@wix/evalforge-evaluator 0.102.0 → 0.103.0

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

@@ -0,0 +1,8 @@
+import type { ConversationMessage } from '@wix/evalforge-types';
+import type { TimestampedMessage } from './execute.js';
+/**
+ * Build a full conversation from raw SDK messages.
+ * Extracts text, thinking, tool_use, and tool_result blocks into a
+ * serializable ConversationMessage array for storage and later display.
+ */
+export declare function buildConversation(timestampedMessages: TimestampedMessage[]): ConversationMessage[];

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

@@ -1,5 +1,18 @@
-import type { SkillWithLatestVersion, TestScenario, LLMTrace } from '@wix/evalforge-types';
+import type { SkillWithLatestVersion, TestScenario, LLMTrace, ConversationMessage } from '@wix/evalforge-types';
 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 { SDKMessage } from '@anthropic-ai/claude-agent-sdk' with { 'resolution-mode': 'import' };
+/**
+ * Message with timestamp — tracks when each message was received.
+ */
+export interface TimestampedMessage {
+    message: SDKMessage;
+    receivedAt: Date;
+}
 /**
  * Execute skills using the Claude Agent SDK.
  *
@@ -15,4 +28,5 @@ import type { ClaudeCodeExecutionOptions, ClaudeCodeExecutionResult } from './ty
 export declare function executeWithClaudeCode(skills: SkillWithLatestVersion[], scenario: TestScenario, options: ClaudeCodeExecutionOptions): Promise<{
     result: ClaudeCodeExecutionResult;
     llmTrace: LLMTrace;
+    conversation: ConversationMessage[];
 }>;

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

@@ -10,5 +10,6 @@
  */
 export { ClaudeCodeAdapter, claudeCodeAdapter } from './claude-code-adapter.js';
 export { executeWithClaudeCode } from './execute.js';
+export { buildConversation } from './build-conversation.js';
 export type { ClaudeCodeExecutionOptions, ClaudeCodeExecutionResult } from './types.js';
 export type { TraceContext } from '@wix/evalforge-types';

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

@@ -8,6 +8,7 @@ export type PartialEvalRunResult = Omit<EvalRunResult, 'assertionResults' | 'pas
  */
 export interface CapturedStep {
     text: string;
+    thinking?: string;
     usage: {
         inputTokens: number;
         outputTokens: number;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@wix/evalforge-evaluator",
-  "version": "0.102.0",
+  "version": "0.103.0",
   "description": "EvalForge Evaluator",
   "bin": "./build/index.js",
   "files": [
@@ -20,8 +20,8 @@
     "@anthropic-ai/claude-agent-sdk": "^0.2.44",
     "@anthropic-ai/claude-code": "^2.1.44",
     "@wix/eval-assertions": "0.22.0",
-    "@wix/evalforge-github-client": "0.20.0",
-    "@wix/evalforge-types": "0.45.0",
+    "@wix/evalforge-github-client": "0.21.0",
+    "@wix/evalforge-types": "0.46.0",
     "ai": "^6.0.6",
     "diff": "^7.0.0",
     "tar": "^7.5.3",
@@ -60,5 +60,5 @@
       "artifactId": "evalforge-evaluator"
     }
   },
-  "falconPackageHash": "db973c328d16212a1aea0a801b5e3d93245f0611fceadb4a64f7d87e"
+  "falconPackageHash": "770ab064ab729180951ced129aaf6e04a6d3bdf65e4cc014a353566e"
 }