agentv 3.2.5 → 3.4.0

package/dist/{chunk-6XTYVCMN.js → chunk-GOZV2HN2.js}

@@ -301,7 +301,7 @@ var require_dist = __commonJS({
   }
 });
 
-// ../../packages/core/dist/chunk-DVFNM65P.js
+// ../../packages/core/dist/chunk-JO4HIAEF.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-DVFNM65P.js
+// ../../packages/core/dist/chunk-JO4HIAEF.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);
@@ -1810,6 +1810,7 @@ var AGENT_PROVIDER_KINDS = [
   "copilot-sdk",
   "copilot-cli",
   "pi-coding-agent",
+  "pi-agent-sdk",
   "claude",
   "claude-cli",
   "claude-sdk",
@@ -14657,19 +14658,26 @@ function logWarning(message) {
   console.warn(`${ANSI_YELLOW2}Warning: ${message}${ANSI_RESET3}`);
 }
 var TEMPLATE_VARIABLES = {
+  /** @deprecated Use OUTPUT_TEXT instead */
   ANSWER: "answer",
   EXPECTED_OUTPUT: "expected_output",
+  /** @deprecated Use INPUT_TEXT instead */
   QUESTION: "question",
   CRITERIA: "criteria",
+  /** @deprecated Use EXPECTED_OUTPUT_TEXT instead */
   REFERENCE_ANSWER: "reference_answer",
   INPUT: "input",
   OUTPUT: "output",
-  FILE_CHANGES: "file_changes"
+  FILE_CHANGES: "file_changes",
+  INPUT_TEXT: "input_text",
+  OUTPUT_TEXT: "output_text",
+  EXPECTED_OUTPUT_TEXT: "expected_output_text"
 };
 var VALID_TEMPLATE_VARIABLES = new Set(Object.values(TEMPLATE_VARIABLES));
 var REQUIRED_TEMPLATE_VARIABLES = /* @__PURE__ */ new Set([
   TEMPLATE_VARIABLES.ANSWER,
-  TEMPLATE_VARIABLES.EXPECTED_OUTPUT
+  TEMPLATE_VARIABLES.EXPECTED_OUTPUT,
+  TEMPLATE_VARIABLES.OUTPUT_TEXT
 ]);
 var ANSI_YELLOW3 = "\x1B[33m";
 var ANSI_RESET4 = "\x1B[0m";
@@ -14690,13 +14698,13 @@ function validateTemplateVariables(content, source) {
     }
     match = variablePattern.exec(content);
   }
-  const hasCandidateAnswer = foundVariables.has(TEMPLATE_VARIABLES.ANSWER);
+  const hasCandidateAnswer = foundVariables.has(TEMPLATE_VARIABLES.ANSWER) || foundVariables.has(TEMPLATE_VARIABLES.OUTPUT_TEXT);
   const hasExpectedOutput = foundVariables.has(TEMPLATE_VARIABLES.EXPECTED_OUTPUT);
   const hasRequiredFields = hasCandidateAnswer || hasExpectedOutput;
   if (!hasRequiredFields) {
     throw new Error(
       `Missing required fields. Must include at least one of:
-  - {{ ${TEMPLATE_VARIABLES.ANSWER} }}
+  - {{ ${TEMPLATE_VARIABLES.ANSWER} }} or {{ ${TEMPLATE_VARIABLES.OUTPUT_TEXT} }}
   - {{ ${TEMPLATE_VARIABLES.EXPECTED_OUTPUT} }}`
     );
   }
@@ -20905,6 +20913,29 @@ var MockProvider = class {
     return this.delayMs;
   }
 };
+function extractPiTextContent(content) {
+  if (typeof content === "string") {
+    return content;
+  }
+  if (!Array.isArray(content)) {
+    return void 0;
+  }
+  const textParts = [];
+  for (const part of content) {
+    if (!part || typeof part !== "object") {
+      continue;
+    }
+    const p = part;
+    if (p.type === "text" && typeof p.text === "string") {
+      textParts.push(p.text);
+    }
+  }
+  return textParts.length > 0 ? textParts.join("\n") : void 0;
+}
+function toFiniteNumber(value) {
+  if (typeof value === "number" && Number.isFinite(value)) return value;
+  return void 0;
+}
 var piAgentModule = null;
 var piAiModule = null;
 async function loadPiModules() {
@@ -20945,7 +20976,8 @@ var PiAgentSdkProvider = class {
       throw new Error("Pi agent SDK request was aborted before execution");
     }
     const { Agent, getModel, getEnvApiKey } = await loadPiModules();
-    const startTime = Date.now();
+    const startTimeIso = (/* @__PURE__ */ new Date()).toISOString();
+    const startMs = Date.now();
     const providerName = this.config.provider ?? "anthropic";
     const modelId = this.config.model ?? "claude-sonnet-4-20250514";
     const model = getModel(providerName, modelId);
@@ -20962,16 +20994,73 @@ var PiAgentSdkProvider = class {
         return this.config.apiKey ?? getEnvApiKey(provider) ?? void 0;
       }
     });
-    const output = [];
-    let finalAssistantContent = "";
+    let tokenUsage;
+    let costUsd;
+    const toolTrackers = /* @__PURE__ */ new Map();
+    const completedToolResults = /* @__PURE__ */ new Map();
     const unsubscribe = agent.subscribe((event) => {
-      if (event.type === "message_end") {
-        const msg = event.message;
-        if (msg.role === "assistant") {
-          const content = extractTextContent3(msg.content);
-          if (content) {
-            finalAssistantContent = content;
+      switch (event.type) {
+        case "message_end": {
+          const msg = event.message;
+          if (msg && typeof msg === "object" && "role" in msg && msg.role === "assistant" && "usage" in msg) {
+            const usage = msg.usage;
+            if (usage && typeof usage === "object") {
+              const u = usage;
+              const input = toFiniteNumber(u.input);
+              const output = toFiniteNumber(u.output);
+              const cached = toFiniteNumber(u.cacheRead);
+              let callDelta;
+              if (input !== void 0 || output !== void 0) {
+                callDelta = {
+                  input: input ?? 0,
+                  output: output ?? 0,
+                  ...cached !== void 0 ? { cached } : {}
+                };
+                tokenUsage = {
+                  input: (tokenUsage?.input ?? 0) + callDelta.input,
+                  output: (tokenUsage?.output ?? 0) + callDelta.output,
+                  ...cached !== void 0 ? { cached: (tokenUsage?.cached ?? 0) + cached } : tokenUsage?.cached !== void 0 ? { cached: tokenUsage.cached } : {}
+                };
+              }
+              const cost = u.cost;
+              if (cost && typeof cost === "object") {
+                const total = toFiniteNumber(cost.total);
+                if (total !== void 0) {
+                  costUsd = (costUsd ?? 0) + total;
+                }
+              }
+              request.streamCallbacks?.onLlmCallEnd?.(modelId, callDelta);
+            }
           }
+          break;
+        }
+        case "tool_execution_start": {
+          toolTrackers.set(event.toolCallId, {
+            toolCallId: event.toolCallId,
+            toolName: event.toolName,
+            args: event.args,
+            startMs: Date.now(),
+            startTime: (/* @__PURE__ */ new Date()).toISOString()
+          });
+          request.streamCallbacks?.onToolCallStart?.(event.toolName, event.toolCallId);
+          break;
+        }
+        case "tool_execution_end": {
+          const tracker = toolTrackers.get(event.toolCallId);
+          const durationMs = tracker ? Date.now() - tracker.startMs : 0;
+          completedToolResults.set(event.toolCallId, {
+            output: event.result,
+            durationMs
+          });
+          request.streamCallbacks?.onToolCallEnd?.(
+            event.toolName,
+            tracker?.args,
+            event.result,
+            durationMs,
+            event.toolCallId
+          );
+          toolTrackers.delete(event.toolCallId);
+          break;
         }
       }
     });
@@ -20990,10 +21079,12 @@ var PiAgentSdkProvider = class {
       }
       await agent.waitForIdle();
       const agentMessages = agent.state.messages;
+      const output = [];
       for (const msg of agentMessages) {
-        output.push(convertAgentMessage(msg));
+        output.push(convertAgentMessage(msg, toolTrackers, completedToolResults));
       }
-      const durationMs = Date.now() - startTime;
+      const endTimeIso = (/* @__PURE__ */ new Date()).toISOString();
+      const durationMs = Date.now() - startMs;
       return {
         raw: {
           messages: agentMessages,
@@ -21002,49 +21093,54 @@ var PiAgentSdkProvider = class {
           provider: this.config.provider
         },
         output,
-        durationMs
+        tokenUsage,
+        costUsd,
+        durationMs,
+        startTime: startTimeIso,
+        endTime: endTimeIso
       };
     } finally {
       unsubscribe();
     }
   }
 };
-function extractTextContent3(content) {
-  if (typeof content === "string") {
-    return content;
-  }
-  if (!Array.isArray(content)) {
-    return void 0;
-  }
-  const textParts = [];
-  for (const part of content) {
-    if (!part || typeof part !== "object") {
-      continue;
-    }
-    const p = part;
-    if (p.type === "text" && typeof p.text === "string") {
-      textParts.push(p.text);
-    }
-  }
-  return textParts.length > 0 ? textParts.join("\n") : void 0;
-}
-function convertAgentMessage(message) {
+function convertAgentMessage(message, toolTrackers, completedToolResults) {
   if (!message || typeof message !== "object") {
     return { role: "unknown", content: String(message) };
   }
   const msg = message;
   const role = typeof msg.role === "string" ? msg.role : "unknown";
-  const content = extractTextContent3(msg.content);
-  const toolCalls = extractToolCalls3(msg.content);
+  const content = extractPiTextContent(msg.content);
+  const toolCalls = extractToolCalls3(msg.content, toolTrackers, completedToolResults);
   const startTime = typeof msg.timestamp === "number" ? new Date(msg.timestamp).toISOString() : typeof msg.timestamp === "string" ? msg.timestamp : void 0;
+  let msgTokenUsage;
+  if (msg.usage && typeof msg.usage === "object") {
+    const u = msg.usage;
+    const input = toFiniteNumber(u.input);
+    const output = toFiniteNumber(u.output);
+    if (input !== void 0 || output !== void 0) {
+      msgTokenUsage = {
+        input: input ?? 0,
+        output: output ?? 0,
+        ...toFiniteNumber(u.cacheRead) !== void 0 ? { cached: toFiniteNumber(u.cacheRead) } : {}
+      };
+    }
+  }
+  const metadata = {};
+  if (msg.api) metadata.api = msg.api;
+  if (msg.provider) metadata.provider = msg.provider;
+  if (msg.model) metadata.model = msg.model;
+  if (msg.stopReason) metadata.stopReason = msg.stopReason;
   return {
     role,
     content,
     toolCalls: toolCalls.length > 0 ? toolCalls : void 0,
-    startTime
+    startTime,
+    metadata: Object.keys(metadata).length > 0 ? metadata : void 0,
+    tokenUsage: msgTokenUsage
   };
 }
-function extractToolCalls3(content) {
+function extractToolCalls3(content, toolTrackers, completedToolResults) {
   if (!Array.isArray(content)) {
     return [];
   }
@@ -21055,10 +21151,17 @@ function extractToolCalls3(content) {
     }
     const p = part;
     if (p.type === "toolCall" && typeof p.name === "string") {
+      const id = typeof p.id === "string" ? p.id : void 0;
+      const tracker = id ? toolTrackers.get(id) : void 0;
+      const completed = id ? completedToolResults.get(id) : void 0;
       toolCalls.push({
         tool: p.name,
         input: p.arguments,
-        id: typeof p.id === "string" ? p.id : void 0
+        id,
+        output: completed?.output,
+        durationMs: completed?.durationMs,
+        startTime: tracker?.startTime,
+        endTime: tracker?.startTime && completed?.durationMs !== void 0 ? new Date(new Date(tracker.startTime).getTime() + completed.durationMs).toISOString() : void 0
       });
     }
   }
@@ -21590,14 +21693,14 @@ function extractTokenUsage(events) {
     const usage = record.usage;
     if (usage && typeof usage === "object") {
       const u = usage;
-      const input = toNumber(u.input_tokens ?? u.inputTokens ?? u.input);
-      const output = toNumber(u.output_tokens ?? u.outputTokens ?? u.output);
+      const input = toFiniteNumber(u.input_tokens ?? u.inputTokens ?? u.input);
+      const output = toFiniteNumber(u.output_tokens ?? u.outputTokens ?? u.output);
       if (input !== void 0 || output !== void 0) {
         const result = {
           input: input ?? 0,
           output: output ?? 0
         };
-        const cached = toNumber(u.cache_read_input_tokens ?? u.cached ?? u.cachedTokens);
+        const cached = toFiniteNumber(u.cache_read_input_tokens ?? u.cached ?? u.cachedTokens);
         if (cached !== void 0) {
           return { ...result, cached };
         }
@@ -21622,13 +21725,13 @@ function aggregateUsageFromMessages(messages) {
     const usage = m.usage;
     if (!usage || typeof usage !== "object") continue;
     const u = usage;
-    const input = toNumber(u.input_tokens ?? u.inputTokens ?? u.input);
-    const output = toNumber(u.output_tokens ?? u.outputTokens ?? u.output);
+    const input = toFiniteNumber(u.input_tokens ?? u.inputTokens ?? u.input);
+    const output = toFiniteNumber(u.output_tokens ?? u.outputTokens ?? u.output);
     if (input !== void 0 || output !== void 0) {
       found = true;
       totalInput += input ?? 0;
       totalOutput += output ?? 0;
-      const cached = toNumber(u.cache_read_input_tokens ?? u.cached ?? u.cachedTokens);
+      const cached = toFiniteNumber(u.cache_read_input_tokens ?? u.cached ?? u.cachedTokens);
       if (cached !== void 0) {
         totalCached = (totalCached ?? 0) + cached;
       }
@@ -21641,10 +21744,6 @@ function aggregateUsageFromMessages(messages) {
   }
   return result;
 }
-function toNumber(value) {
-  if (typeof value === "number" && Number.isFinite(value)) return value;
-  return void 0;
-}
 function convertPiMessage(message) {
   if (!message || typeof message !== "object") {
     return void 0;
@@ -21654,7 +21753,7 @@ function convertPiMessage(message) {
   if (typeof role !== "string") {
     return void 0;
   }
-  const content = extractTextContent4(msg.content);
+  const content = extractPiTextContent(msg.content);
   const toolCalls = extractToolCalls4(msg.content);
   const startTime = typeof msg.timestamp === "number" ? new Date(msg.timestamp).toISOString() : typeof msg.timestamp === "string" ? msg.timestamp : void 0;
   const metadata = {};
@@ -21671,25 +21770,6 @@ function convertPiMessage(message) {
     metadata: Object.keys(metadata).length > 0 ? metadata : void 0
   };
 }
-function extractTextContent4(content) {
-  if (typeof content === "string") {
-    return content;
-  }
-  if (!Array.isArray(content)) {
-    return void 0;
-  }
-  const textParts = [];
-  for (const part of content) {
-    if (!part || typeof part !== "object") {
-      continue;
-    }
-    const p = part;
-    if (p.type === "text" && typeof p.text === "string") {
-      textParts.push(p.text);
-    }
-  }
-  return textParts.length > 0 ? textParts.join("\n") : void 0;
-}
 function extractToolCalls4(content) {
   if (!Array.isArray(content)) {
     return [];
@@ -23868,7 +23948,11 @@ var CodeEvaluator = class {
       endTime: context2.endTime ?? null,
       fileChanges: context2.fileChanges ?? null,
       workspacePath: context2.workspacePath ?? null,
-      config: this.config ?? null
+      config: this.config ?? null,
+      // Text convenience accessors (new names, always strings)
+      inputText: context2.evalCase.question,
+      outputText: context2.candidate,
+      expectedOutputText: context2.evalCase.reference_answer ?? ""
     };
     const inputPayload = JSON.stringify(toSnakeCaseDeep(payload), null, 2);
     let proxyEnv;
@@ -24110,7 +24194,11 @@ var LlmGraderEvaluator = class {
       [TEMPLATE_VARIABLES.REFERENCE_ANSWER]: (context2.evalCase.reference_answer ?? "").trim(),
       [TEMPLATE_VARIABLES.CRITERIA]: context2.evalCase.criteria.trim(),
       [TEMPLATE_VARIABLES.QUESTION]: formattedQuestion.trim(),
-      [TEMPLATE_VARIABLES.FILE_CHANGES]: context2.fileChanges ?? ""
+      [TEMPLATE_VARIABLES.FILE_CHANGES]: context2.fileChanges ?? "",
+      // Text convenience accessors (new names, always strings)
+      [TEMPLATE_VARIABLES.INPUT_TEXT]: formattedQuestion.trim(),
+      [TEMPLATE_VARIABLES.OUTPUT_TEXT]: context2.candidate.trim(),
+      [TEMPLATE_VARIABLES.EXPECTED_OUTPUT_TEXT]: (context2.evalCase.reference_answer ?? "").trim()
     };
     const systemPrompt = buildOutputSchema();
     const evaluatorTemplate = context2.evaluatorTemplateOverride ?? this.evaluatorTemplate ?? DEFAULT_EVALUATOR_TEMPLATE;
@@ -25694,8 +25782,8 @@ var FieldAccuracyEvaluator = class {
    */
   compareNumericTolerance(path46, candidateValue, expectedValue, fieldConfig, weight) {
     const { tolerance = 0, relative = false } = fieldConfig;
-    const candidateNum = toNumber2(candidateValue);
-    const expectedNum = toNumber2(expectedValue);
+    const candidateNum = toNumber(candidateValue);
+    const expectedNum = toNumber(expectedValue);
     if (candidateNum === null || expectedNum === null) {
       return {
         path: path46,
@@ -25841,7 +25929,7 @@ function resolvePath(obj, path46) {
   }
   return current;
 }
-function toNumber2(value) {
+function toNumber(value) {
   if (typeof value === "number") {
     return value;
   }
@@ -25950,28 +26038,60 @@ var LatencyEvaluator = class {
     };
   }
 };
+var CLAUDE_MATCHER = {
+  skillTools: ["Skill"],
+  skillInputField: "skill",
+  readTools: ["Read"],
+  readInputField: "file_path"
+};
+var COPILOT_MATCHER = {
+  skillTools: ["Skill", "skill"],
+  skillInputField: "skill",
+  readTools: ["Read File", "readFile", "Read", "readTextFile"],
+  readInputField: "file_path"
+};
+var PROVIDER_TOOL_SEMANTICS = {
+  claude: CLAUDE_MATCHER,
+  "claude-cli": CLAUDE_MATCHER,
+  "claude-sdk": CLAUDE_MATCHER,
+  "pi-coding-agent": CLAUDE_MATCHER,
+  "pi-agent-sdk": CLAUDE_MATCHER,
+  "copilot-cli": COPILOT_MATCHER,
+  "copilot-sdk": COPILOT_MATCHER,
+  vscode: COPILOT_MATCHER,
+  "vscode-insiders": COPILOT_MATCHER
+};
 var SkillTriggerEvaluator = class {
   kind = "skill-trigger";
   config;
   constructor(config) {
     this.config = config;
   }
+  resolveMatcher(providerKind) {
+    if (providerKind) {
+      const match = PROVIDER_TOOL_SEMANTICS[providerKind];
+      if (match) return match;
+    }
+    return CLAUDE_MATCHER;
+  }
   evaluate(context2) {
     const skillName = this.config.skill;
     const shouldTrigger = this.config.should_trigger !== false;
+    const providerKind = context2.provider?.kind;
+    const matcher = this.resolveMatcher(providerKind);
     const firstTool = (context2.output ?? []).flatMap((msg) => msg.toolCalls ?? [])[0];
     let triggered = false;
     let evidence = "";
     if (firstTool) {
       const input = firstTool.input ?? {};
-      if (firstTool.tool === "Skill") {
-        const skillArg = String(input.skill ?? "");
+      if (matcher.skillTools.includes(firstTool.tool)) {
+        const skillArg = String(input[matcher.skillInputField] ?? "");
         if (skillArg.includes(skillName)) {
           triggered = true;
-          evidence = `Skill tool invoked with skill="${skillArg}"`;
+          evidence = `Skill tool invoked with ${matcher.skillInputField}="${skillArg}"`;
         }
-      } else if (firstTool.tool === "Read") {
-        const filePath = String(input.file_path ?? "");
+      } else if (matcher.readTools.includes(firstTool.tool)) {
+        const filePath = String(input[matcher.readInputField] ?? "");
         if (filePath.includes(skillName)) {
           triggered = true;
           evidence = `Read tool loaded skill file: ${filePath}`;
@@ -25996,7 +26116,7 @@ var SkillTriggerEvaluator = class {
       verdict: "fail",
       hits: [],
       misses: [
-        shouldTrigger ? firstTool ? `First tool was "${firstTool.tool}" \u2014 not Skill/Read for "${skillName}"` : "No tool calls recorded" : evidence || `Skill "${skillName}" triggered unexpectedly`
+        shouldTrigger ? firstTool ? `First tool was "${firstTool.tool}" \u2014 not a skill/read tool for "${skillName}"` : "No tool calls recorded" : evidence || `Skill "${skillName}" triggered unexpectedly`
       ],
       expectedAspectCount: 1,
       reasoning: shouldTrigger ? `Skill "${skillName}" was not triggered` : "False trigger: skill fired when it should not have"
@@ -26038,7 +26158,11 @@ function assembleFreeform(evalCase, candidate, promptInputs, fileChanges, evalua
     [TEMPLATE_VARIABLES.REFERENCE_ANSWER]: (evalCase.reference_answer ?? "").trim(),
     [TEMPLATE_VARIABLES.CRITERIA]: evalCase.criteria.trim(),
     [TEMPLATE_VARIABLES.QUESTION]: formattedQuestion.trim(),
-    [TEMPLATE_VARIABLES.FILE_CHANGES]: fileChanges ?? ""
+    [TEMPLATE_VARIABLES.FILE_CHANGES]: fileChanges ?? "",
+    // Text convenience accessors (new names, always strings)
+    [TEMPLATE_VARIABLES.INPUT_TEXT]: formattedQuestion.trim(),
+    [TEMPLATE_VARIABLES.OUTPUT_TEXT]: candidate.trim(),
+    [TEMPLATE_VARIABLES.EXPECTED_OUTPUT_TEXT]: (evalCase.reference_answer ?? "").trim()
   };
   const systemPrompt = buildOutputSchema();
   const template = evaluatorTemplateOverride ?? DEFAULT_EVALUATOR_TEMPLATE;
@@ -27029,7 +27153,11 @@ async function executePromptTemplate(script, context2, config, timeoutMs) {
     trace: context2.trace ?? null,
     fileChanges: context2.fileChanges ?? null,
     workspacePath: context2.workspacePath ?? null,
-    config: config ?? context2.config ?? null
+    config: config ?? context2.config ?? null,
+    // Text convenience accessors (new names, always strings)
+    inputText: context2.evalCase.question,
+    outputText: context2.candidate,
+    expectedOutputText: context2.evalCase.reference_answer ?? ""
   };
   const inputJson = JSON.stringify(toSnakeCaseDeep(payload), null, 2);
   const scriptPath = script[script.length - 1];
@@ -28689,7 +28817,9 @@ async function runEvaluation(options) {
               testId: evalCase.id,
               status: "failed",
               completedAt: Date.now(),
-              error: budgetResult.error
+              error: budgetResult.error,
+              score: budgetResult.score,
+              executionStatus: budgetResult.executionStatus
             });
           }
           if (onResult) {
@@ -28720,7 +28850,9 @@ async function runEvaluation(options) {
               testId: evalCase.id,
               status: "failed",
               completedAt: Date.now(),
-              error: haltResult.error
+              error: haltResult.error,
+              score: haltResult.score,
+              executionStatus: haltResult.executionStatus
             });
           }
           if (onResult) {
@@ -28800,7 +28932,9 @@ async function runEvaluation(options) {
               startedAt: 0,
               // Not used for completed status
               completedAt: Date.now(),
-              error: result.error
+              error: result.error,
+              score: result.score,
+              executionStatus: result.executionStatus
             });
           }
           if (onResult) {
@@ -28971,7 +29105,9 @@ async function runBatchEvaluation(options) {
     const merged = computed ? mergeExecutionMetrics(computed, {
       tokenUsage: providerResponse.tokenUsage,
       costUsd: providerResponse.costUsd,
-      durationMs: providerResponse.durationMs
+      durationMs: providerResponse.durationMs,
+      startTime: providerResponse.startTime,
+      endTime: providerResponse.endTime
     }) : void 0;
     const trace2 = merged?.trace;
     const costUsd = merged?.costUsd;
@@ -29036,7 +29172,9 @@ async function runBatchEvaluation(options) {
           testId: evalCase.id,
           status: "failed",
           completedAt: Date.now(),
-          error: error instanceof Error ? error.message : String(error)
+          error: error instanceof Error ? error.message : String(error),
+          score: errorResult.score,
+          executionStatus: errorResult.executionStatus
         });
       }
       continue;
@@ -29052,7 +29190,9 @@ async function runBatchEvaluation(options) {
         status: result.error ? "failed" : "completed",
         startedAt: 0,
         completedAt: Date.now(),
-        error: result.error
+        error: result.error,
+        score: result.score,
+        executionStatus: result.executionStatus
       });
     }
   }
@@ -29362,7 +29502,9 @@ async function runEvalCase(options) {
   const merged = computed ? mergeExecutionMetrics(computed, {
     tokenUsage: providerResponse.tokenUsage,
     costUsd: providerResponse.costUsd,
-    durationMs: providerResponse.durationMs
+    durationMs: providerResponse.durationMs,
+    startTime: providerResponse.startTime,
+    endTime: providerResponse.endTime
   }) : void 0;
   const trace2 = merged?.trace;
   const costUsd = merged?.costUsd;
@@ -31148,4 +31290,4 @@ export {
   OtelStreamingObserver,
   createAgentKernel
 };
-//# sourceMappingURL=chunk-6XTYVCMN.js.map
+//# sourceMappingURL=chunk-GOZV2HN2.js.map