@agentv/core 0.15.0 → 0.17.0

package/dist/index.d.cts

@@ -101,7 +101,7 @@ interface EvalCase {
     readonly question: string;
     readonly input_messages: readonly TestMessage[];
     readonly input_segments: readonly JsonObject[];
-    readonly output_segments: readonly JsonObject[];
+    readonly expected_segments: readonly JsonObject[];
     readonly reference_answer?: string;
     readonly guideline_paths: readonly string[];
     readonly guideline_patterns?: readonly string[];
@@ -262,6 +262,17 @@ interface TargetDefinition {
     readonly retryStatusCodes?: unknown | undefined;
 }
 
+/**
+ * Formatting mode for segment content.
+ * - 'agent': File references only (for providers with filesystem access)
+ * - 'lm': Embedded file content with XML tags (for language model providers)
+ */
+type FormattingMode = "agent" | "lm";
+/**
+ * Extract fenced code blocks from AgentV user segments.
+ */
+declare function extractCodeBlocks(segments: readonly JsonObject[]): readonly string[];
+
 /**
  * Build prompt inputs by consolidating user request context and guideline content.
  */
@@ -271,12 +282,13 @@ interface PromptInputs {
     readonly chatPrompt?: ChatPrompt;
     readonly systemMessage?: string;
 }
-declare function buildPromptInputs(testCase: EvalCase): Promise<PromptInputs>;
-
 /**
- * Extract fenced code blocks from AgentV user segments.
+ * Build prompt inputs by consolidating user request context and guideline content.
+ *
+ * @param testCase - The evaluation test case
+ * @param mode - Formatting mode: 'agent' for file references, 'lm' for embedded content (default: 'lm')
  */
-declare function extractCodeBlocks(segments: readonly JsonObject[]): readonly string[];
+declare function buildPromptInputs(testCase: EvalCase, mode?: FormattingMode): Promise<PromptInputs>;
 
 /**
  * Determine whether a path references guideline content (instructions or prompts).
@@ -605,6 +617,7 @@ interface RunEvaluationOptions {
     readonly evalId?: string;
     readonly verbose?: boolean;
     readonly maxConcurrency?: number;
+    readonly evalCases?: readonly EvalCase[];
     readonly onResult?: (result: EvaluationResult) => MaybePromise<void>;
     readonly onProgress?: (event: ProgressEvent) => MaybePromise<void>;
 }

package/dist/index.d.ts

@@ -101,7 +101,7 @@ interface EvalCase {
     readonly question: string;
     readonly input_messages: readonly TestMessage[];
     readonly input_segments: readonly JsonObject[];
-    readonly output_segments: readonly JsonObject[];
+    readonly expected_segments: readonly JsonObject[];
     readonly reference_answer?: string;
     readonly guideline_paths: readonly string[];
     readonly guideline_patterns?: readonly string[];
@@ -262,6 +262,17 @@ interface TargetDefinition {
     readonly retryStatusCodes?: unknown | undefined;
 }
 
+/**
+ * Formatting mode for segment content.
+ * - 'agent': File references only (for providers with filesystem access)
+ * - 'lm': Embedded file content with XML tags (for language model providers)
+ */
+type FormattingMode = "agent" | "lm";
+/**
+ * Extract fenced code blocks from AgentV user segments.
+ */
+declare function extractCodeBlocks(segments: readonly JsonObject[]): readonly string[];
+
 /**
  * Build prompt inputs by consolidating user request context and guideline content.
  */
@@ -271,12 +282,13 @@ interface PromptInputs {
     readonly chatPrompt?: ChatPrompt;
     readonly systemMessage?: string;
 }
-declare function buildPromptInputs(testCase: EvalCase): Promise<PromptInputs>;
-
 /**
- * Extract fenced code blocks from AgentV user segments.
+ * Build prompt inputs by consolidating user request context and guideline content.
+ *
+ * @param testCase - The evaluation test case
+ * @param mode - Formatting mode: 'agent' for file references, 'lm' for embedded content (default: 'lm')
  */
-declare function extractCodeBlocks(segments: readonly JsonObject[]): readonly string[];
+declare function buildPromptInputs(testCase: EvalCase, mode?: FormattingMode): Promise<PromptInputs>;
 
 /**
  * Determine whether a path references guideline content (instructions or prompts).
@@ -605,6 +617,7 @@ interface RunEvaluationOptions {
     readonly evalId?: string;
     readonly verbose?: boolean;
     readonly maxConcurrency?: number;
+    readonly evalCases?: readonly EvalCase[];
     readonly onResult?: (result: EvaluationResult) => MaybePromise<void>;
     readonly onProgress?: (event: ProgressEvent) => MaybePromise<void>;
 }