@wix/evalforge-evaluator 0.183.0 → 0.184.0

package/build/types/run-scenario/agents/claude-code/execute.d.ts

@@ -1,11 +1,12 @@
 import type { SkillWithLatestVersion, TestScenario, LLMTrace, ConversationMessage, TriggerPromptImage } from '@wix/evalforge-types';
+import type { CapturedStep } from '../../types.js';
 import type { ClaudeCodeExecutionOptions, ClaudeCodeExecutionResult } from './types.js';
 /**
  * Import SDK types directly from Claude Agent SDK.
  * Type-only imports are erased at compile time - zero runtime overhead.
  * The SDK is still dynamically imported at runtime in executeWithClaudeCode().
  */
-import type { SDKUserMessage, SDKMessage } from '@anthropic-ai/claude-agent-sdk' with { 'resolution-mode': 'import' };
+import type { SDKResultMessage, SDKUserMessage, SDKMessage } from '@anthropic-ai/claude-agent-sdk' with { 'resolution-mode': 'import' };
 /**
  * Message with timestamp — tracks when each message was received.
  */
@@ -46,3 +47,24 @@ export declare function executeWithClaudeCode(skills: SkillWithLatestVersion[],
     llmTrace: LLMTrace;
     conversation: ConversationMessage[];
 }>;
+/**
+ * Process SDK messages into CapturedSteps for LLM trace building.
+ * Uses actual timestamps from when messages were received to calculate durations.
+ */
+export declare function processMessages(timestampedMessages: TimestampedMessage[], startTime: Date, endTime: Date): {
+    steps: CapturedStep[];
+    result?: SDKResultMessage;
+};
+/**
+ * Build LLM trace from captured steps.
+ * Calculates per-step costs using model pricing and sums tokens from steps as fallback.
+ */
+export declare function buildLLMTraceFromSteps(steps: CapturedStep[], totalDurationMs: number, usage: {
+    inputTokens: number;
+    outputTokens: number;
+    totalTokens: number;
+    costUsd?: number;
+    cacheReadTokens?: number;
+    cacheWriteTokens?: number;
+    durationApiMs?: number;
+}, model: string): LLMTrace;

package/build/types/run-scenario/types.d.ts

@@ -21,6 +21,8 @@ export interface CapturedStep {
         toolName: string;
         toolUseId?: string;
         args: unknown;
+        isError?: boolean;
+        errorContent?: string;
     }>;
     toolResults?: unknown[];
     /** True if any tool result for this step's tool calls was an error */

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@wix/evalforge-evaluator",
-  "version": "0.183.0",
+  "version": "0.184.0",
   "description": "EvalForge Evaluator",
   "bin": "./build/index.js",
   "files": [
@@ -71,5 +71,5 @@
       "artifactId": "evalforge-evaluator"
     }
   },
-  "falconPackageHash": "29debc912e78ca8ff33cbeeafe55eca432be7646d7ccc604f9a2f57b"
+  "falconPackageHash": "e51797f0d074a5b087399ed74f317c6a01d157d49801b14659740125"
 }