@ls-stack/agent-eval 0.58.5 → 0.59.0

package/dist/{app-DLNmRUqH.mjs → app-B3PEtWqH.mjs}

@@ -1,7 +1,7 @@
-import { Dt as getCaseRowCaseKey, ft as getEvalTitle, it as extractCacheEntries, kt as caseRowSchema, nt as createRunRequestSchema, rt as updateManualScoreRequestSchema } from "./runExecution-BMnJXWhN.mjs";
-import { o as stageManualInputFile } from "./cli-ClAkjTvo.mjs";
-import "./src-DfzidkYr.mjs";
-import { t as getRunnerInstance } from "./runner-BfHgVhGS.mjs";
+import { Dt as getCaseRowCaseKey, ft as getEvalTitle, it as extractCacheEntries, kt as caseRowSchema, nt as createRunRequestSchema, rt as updateManualScoreRequestSchema } from "./runExecution-C3XVZHRC.mjs";
+import { o as stageManualInputFile } from "./cli-Dkp2-rBm.mjs";
+import "./src-8dGXUULC.mjs";
+import { t as getRunnerInstance } from "./runner-C9J-1fkp.mjs";
 import { z } from "zod/v4";
 import { resultify } from "t-result";
 import { readFile } from "node:fs/promises";

package/dist/bin.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { t as runCli } from "./cli-ClAkjTvo.mjs";
+import { t as runCli } from "./cli-Dkp2-rBm.mjs";
 import { spawn } from "node:child_process";
 //#region src/bin.ts
 const moduleMocksFlag = "--experimental-test-module-mocks";

package/dist/caseChild.mjs

@@ -1,4 +1,4 @@
-import { Ct as resolveApiCallsConfig, Lt as runWithEvalRegistry, R as configureEvalRunLogs, Y as runInEvalRuntimeScope, _ as createBufferedCacheStore, a as isCaseChildParentMessage, d as loadEvalModule, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, r as runCase, v as createFsCacheStore, wt as resolveLlmCallsConfig, y as getCacheRetentionOptions } from "./runExecution-BMnJXWhN.mjs";
+import { Ct as resolveApiCallsConfig, Lt as runWithEvalRegistry, R as configureEvalRunLogs, Y as runInEvalRuntimeScope, _ as createBufferedCacheStore, a as isCaseChildParentMessage, d as loadEvalModule, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, r as runCase, v as createFsCacheStore, wt as resolveLlmCallsConfig, y as getCacheRetentionOptions } from "./runExecution-C3XVZHRC.mjs";
 //#region ../runner/src/caseChild.ts
 let fatalErrorReported = false;
 let disconnectExpected = false;

package/dist/{cli-ClAkjTvo.mjs → cli-Dkp2-rBm.mjs}

@@ -1,5 +1,5 @@
-import { Ct as resolveApiCallsConfig, Dt as getCaseRowCaseKey, Et as buildEvalKey, It as getEvalRegistry, St as runSummarySchema, c as resolveArtifactPath, f as resolveEvalDefaultConfig, ft as getEvalTitle, h as normalizeScoreDef, kt as caseRowSchema, m as buildDeclaredColumnDefs, mt as deriveScopedSummaryFromCases, o as stripTerminalControlCodes, p as loadConfig, pt as getEvalDisplayStatus, s as resolveTracePresentation, ut as applyDerivedCallAttributes, v as createFsCacheStore, vt as matchesTagsFilter, wt as resolveLlmCallsConfig, y as getCacheRetentionOptions } from "./runExecution-BMnJXWhN.mjs";
-import { C as validateCharts, S as parseEvalDiscovery, _ as runTouchesEval, a as validateTagsFilters, b as deriveEvalFreshness, c as getLatestRunInfos, d as nextShortIdFromSnapshots, f as persistCaseDetail, g as recomputePersistedCaseStatus, h as recomputeEvalStatusesInRuns, i as resolveEvalTags, l as loadPersistedRunSnapshot, m as persistRunState, n as getTargetEvalKeys, o as generateRunId, p as deleteTemporaryRuns, s as getLastRunStatuses, u as loadPersistedRunSnapshots, v as buildManualInputDescriptor, x as loadIsolatedEvalRegistry, y as parseManualInputValues } from "./runOrchestration-CvbTAoEb.mjs";
+import { Ct as resolveApiCallsConfig, Dt as getCaseRowCaseKey, Et as buildEvalKey, It as getEvalRegistry, St as runSummarySchema, c as resolveArtifactPath, f as resolveEvalDefaultConfig, ft as getEvalTitle, h as normalizeScoreDef, kt as caseRowSchema, m as buildDeclaredColumnDefs, mt as deriveScopedSummaryFromCases, o as stripTerminalControlCodes, p as loadConfig, pt as getEvalDisplayStatus, s as resolveTracePresentation, ut as applyDerivedCallAttributes, v as createFsCacheStore, vt as matchesTagsFilter, wt as resolveLlmCallsConfig, y as getCacheRetentionOptions } from "./runExecution-C3XVZHRC.mjs";
+import { C as validateCharts, S as parseEvalDiscovery, _ as runTouchesEval, a as validateTagsFilters, b as deriveEvalFreshness, c as getLatestRunInfos, d as nextShortIdFromSnapshots, f as persistCaseDetail, g as recomputePersistedCaseStatus, h as recomputeEvalStatusesInRuns, i as resolveEvalTags, l as loadPersistedRunSnapshot, m as persistRunState, n as getTargetEvalKeys, o as generateRunId, p as deleteTemporaryRuns, s as getLastRunStatuses, u as loadPersistedRunSnapshots, v as buildManualInputDescriptor, x as loadIsolatedEvalRegistry, y as parseManualInputValues } from "./runOrchestration-B5An-AEi.mjs";
 import { resultify } from "t-result";
 import { copyFile, mkdir, readFile, rm, writeFile } from "node:fs/promises";
 import { basename, dirname, extname, isAbsolute, join, relative, resolve, sep } from "node:path";
@@ -2199,8 +2199,8 @@ async function commandApp(args) {
 	const { serve } = await import("@hono/node-server");
 	const bundledWebDist = resolve(currentDir, "apps/web/dist");
 	if (existsSync(bundledWebDist)) process.env.AGENT_EVALS_WEB_DIST = bundledWebDist;
-	const appModule = await import("./app-DLNmRUqH.mjs");
-	const runnerModule = await import("./runner-DJJekv9f.mjs");
+	const appModule = await import("./app-B3PEtWqH.mjs");
+	const runnerModule = await import("./runner-BJXz_V_V.mjs");
 	if (!isHonoAppModule(appModule)) throw new Error("Server app module is invalid");
 	if (!isServerRunnerModule(runnerModule)) throw new Error("Server runner module is invalid");
 	await runnerModule.initRunner();

package/dist/index.d.mts

@@ -398,6 +398,18 @@ type EvalCase$1$1<TInput = unknown> = {
   input: TInput;
   tags?: string[];
 };
+/** Normalized view of one tool-call span and its common tool metadata. */
+type EvalToolCallSpan = {
+  /** Preferred tool name, using GenAI/Mastra identity metadata when present. */name: string; /** Original trace span display name. */
+  spanName: string; /** Original trace span kind. */
+  kind: string; /** Parsed tool-call arguments, or the raw value when parsing is not possible. */
+  arguments: unknown; /** Parsed tool-call result, or the raw value when parsing is not possible. */
+  result: unknown; /** Tool description from GenAI/Mastra metadata when present. */
+  description: string | undefined; /** Tool type from GenAI/Mastra metadata when present. */
+  toolType: string | undefined; /** Original span attributes. */
+  attributes: Record<string, unknown> | undefined; /** Original trace span for fields not normalized above. */
+  span: EvalTraceSpan$2;
+};
 /** Query helpers built from the flattened trace recorded for one eval case. */
 type EvalTraceTree = {
   /** Flat span list in creation order. */spans: EvalTraceSpan$2[]; /** Top-level spans whose `parentId` is `null`. */
@@ -406,10 +418,16 @@ type EvalTraceTree = {
   findSpans: (name: string) => EvalTraceSpan$2[]; /** Return whether any span name exactly matches `name`. */
   hasSpan: (name: string) => boolean; /** Return every span whose kind exactly matches `kind`. */
   findSpansByKind: (kind: string) => EvalTraceSpan$2[]; /** Return every span with `kind: 'tool'` or `kind: 'tool_call'`. */
-  findToolCallSpans: () => EvalTraceSpan$2[]; /** Return the names of every span with `kind: 'tool'` or `kind: 'tool_call'`. */
-  listToolCallSpanNames: () => string[]; /** Return whether a tool-call span has a name exactly matching `name`. */
-  hasToolCallSpan: (name: string) => boolean; /** Return whether a tool-call span name appears exactly `expectedCalls` times. */
-  hasNToolCallSpans: (toolName: string, expectedCalls: number) => boolean; /** Return span names in creation order, optionally filtered by kind. */
+  findToolCallSpans: () => EvalTraceSpan$2[];
+  /**
+   * Return tool-call names, preferring GenAI/Mastra tool identity attributes
+   * when available.
+   */
+  listToolCallSpanNames: () => string[]; /** Return whether a tool-call span name or tool identity matches `name`. */
+  hasToolCallSpan: (name: string) => boolean; /** Return normalized tool-call spans whose name or tool identity matches `name`. */
+  getToolCallSpans: (name: string) => EvalToolCallSpan[]; /** Return how many tool-call spans have a name or tool identity matching `toolName`. */
+  getToolCallSpanCount: (toolName: string) => number; /** Return whether a tool-call span name or tool identity appears exactly `expectedCalls` times. */
+  hasToolCallSpanCount: (toolName: string, expectedCalls: number) => boolean; /** Return span names in creation order, optionally filtered by kind. */
   listSpanNames: (kind?: string) => string[]; /** Return span names in depth-first tree order, optionally filtered by kind. */
   listSpanNamesDfs: (kind?: string) => string[]; /** Return all spans in depth-first tree order. */
   flattenDfs: () => EvalTraceSpan$2[];
@@ -3688,6 +3706,18 @@ type EvalCase$1<TInput = unknown> = {
   input: TInput;
   tags?: string[];
 };
+/** Normalized view of one tool-call span and its common tool metadata. */
+type EvalToolCallSpan$1 = {
+  /** Preferred tool name, using GenAI/Mastra identity metadata when present. */name: string; /** Original trace span display name. */
+  spanName: string; /** Original trace span kind. */
+  kind: string; /** Parsed tool-call arguments, or the raw value when parsing is not possible. */
+  arguments: unknown; /** Parsed tool-call result, or the raw value when parsing is not possible. */
+  result: unknown; /** Tool description from GenAI/Mastra metadata when present. */
+  description: string | undefined; /** Tool type from GenAI/Mastra metadata when present. */
+  toolType: string | undefined; /** Original span attributes. */
+  attributes: Record<string, unknown> | undefined; /** Original trace span for fields not normalized above. */
+  span: EvalTraceSpan$1;
+};
 /** Query helpers built from the flattened trace recorded for one eval case. */
 type EvalTraceTree$1 = {
   /** Flat span list in creation order. */spans: EvalTraceSpan$1[]; /** Top-level spans whose `parentId` is `null`. */
@@ -3696,10 +3726,16 @@ type EvalTraceTree$1 = {
   findSpans: (name: string) => EvalTraceSpan$1[]; /** Return whether any span name exactly matches `name`. */
   hasSpan: (name: string) => boolean; /** Return every span whose kind exactly matches `kind`. */
   findSpansByKind: (kind: string) => EvalTraceSpan$1[]; /** Return every span with `kind: 'tool'` or `kind: 'tool_call'`. */
-  findToolCallSpans: () => EvalTraceSpan$1[]; /** Return the names of every span with `kind: 'tool'` or `kind: 'tool_call'`. */
-  listToolCallSpanNames: () => string[]; /** Return whether a tool-call span has a name exactly matching `name`. */
-  hasToolCallSpan: (name: string) => boolean; /** Return whether a tool-call span name appears exactly `expectedCalls` times. */
-  hasNToolCallSpans: (toolName: string, expectedCalls: number) => boolean; /** Return span names in creation order, optionally filtered by kind. */
+  findToolCallSpans: () => EvalTraceSpan$1[];
+  /**
+   * Return tool-call names, preferring GenAI/Mastra tool identity attributes
+   * when available.
+   */
+  listToolCallSpanNames: () => string[]; /** Return whether a tool-call span name or tool identity matches `name`. */
+  hasToolCallSpan: (name: string) => boolean; /** Return normalized tool-call spans whose name or tool identity matches `name`. */
+  getToolCallSpans: (name: string) => EvalToolCallSpan$1[]; /** Return how many tool-call spans have a name or tool identity matching `toolName`. */
+  getToolCallSpanCount: (toolName: string) => number; /** Return whether a tool-call span name or tool identity appears exactly `expectedCalls` times. */
+  hasToolCallSpanCount: (toolName: string, expectedCalls: number) => boolean; /** Return span names in creation order, optionally filtered by kind. */
   listSpanNames: (kind?: string) => string[]; /** Return span names in depth-first tree order, optionally filtered by kind. */
   listSpanNamesDfs: (kind?: string) => string[]; /** Return all spans in depth-first tree order. */
   flattenDfs: () => EvalTraceSpan$1[];
@@ -7294,4 +7330,4 @@ declare function defineEval<TInput = unknown, TOutputs extends EvalOutputs = Eva
 /** Return whether the active eval case has tags matching the typed input. */
 declare function matchesEvalTags(input: EvalTagMatchInput): boolean;
 //#endregion
-export { AgentEvalTagRegistry, AgentEvalsConfig, type ApiCallEntry, type ApiCallMetric, type ApiCallMetricFormat, type ApiCallMetricPlacement, type ApiCallMetricValue, type ApiCallsConfigInput, type AssertionFailure, type CacheActivityEntry, type CacheAdapter, type CacheDebugKeyEntry, type CacheDebugKeyFile, type CacheDebugKeyWrite, type CacheEntry, type CacheEntryWithDebugKey, type CacheFile, type CacheHitEntry, type CacheKeyHashInput, type CacheKeyHashOptions, type CacheListItem, type CacheMode, type CacheOperationType, type CacheRecording, type CacheRecordingFrame, type CacheRecordingOp, type CacheRepairSummary, type CacheScopeContext, type CacheSerializationOptions, type CacheStatus, type CallDerivedAttribute, type CallDerivedAttributeContext, type CallDerivedAttributesConfig, type CallDerivedAttributesFn, type CaptureEvalSpanErrorLevel, type CaptureEvalSpanErrorOptions, type CaseDetail, type CaseRow, type CellValue, type ColumnDef, type ColumnFormat, type ColumnKind, type ConfigReloadState, type ConfigReloadStatus, type CreateRunRequest, type DefaultConfigKey, type DerivedStatus, type DiscoveryIssue, EvalAssertionError, type EvalCacheConfig, EvalCase, type EvalCaseScope, type EvalChartAggregate, type EvalChartAxis, type EvalChartBuiltinMetric, type EvalChartColor, type EvalChartConfig, type EvalChartMetric, type EvalChartTooltipExtra, type EvalChartType, type EvalChartsConfig, type EvalColumnOverride, type EvalColumns, EvalDefinition, type EvalDeriveConfig, type EvalDeriveContext, type EvalDeriveFn, type EvalDeriveMap, type EvalDeriveValueFn, type EvalDisplayStatus, type EvalExecuteContext, type EvalExpectation, type EvalFreshnessStatus, type EvalManualInputConfig, type EvalManualScoreDef, type EvalOutputs, type EvalOutputsSchema, type EvalRunner, type EvalRuntimeScope, EvalRuntimeUsageError, type EvalScoreContext, type EvalScoreDef, type EvalScoreFn, type EvalSetOutput, type EvalStartTime, type EvalStatAggregate, type EvalStatItem, type EvalStatsConfig, type EvalSummary, EvalTag, EvalTagMatchInput, type EvalTraceTree, type EvalTracingAssertionsConfig, type EvalTracingAssertionsFn, type JsonCell, type LlmCallCostBreakdown, type LlmCallCostCurrency, type LlmCallEntry, type LlmCallMetric, type LlmCallMetricFormat, type LlmCallMetricPlacement, type LlmCallMetricValue, type LlmCallPricing, type LlmCallPricingRate, type LlmCallPricingRegistry, type LlmCallSimulatedTokens, type LlmCallsConfigInput, type LlmCostScenario, type ManualInputDescriptor, type ManualInputFieldDescriptor, type ManualInputFieldKind, type ManualInputFieldOverride, type ManualInputFieldsConfig, type ManualInputFileValue, type ManualInputSelectOption, type MaterializeManualInputFilesResult, type NumberDisplayOptions, type ReadManualInputFileResult, type RemoveDefaultConfig, type ResolvedApiCallMetric, type ResolvedApiCallsConfig, type ResolvedCallDerivedAttribute, type ResolvedLlmCallCostCurrency, type ResolvedLlmCallMetric, type ResolvedLlmCallPricing, type ResolvedLlmCallsConfig, type RunInEvalScopeOptions, type RunLogEntry, type RunLogLevel, type RunLogLocation, type RunLogPhase, type RunLogsConfigInput, type RunManifest, type RunSummary, type ScalarCell, type ScopedCaseSummary, type ScoreTrace, type SerializedCacheSpan, type SerializedCacheValue, type SpanCacheOptions, type SseEnvelope, type SseEventType, type TraceActiveSpan, type TraceAttributeDisplay, type TraceAttributeDisplayFormat, type TraceAttributeDisplayInput, type TraceAttributeDisplayPlacement, type TraceAttributeTransform, type TraceAttributeTransformContext, type TraceCacheInfo, type TraceCacheRef, type TraceDisplayConfig, type TraceDisplayInputConfig, type TraceSpanInfo, type TrialSelectionMode, type UpdateManualScoreRequest, appendToEvalOutput, buildTraceTree, captureEvalSpanError, cleanupStagedManualInputFiles, createRunner, defineEval, deserializeCacheRecording, deserializeCacheValue, evalAssert, evalExpect, evalLog, evalSpan, evalTime, evalTracer, extractApiCalls, extractCacheEntries, extractCacheHits, extractLlmCalls, getCurrentScope, getEvalCaseInput, getEvalRegistry, getNestedAttribute, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, isManualInputFileValue, manualInputFileValueSchema, matchesEvalTags, materializeManualInputFiles, mergeEvalOutput, nextEvalId, readManualInputFile, repoFile, runCli, runInEvalRuntimeScope, runInEvalScope, runInExistingEvalScope, serializeCacheRecording, serializeCacheValue, setEvalOutput, setScopeCacheContext, simulateLlmCallCost, simulateTokenAllocation, stageManualInputFile, stageManualInputFileFromPath, startEvalBackgroundJob, z };
+export { AgentEvalTagRegistry, AgentEvalsConfig, type ApiCallEntry, type ApiCallMetric, type ApiCallMetricFormat, type ApiCallMetricPlacement, type ApiCallMetricValue, type ApiCallsConfigInput, type AssertionFailure, type CacheActivityEntry, type CacheAdapter, type CacheDebugKeyEntry, type CacheDebugKeyFile, type CacheDebugKeyWrite, type CacheEntry, type CacheEntryWithDebugKey, type CacheFile, type CacheHitEntry, type CacheKeyHashInput, type CacheKeyHashOptions, type CacheListItem, type CacheMode, type CacheOperationType, type CacheRecording, type CacheRecordingFrame, type CacheRecordingOp, type CacheRepairSummary, type CacheScopeContext, type CacheSerializationOptions, type CacheStatus, type CallDerivedAttribute, type CallDerivedAttributeContext, type CallDerivedAttributesConfig, type CallDerivedAttributesFn, type CaptureEvalSpanErrorLevel, type CaptureEvalSpanErrorOptions, type CaseDetail, type CaseRow, type CellValue, type ColumnDef, type ColumnFormat, type ColumnKind, type ConfigReloadState, type ConfigReloadStatus, type CreateRunRequest, type DefaultConfigKey, type DerivedStatus, type DiscoveryIssue, EvalAssertionError, type EvalCacheConfig, EvalCase, type EvalCaseScope, type EvalChartAggregate, type EvalChartAxis, type EvalChartBuiltinMetric, type EvalChartColor, type EvalChartConfig, type EvalChartMetric, type EvalChartTooltipExtra, type EvalChartType, type EvalChartsConfig, type EvalColumnOverride, type EvalColumns, EvalDefinition, type EvalDeriveConfig, type EvalDeriveContext, type EvalDeriveFn, type EvalDeriveMap, type EvalDeriveValueFn, type EvalDisplayStatus, type EvalExecuteContext, type EvalExpectation, type EvalFreshnessStatus, type EvalManualInputConfig, type EvalManualScoreDef, type EvalOutputs, type EvalOutputsSchema, type EvalRunner, type EvalRuntimeScope, EvalRuntimeUsageError, type EvalScoreContext, type EvalScoreDef, type EvalScoreFn, type EvalSetOutput, type EvalStartTime, type EvalStatAggregate, type EvalStatItem, type EvalStatsConfig, type EvalSummary, EvalTag, EvalTagMatchInput, type EvalToolCallSpan, type EvalTraceTree, type EvalTracingAssertionsConfig, type EvalTracingAssertionsFn, type JsonCell, type LlmCallCostBreakdown, type LlmCallCostCurrency, type LlmCallEntry, type LlmCallMetric, type LlmCallMetricFormat, type LlmCallMetricPlacement, type LlmCallMetricValue, type LlmCallPricing, type LlmCallPricingRate, type LlmCallPricingRegistry, type LlmCallSimulatedTokens, type LlmCallsConfigInput, type LlmCostScenario, type ManualInputDescriptor, type ManualInputFieldDescriptor, type ManualInputFieldKind, type ManualInputFieldOverride, type ManualInputFieldsConfig, type ManualInputFileValue, type ManualInputSelectOption, type MaterializeManualInputFilesResult, type NumberDisplayOptions, type ReadManualInputFileResult, type RemoveDefaultConfig, type ResolvedApiCallMetric, type ResolvedApiCallsConfig, type ResolvedCallDerivedAttribute, type ResolvedLlmCallCostCurrency, type ResolvedLlmCallMetric, type ResolvedLlmCallPricing, type ResolvedLlmCallsConfig, type RunInEvalScopeOptions, type RunLogEntry, type RunLogLevel, type RunLogLocation, type RunLogPhase, type RunLogsConfigInput, type RunManifest, type RunSummary, type ScalarCell, type ScopedCaseSummary, type ScoreTrace, type SerializedCacheSpan, type SerializedCacheValue, type SpanCacheOptions, type SseEnvelope, type SseEventType, type TraceActiveSpan, type TraceAttributeDisplay, type TraceAttributeDisplayFormat, type TraceAttributeDisplayInput, type TraceAttributeDisplayPlacement, type TraceAttributeTransform, type TraceAttributeTransformContext, type TraceCacheInfo, type TraceCacheRef, type TraceDisplayConfig, type TraceDisplayInputConfig, type TraceSpanInfo, type TrialSelectionMode, type UpdateManualScoreRequest, appendToEvalOutput, buildTraceTree, captureEvalSpanError, cleanupStagedManualInputFiles, createRunner, defineEval, deserializeCacheRecording, deserializeCacheValue, evalAssert, evalExpect, evalLog, evalSpan, evalTime, evalTracer, extractApiCalls, extractCacheEntries, extractCacheHits, extractLlmCalls, getCurrentScope, getEvalCaseInput, getEvalRegistry, getNestedAttribute, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, isManualInputFileValue, manualInputFileValueSchema, matchesEvalTags, materializeManualInputFiles, mergeEvalOutput, nextEvalId, readManualInputFile, repoFile, runCli, runInEvalRuntimeScope, runInEvalScope, runInExistingEvalScope, serializeCacheRecording, serializeCacheValue, setEvalOutput, setScopeCacheContext, simulateLlmCallCost, simulateTokenAllocation, stageManualInputFile, stageManualInputFileFromPath, startEvalBackgroundJob, z };

package/dist/index.mjs

@@ -1,4 +1,4 @@
-import { $ as setEvalOutput, A as serializeCacheValue, B as evalLog, C as evalSpan, D as deserializeCacheRecording, E as hashCacheKeySync, F as EvalAssertionError, G as isInEvalScope, H as getCurrentScope, I as EvalRuntimeUsageError, It as getEvalRegistry, J as nextEvalId, L as appendToEvalOutput, M as manualInputFileValueSchema, N as readManualInputFile, O as deserializeCacheValue, P as evalExpect, S as captureEvalSpanError, T as hashCacheKey, U as getEvalCaseInput, V as evalTime, W as incrementEvalOutput, X as runInEvalScope, Y as runInEvalRuntimeScope, Z as runInExistingEvalScope, at as extractCacheHits, b as z, ct as simulateLlmCallCost, dt as getNestedAttribute, et as setScopeCacheContext, it as extractCacheEntries, j as repoFile, k as serializeCacheRecording, lt as simulateTokenAllocation, ot as extractApiCalls, q as mergeEvalOutput, st as extractLlmCalls, tt as startEvalBackgroundJob, w as evalTracer, x as buildTraceTree, z as evalAssert } from "./runExecution-BMnJXWhN.mjs";
-import { a as materializeManualInputFiles, i as isManualInputFileValue, n as createRunner, o as stageManualInputFile, r as cleanupStagedManualInputFiles, s as stageManualInputFileFromPath, t as runCli } from "./cli-ClAkjTvo.mjs";
-import { n as matchesEvalTags, t as defineEval } from "./src-DfzidkYr.mjs";
+import { $ as setEvalOutput, A as serializeCacheValue, B as evalLog, C as evalSpan, D as deserializeCacheRecording, E as hashCacheKeySync, F as EvalAssertionError, G as isInEvalScope, H as getCurrentScope, I as EvalRuntimeUsageError, It as getEvalRegistry, J as nextEvalId, L as appendToEvalOutput, M as manualInputFileValueSchema, N as readManualInputFile, O as deserializeCacheValue, P as evalExpect, S as captureEvalSpanError, T as hashCacheKey, U as getEvalCaseInput, V as evalTime, W as incrementEvalOutput, X as runInEvalScope, Y as runInEvalRuntimeScope, Z as runInExistingEvalScope, at as extractCacheHits, b as z, ct as simulateLlmCallCost, dt as getNestedAttribute, et as setScopeCacheContext, it as extractCacheEntries, j as repoFile, k as serializeCacheRecording, lt as simulateTokenAllocation, ot as extractApiCalls, q as mergeEvalOutput, st as extractLlmCalls, tt as startEvalBackgroundJob, w as evalTracer, x as buildTraceTree, z as evalAssert } from "./runExecution-C3XVZHRC.mjs";
+import { a as materializeManualInputFiles, i as isManualInputFileValue, n as createRunner, o as stageManualInputFile, r as cleanupStagedManualInputFiles, s as stageManualInputFileFromPath, t as runCli } from "./cli-Dkp2-rBm.mjs";
+import { n as matchesEvalTags, t as defineEval } from "./src-8dGXUULC.mjs";
 export { EvalAssertionError, EvalRuntimeUsageError, appendToEvalOutput, buildTraceTree, captureEvalSpanError, cleanupStagedManualInputFiles, createRunner, defineEval, deserializeCacheRecording, deserializeCacheValue, evalAssert, evalExpect, evalLog, evalSpan, evalTime, evalTracer, extractApiCalls, extractCacheEntries, extractCacheHits, extractLlmCalls, getCurrentScope, getEvalCaseInput, getEvalRegistry, getNestedAttribute, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, isManualInputFileValue, manualInputFileValueSchema, matchesEvalTags, materializeManualInputFiles, mergeEvalOutput, nextEvalId, readManualInputFile, repoFile, runCli, runInEvalRuntimeScope, runInEvalScope, runInExistingEvalScope, serializeCacheRecording, serializeCacheValue, setEvalOutput, setScopeCacheContext, simulateLlmCallCost, simulateTokenAllocation, stageManualInputFile, stageManualInputFileFromPath, startEvalBackgroundJob, z };

package/dist/runChild.mjs

@@ -1,5 +1,5 @@
-import { At as evalStatAggregateSchema, Et as buildEvalKey, Mt as manualInputDescriptorSchema, Nt as evalChartsConfigSchema, Pt as columnDefSchema, R as configureEvalRunLogs, St as runSummarySchema, jt as evalStatsConfigSchema, l as registerAgentEvalsPackageResolutionHooks, nt as createRunRequestSchema, p as loadConfig, v as createFsCacheStore, xt as runManifestSchema, y as getCacheRetentionOptions } from "./runExecution-BMnJXWhN.mjs";
-import { S as parseEvalDiscovery, m as persistRunState, r as getTargetEvals$1, t as executeRun } from "./runOrchestration-CvbTAoEb.mjs";
+import { At as evalStatAggregateSchema, Et as buildEvalKey, Mt as manualInputDescriptorSchema, Nt as evalChartsConfigSchema, Pt as columnDefSchema, R as configureEvalRunLogs, St as runSummarySchema, jt as evalStatsConfigSchema, l as registerAgentEvalsPackageResolutionHooks, nt as createRunRequestSchema, p as loadConfig, v as createFsCacheStore, xt as runManifestSchema, y as getCacheRetentionOptions } from "./runExecution-C3XVZHRC.mjs";
+import { S as parseEvalDiscovery, m as persistRunState, r as getTargetEvals$1, t as executeRun } from "./runOrchestration-B5An-AEi.mjs";
 import { z } from "zod/v4";
 import { readFile } from "node:fs/promises";
 import { relative } from "node:path";

package/dist/{runExecution-BMnJXWhN.mjs → runExecution-C3XVZHRC.mjs}

@@ -5142,8 +5142,88 @@ function buildTraceTree(spans, checkpoints) {
 		visit(null);
 		return result;
 	};
+	const isRecord = (value) => {
+		return typeof value === "object" && value !== null;
+	};
+	const readRecordValue = (value, key) => {
+		if (!isRecord(value)) return void 0;
+		const child = value[key];
+		return isRecord(child) ? child : void 0;
+	};
+	const readStringValue = (value, key) => {
+		if (!isRecord(value)) return void 0;
+		const child = value[key];
+		return typeof child === "string" && child.length > 0 ? child : void 0;
+	};
+	const readValue = (value, key) => {
+		if (!isRecord(value)) return void 0;
+		return value[key];
+	};
+	const parseMaybeJson = (value) => {
+		if (typeof value !== "string") return value;
+		const parsed = resultify(() => JSON.parse(value));
+		return parsed.error ? value : parsed.value;
+	};
+	const firstDefined = (values) => {
+		return values.find((value) => value !== void 0);
+	};
+	const getToolCallMetadata = (span) => {
+		const attributes = span.attributes;
+		return {
+			attributes,
+			genAI: readRecordValue(attributes, "genAI"),
+			mastra: readRecordValue(attributes, "mastra"),
+			toolAttributes: readRecordValue(attributes, "attributes")
+		};
+	};
 	const isToolCallSpan = (span) => {
-		return span.kind === "tool" || span.kind === "tool_call";
+		const { attributes, genAI, mastra } = getToolCallMetadata(span);
+		return span.kind === "tool" || span.kind === "tool_call" || readStringValue(attributes, "gen_ai.tool.type") === "tool" || readStringValue(genAI, "gen_ai.tool.type") === "tool" || readStringValue(genAI, "mastra.span.type") === "tool_call" || readStringValue(mastra, "type") === "tool_call" || readStringValue(mastra, "entityType") === "tool";
+	};
+	const getToolCallIdentityNames = (span) => {
+		const { attributes, genAI, mastra } = getToolCallMetadata(span);
+		return [
+			readStringValue(attributes, "gen_ai.tool.name"),
+			readStringValue(genAI, "gen_ai.tool.name"),
+			readStringValue(mastra, "entityName"),
+			readStringValue(mastra, "entityId"),
+			span.name
+		].filter((name) => name !== void 0);
+	};
+	const getPreferredToolCallName = (span) => {
+		return getToolCallIdentityNames(span)[0] ?? span.name;
+	};
+	const toolCallSpanMatchesName = (span, toolName) => {
+		return getToolCallIdentityNames(span).includes(toolName);
+	};
+	const countToolCallSpans = (toolName) => {
+		return spans.filter((span) => {
+			return isToolCallSpan(span) && toolCallSpanMatchesName(span, toolName);
+		}).length;
+	};
+	const buildToolCallSpan = (span) => {
+		const { attributes, genAI, toolAttributes } = getToolCallMetadata(span);
+		return {
+			name: getPreferredToolCallName(span),
+			spanName: span.name,
+			kind: span.kind,
+			arguments: parseMaybeJson(firstDefined([
+				readValue(attributes, "gen_ai.tool.call.arguments"),
+				readValue(genAI, "gen_ai.tool.call.arguments"),
+				readValue(attributes, "arguments"),
+				readValue(attributes, "input")
+			])),
+			result: parseMaybeJson(firstDefined([
+				readValue(attributes, "gen_ai.tool.call.result"),
+				readValue(genAI, "gen_ai.tool.call.result"),
+				readValue(attributes, "result"),
+				readValue(attributes, "output")
+			])),
+			description: readStringValue(attributes, "gen_ai.tool.description") ?? readStringValue(genAI, "gen_ai.tool.description") ?? readStringValue(toolAttributes, "toolDescription"),
+			toolType: readStringValue(attributes, "gen_ai.tool.type") ?? readStringValue(genAI, "gen_ai.tool.type") ?? readStringValue(toolAttributes, "toolType"),
+			attributes,
+			span
+		};
 	};
 	const filterSpanNames = (sourceSpans, kind) => {
 		return sourceSpans.filter((span) => kind === void 0 || span.kind === kind).map((span) => span.name);
@@ -5167,15 +5247,23 @@ function buildTraceTree(spans, checkpoints) {
 			return spans.filter(isToolCallSpan);
 		},
 		listToolCallSpanNames() {
-			return spans.filter(isToolCallSpan).map((span) => span.name);
+			return spans.filter(isToolCallSpan).map(getPreferredToolCallName);
 		},
 		hasToolCallSpan(name) {
-			return spans.some((s) => isToolCallSpan(s) && s.name === name);
+			return spans.some((s) => {
+				return isToolCallSpan(s) && toolCallSpanMatchesName(s, name);
+			});
 		},
-		hasNToolCallSpans(toolName, expectedCalls) {
+		getToolCallSpans(name) {
 			return spans.filter((span) => {
-				return isToolCallSpan(span) && span.name === toolName;
-			}).length === expectedCalls;
+				return isToolCallSpan(span) && toolCallSpanMatchesName(span, name);
+			}).map(buildToolCallSpan);
+		},
+		getToolCallSpanCount(toolName) {
+			return countToolCallSpans(toolName);
+		},
+		hasToolCallSpanCount(toolName, expectedCalls) {
+			return countToolCallSpans(toolName) === expectedCalls;
 		},
 		listSpanNames(kind) {
 			return filterSpanNames(spans, kind);

package/dist/{runOrchestration-CvbTAoEb.mjs → runOrchestration-B5An-AEi.mjs}

@@ -1,4 +1,4 @@
-import { Dt as getCaseRowCaseKey, Lt as runWithEvalRegistry, Ot as caseDetailSchema, Q as runWithEvalClock, St as runSummarySchema, Tt as buildCaseKey, Y as runInEvalRuntimeScope, _t as dedupeEvalTags, bt as validateTagsFilterExpression, d as loadEvalModule, f as resolveEvalDefaultConfig, g as commitPendingCacheWrites, gt as deriveStatusFromChildStatuses, ht as deriveStatusFromCaseRows, i as isCaseChildMessage, kt as caseRowSchema, m as buildDeclaredColumnDefs, mt as deriveScopedSummaryFromCases, n as resolveRunnableEvalCases, o as stripTerminalControlCodes, t as filterEvalCases, u as runWithModuleIsolation, vt as matchesTagsFilter, xt as runManifestSchema, yt as validateEvalTagName } from "./runExecution-BMnJXWhN.mjs";
+import { Dt as getCaseRowCaseKey, Lt as runWithEvalRegistry, Ot as caseDetailSchema, Q as runWithEvalClock, St as runSummarySchema, Tt as buildCaseKey, Y as runInEvalRuntimeScope, _t as dedupeEvalTags, bt as validateTagsFilterExpression, d as loadEvalModule, f as resolveEvalDefaultConfig, g as commitPendingCacheWrites, gt as deriveStatusFromChildStatuses, ht as deriveStatusFromCaseRows, i as isCaseChildMessage, kt as caseRowSchema, m as buildDeclaredColumnDefs, mt as deriveScopedSummaryFromCases, n as resolveRunnableEvalCases, o as stripTerminalControlCodes, t as filterEvalCases, u as runWithModuleIsolation, vt as matchesTagsFilter, xt as runManifestSchema, yt as validateEvalTagName } from "./runExecution-C3XVZHRC.mjs";
 import { Result, resultify } from "t-result";
 import { readFile, readdir, rm, writeFile } from "node:fs/promises";
 import { dirname, join } from "node:path";

package/dist/{runner-DJJekv9f.mjs → runner-BJXz_V_V.mjs}

@@ -1,2 +1,2 @@
-import { n as initRunner, t as getRunnerInstance } from "./runner-BfHgVhGS.mjs";
+import { n as initRunner, t as getRunnerInstance } from "./runner-C9J-1fkp.mjs";
 export { getRunnerInstance, initRunner };

package/dist/{runner-BfHgVhGS.mjs → runner-C9J-1fkp.mjs}

@@ -1,5 +1,5 @@
-import { n as createRunner } from "./cli-ClAkjTvo.mjs";
-import "./src-DfzidkYr.mjs";
+import { n as createRunner } from "./cli-Dkp2-rBm.mjs";
+import "./src-8dGXUULC.mjs";
 //#region ../../apps/server/src/runner.ts
 let runnerInstance = null;
 function getRunnerInstance() {

package/dist/{src-DfzidkYr.mjs → src-8dGXUULC.mjs}

@@ -1,5 +1,5 @@
-import { Ft as defineEval$1, K as matchesEvalTags$1 } from "./runExecution-BMnJXWhN.mjs";
-import "./cli-ClAkjTvo.mjs";
+import { Ft as defineEval$1, K as matchesEvalTags$1 } from "./runExecution-C3XVZHRC.mjs";
+import "./cli-Dkp2-rBm.mjs";
 //#region src/index.ts
 /** Register an eval definition with typed tag support. */
 function defineEval(definition) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ls-stack/agent-eval",
-  "version": "0.58.5",
+  "version": "0.59.0",
   "type": "module",
   "bin": {
     "agent-evals": "./dist/bin.mjs"
@@ -32,9 +32,9 @@
   "devDependencies": {
     "@types/node": "^24.7.2",
     "typescript": "^5.9.2",
-    "@agent-evals/sdk": "0.0.1",
     "@agent-evals/runner": "0.0.1",
-    "@agent-evals/shared": "0.0.1"
+    "@agent-evals/shared": "0.0.1",
+    "@agent-evals/sdk": "0.0.1"
   },
   "scripts": {
     "build": "pnpm --filter @agent-evals/web build && pnpm --filter @agent-evals/shared build && pnpm --filter @agent-evals/sdk build && pnpm --filter @agent-evals/runner build && tsdown --filter cli-js && tsdown --filter cli-types",

package/skills/agent-eval/SKILL.md

@@ -375,11 +375,19 @@ See `EvalScoreDef` / `EvalManualScoreDef` in the types for the full shape
   `trace.hasSpan(name)`, `trace.findSpansByKind(kind)`,
   `trace.findToolCallSpans()`, `trace.listToolCallSpanNames()`,
   `trace.hasToolCallSpan(name)`,
-  `trace.hasNToolCallSpans(toolName, expectedCalls)`,
+  `trace.getToolCallSpans(name)`,
+  `trace.getToolCallSpanCount(toolName)`,
+  `trace.hasToolCallSpanCount(toolName, expectedCalls)`,
   `trace.listSpanNames(kind?)`, `trace.listSpanNamesDfs(kind?)`, and
   `trace.flattenDfs()`.
   The tool-call helpers include both `kind: 'tool'` spans and imported
-  execution spans recorded as `kind: 'tool_call'`.
+  execution spans recorded as `kind: 'tool_call'`. Tool-name checks and counts
+  match the span `name` as well as GenAI/Mastra identity attributes such as
+  `genAI["gen_ai.tool.name"]` and `mastra.entityName`; list helpers prefer
+  those tool identity attributes when present. `getToolCallSpans(name)`
+  returns one normalized object per matching call, including parsed
+  `arguments`, parsed `result`, `description`, `toolType`, `attributes`, and
+  the original `span`.
 - `traceDisplay` promotes selected span attributes into the trace tree and
   detail pane; it supports aggregation across subtrees (`scope`, `mode`) and
   user-defined `transform(...)` for derived views (e.g. currency conversion).