agentv 3.2.4 → 3.2.5

package/dist/{chunk-VBGYESW7.js → chunk-6XTYVCMN.js}

@@ -301,7 +301,7 @@ var require_dist = __commonJS({
   }
 });
 
-// ../../packages/core/dist/chunk-5SQK5FXC.js
+// ../../packages/core/dist/chunk-DVFNM65P.js
 import { constants } from "node:fs";
 import { access, readFile } from "node:fs/promises";
 import path from "node:path";
@@ -419,7 +419,7 @@ __export(external_exports2, {
   void: () => voidType
 });
 
-// ../../packages/core/dist/chunk-5SQK5FXC.js
+// ../../packages/core/dist/chunk-DVFNM65P.js
 var TEST_MESSAGE_ROLE_VALUES = ["system", "user", "assistant", "tool"];
 var TEST_MESSAGE_ROLES = TEST_MESSAGE_ROLE_VALUES;
 var TEST_MESSAGE_ROLE_SET = new Set(TEST_MESSAGE_ROLE_VALUES);
@@ -17856,7 +17856,14 @@ async function invokeModel(options) {
 function mapResponse(result) {
   const content = result.text ?? "";
   const rawUsage = result.totalUsage ?? result.usage;
-  const tokenUsage = rawUsage?.inputTokens != null && rawUsage?.outputTokens != null ? { input: rawUsage.inputTokens, output: rawUsage.outputTokens } : void 0;
+  const reasoning = rawUsage?.outputTokenDetails?.reasoningTokens ?? void 0;
+  const cached = rawUsage?.inputTokenDetails?.cacheReadTokens ?? void 0;
+  const tokenUsage = rawUsage?.inputTokens != null && rawUsage?.outputTokens != null ? {
+    input: rawUsage.inputTokens,
+    output: rawUsage.outputTokens,
+    ...reasoning != null ? { reasoning } : {},
+    ...cached != null ? { cached } : {}
+  } : void 0;
   return {
     raw: result,
     usage: toJsonObject(rawUsage),
@@ -18191,10 +18198,12 @@ var ClaudeCliProvider = class {
             if (usage) {
               const inputTokens = (usage.input_tokens ?? 0) + (usage.cache_read_input_tokens ?? 0) + (usage.cache_creation_input_tokens ?? 0);
               const outputTokens = usage.output_tokens ?? 0;
+              const reasoningTokens = usage.reasoning_tokens ?? void 0;
               tokenUsage = {
                 input: inputTokens,
                 output: outputTokens,
-                cached: usage.cache_read_input_tokens ?? void 0
+                cached: usage.cache_read_input_tokens ?? void 0,
+                reasoning: reasoningTokens
               };
               request.streamCallbacks?.onLlmCallEnd?.(this.config.model ?? "claude", tokenUsage);
             }
@@ -29279,6 +29288,7 @@ async function runEvalCase(options) {
     } catch {
     }
   }
+  const caseStartMs = Date.now();
   const attemptBudget = (maxRetries ?? 0) + 1;
   let attempt = 0;
   let providerResponse = cachedResponse;
@@ -29427,9 +29437,22 @@ async function runEvalCase(options) {
       fileChanges,
       workspacePath
     });
+    const totalDurationMs = Date.now() - caseStartMs;
+    const graderTokens = aggregateEvaluatorTokenUsage(result.scores);
+    const evalRunTokenUsage = tokenUsage || graderTokens ? {
+      input: (tokenUsage?.input ?? 0) + (graderTokens?.input ?? 0),
+      output: (tokenUsage?.output ?? 0) + (graderTokens?.output ?? 0),
+      ...tokenUsage?.reasoning != null || graderTokens?.reasoning != null ? { reasoning: (tokenUsage?.reasoning ?? 0) + (graderTokens?.reasoning ?? 0) } : {},
+      ...tokenUsage?.cached != null || graderTokens?.cached != null ? { cached: (tokenUsage?.cached ?? 0) + (graderTokens?.cached ?? 0) } : {}
+    } : void 0;
+    const evalRun = {
+      durationMs: totalDurationMs,
+      ...evalRunTokenUsage ? { tokenUsage: evalRunTokenUsage } : {}
+    };
     const executionStatus = providerError ? "execution_error" : classifyQualityStatus(result.score);
     const finalResult = providerError ? {
       ...result,
+      evalRun,
       error: providerError,
       executionStatus,
       failureStage: "agent",
@@ -29438,7 +29461,7 @@ async function runEvalCase(options) {
       beforeAllOutput,
       beforeEachOutput,
       afterEachOutput
-    } : { ...result, executionStatus, beforeAllOutput, beforeEachOutput, afterEachOutput };
+    } : { ...result, evalRun, executionStatus, beforeAllOutput, beforeEachOutput, afterEachOutput };
     const isFailure = !!finalResult.error || finalResult.score < 0.5;
     if (workspacePath && !isSharedWorkspace) {
       if (forceCleanup) {
@@ -29458,6 +29481,7 @@ async function runEvalCase(options) {
     }
     return finalResult;
   } catch (error) {
+    const evalRun = { durationMs: Date.now() - caseStartMs };
     const errorResult = buildErrorResult(
       evalCase,
       target.name,
@@ -29473,10 +29497,10 @@ async function runEvalCase(options) {
         await cleanupWorkspace(workspacePath).catch(() => {
         });
       } else {
-        return { ...errorResult, workspacePath, beforeEachOutput, afterEachOutput };
+        return { ...errorResult, evalRun, workspacePath, beforeEachOutput, afterEachOutput };
       }
     }
-    return { ...errorResult, beforeEachOutput, afterEachOutput };
+    return { ...errorResult, evalRun, beforeEachOutput, afterEachOutput };
   }
 }
 async function runEvalCaseWithTrials(options, trialsConfig) {
@@ -30051,6 +30075,44 @@ function buildResultInput(promptInputs) {
   }
   return promptInputs.question;
 }
+function aggregateEvaluatorTokenUsage(scores) {
+  if (!scores || scores.length === 0) return void 0;
+  let hasAny = false;
+  let input = 0;
+  let output = 0;
+  let reasoning = 0;
+  let cached = 0;
+  let hasReasoning = false;
+  let hasCached = false;
+  const visit = (items) => {
+    for (const item of items) {
+      if (item.tokenUsage) {
+        hasAny = true;
+        input += item.tokenUsage.input;
+        output += item.tokenUsage.output;
+        if (item.tokenUsage.reasoning != null) {
+          hasReasoning = true;
+          reasoning += item.tokenUsage.reasoning;
+        }
+        if (item.tokenUsage.cached != null) {
+          hasCached = true;
+          cached += item.tokenUsage.cached;
+        }
+      }
+      if (item.scores) {
+        visit(item.scores);
+      }
+    }
+  };
+  visit(scores);
+  if (!hasAny) return void 0;
+  return {
+    input,
+    output,
+    ...hasReasoning ? { reasoning } : {},
+    ...hasCached ? { cached } : {}
+  };
+}
 function isTimeoutLike(error) {
   if (!error) {
     return false;
@@ -31086,4 +31148,4 @@ export {
   OtelStreamingObserver,
   createAgentKernel
 };
-//# sourceMappingURL=chunk-VBGYESW7.js.map
+//# sourceMappingURL=chunk-6XTYVCMN.js.map