
251
|
interface TraceSummary {
|
|
252
252
|
/** Total number of events in trace */
|
|
253
253
|
readonly eventCount: number;
|
|
254
|
-
/** Unique tool names, sorted alphabetically */
|
|
255
|
-
readonly toolNames: readonly string[];
|
|
256
254
|
/** Map of tool name to call count */
|
|
257
|
-
readonly
|
|
255
|
+
readonly toolCalls: Readonly<Record<string, number>>;
|
|
258
256
|
/** Number of error events */
|
|
259
257
|
readonly errorCount: number;
|
|
260
258
|
/** Per-tool duration arrays in milliseconds (optional) */
|
package/dist/index.js
CHANGED
|
@@ -83,12 +83,10 @@ function computeTraceSummary(messages) {
|
|
|
83
83
|
}
|
|
84
84
|
}
|
|
85
85
|
}
|
|
86
|
-
const toolNames = Object.keys(toolCallCounts).sort();
|
|
87
86
|
return {
|
|
88
87
|
trace: {
|
|
89
88
|
eventCount: totalToolCalls,
|
|
90
|
-
|
|
91
|
-
toolCallsByName: toolCallCounts,
|
|
89
|
+
toolCalls: toolCallCounts,
|
|
92
90
|
errorCount: 0,
|
|
93
91
|
llmCallCount,
|
|
94
92
|
...hasAnyDuration ? { toolDurations } : {}
|
|
@@ -112,7 +110,7 @@ var DEFAULT_EXPLORATION_TOOLS = [
|
|
|
112
110
|
function explorationRatio(summary, explorationTools = DEFAULT_EXPLORATION_TOOLS) {
|
|
113
111
|
if (summary.eventCount === 0) return void 0;
|
|
114
112
|
const explorationCalls = explorationTools.reduce(
|
|
115
|
-
(sum, tool2) => sum + (summary.
|
|
113
|
+
(sum, tool2) => sum + (summary.toolCalls[tool2] ?? 0),
|
|
116
114
|
0
|
|
117
115
|
);
|
|
118
116
|
return explorationCalls / summary.eventCount;
|
|
@@ -12780,11 +12778,9 @@ var ToolTrajectoryEvaluator = class {
|
|
|
12780
12778
|
for (const call of toolCalls) {
|
|
12781
12779
|
toolCallsByName[call.name] = (toolCallsByName[call.name] ?? 0) + 1;
|
|
12782
12780
|
}
|
|
12783
|
-
const toolNames = Object.keys(toolCallsByName).sort();
|
|
12784
12781
|
return {
|
|
12785
12782
|
eventCount: toolCalls.length,
|
|
12786
|
-
|
|
12787
|
-
toolCallsByName,
|
|
12783
|
+
toolCalls: toolCallsByName,
|
|
12788
12784
|
errorCount: 0
|
|
12789
12785
|
};
|
|
12790
12786
|
}
|
|
@@ -12802,7 +12798,7 @@ var ToolTrajectoryEvaluator = class {
|
|
|
12802
12798
|
const assertions = [];
|
|
12803
12799
|
for (const toolName of toolNames) {
|
|
12804
12800
|
const required = minimums[toolName];
|
|
12805
|
-
const actual = summary.
|
|
12801
|
+
const actual = summary.toolCalls[toolName] ?? 0;
|
|
12806
12802
|
if (actual >= required) {
|
|
12807
12803
|
assertions.push({
|
|
12808
12804
|
text: `${toolName}: called ${actual} times (required >=${required})`,
|
|
@@ -15483,7 +15479,7 @@ async function runBatchEvaluation(options) {
|
|
|
15483
15479
|
const providerResponse = batchResponse[i];
|
|
15484
15480
|
const output = providerResponse.output;
|
|
15485
15481
|
const hasExecutionMetrics = providerResponse.tokenUsage !== void 0 || providerResponse.costUsd !== void 0 || providerResponse.durationMs !== void 0;
|
|
15486
|
-
const computed = output ? computeTraceSummary(output) : hasExecutionMetrics ? { trace: { eventCount: 0,
|
|
15482
|
+
const computed = output ? computeTraceSummary(output) : hasExecutionMetrics ? { trace: { eventCount: 0, toolCalls: {}, errorCount: 0 } } : void 0;
|
|
15487
15483
|
const merged = computed ? mergeExecutionMetrics(computed, {
|
|
15488
15484
|
tokenUsage: providerResponse.tokenUsage,
|
|
15489
15485
|
costUsd: providerResponse.costUsd,
|
|
@@ -15880,7 +15876,7 @@ async function runEvalCase(options) {
|
|
|
15880
15876
|
}
|
|
15881
15877
|
const output = providerResponse.output;
|
|
15882
15878
|
const hasExecutionMetrics = providerResponse.tokenUsage !== void 0 || providerResponse.costUsd !== void 0 || providerResponse.durationMs !== void 0;
|
|
15883
|
-
const computed = output ? computeTraceSummary(output) : hasExecutionMetrics ? { trace: { eventCount: 0,
|
|
15879
|
+
const computed = output ? computeTraceSummary(output) : hasExecutionMetrics ? { trace: { eventCount: 0, toolCalls: {}, errorCount: 0 } } : void 0;
|
|
15884
15880
|
const merged = computed ? mergeExecutionMetrics(computed, {
|
|
15885
15881
|
tokenUsage: providerResponse.tokenUsage,
|
|
15886
15882
|
costUsd: providerResponse.costUsd,
|
|
@@ -17260,7 +17256,10 @@ var OtelTraceExporter = class {
|
|
|
17260
17256
|
if (result.trace) {
|
|
17261
17257
|
const t = result.trace;
|
|
17262
17258
|
rootSpan.setAttribute("agentv.trace.event_count", t.eventCount);
|
|
17263
|
-
rootSpan.setAttribute(
|
|
17259
|
+
rootSpan.setAttribute(
|
|
17260
|
+
"agentv.trace.tool_names",
|
|
17261
|
+
Object.keys(t.toolCalls).sort().join(",")
|
|
17262
|
+
);
|
|
17264
17263
|
if (t.llmCallCount != null)
|
|
17265
17264
|
rootSpan.setAttribute("agentv.trace.llm_call_count", t.llmCallCount);
|
|
17266
17265
|
}
|