@agentv/core 0.7.2 → 0.7.3

package/dist/index.d.cts

@@ -95,7 +95,7 @@ type LlmJudgeEvaluatorConfig = {
 };
 type EvaluatorConfig = CodeEvaluatorConfig | LlmJudgeEvaluatorConfig;
 /**
- * Test case definition sourced from AgentV specs.
+ * Eval case definition sourced from AgentV specs.
  */
 interface EvalCase {
     readonly id: string;
@@ -104,7 +104,6 @@ interface EvalCase {
     readonly question: string;
     readonly input_segments: readonly JsonObject[];
     readonly output_segments: readonly JsonObject[];
-    readonly system_message?: string;
     readonly reference_answer: string;
     readonly guideline_paths: readonly string[];
     readonly guideline_patterns?: readonly string[];
@@ -115,7 +114,7 @@ interface EvalCase {
     readonly evaluators?: readonly EvaluatorConfig[];
 }
 /**
- * Evaluator scorecard for a single test case run.
+ * Evaluator scorecard for a single eval case run.
  */
 interface EvaluationResult {
     readonly eval_id: string;
@@ -159,6 +158,7 @@ declare function isGuidelineFile(filePath: string, patterns?: readonly string[])
 declare function extractCodeBlocks(segments: readonly JsonObject[]): readonly string[];
 type LoadOptions = {
     readonly verbose?: boolean;
+    readonly evalId?: string;
 };
 /**
  * Load eval cases from a AgentV YAML specification file.

package/dist/index.d.ts

@@ -95,7 +95,7 @@ type LlmJudgeEvaluatorConfig = {
 };
 type EvaluatorConfig = CodeEvaluatorConfig | LlmJudgeEvaluatorConfig;
 /**
- * Test case definition sourced from AgentV specs.
+ * Eval case definition sourced from AgentV specs.
  */
 interface EvalCase {
     readonly id: string;
@@ -104,7 +104,6 @@ interface EvalCase {
     readonly question: string;
     readonly input_segments: readonly JsonObject[];
     readonly output_segments: readonly JsonObject[];
-    readonly system_message?: string;
     readonly reference_answer: string;
     readonly guideline_paths: readonly string[];
     readonly guideline_patterns?: readonly string[];
@@ -115,7 +114,7 @@ interface EvalCase {
     readonly evaluators?: readonly EvaluatorConfig[];
 }
 /**
- * Evaluator scorecard for a single test case run.
+ * Evaluator scorecard for a single eval case run.
  */
 interface EvaluationResult {
     readonly eval_id: string;
@@ -159,6 +158,7 @@ declare function isGuidelineFile(filePath: string, patterns?: readonly string[])
 declare function extractCodeBlocks(segments: readonly JsonObject[]): readonly string[];
 type LoadOptions = {
     readonly verbose?: boolean;
+    readonly evalId?: string;
 };
 /**
  * Load eval cases from a AgentV YAML specification file.

package/dist/index.js

@@ -218,6 +218,7 @@ async function processMessages(options) {
 }
 async function loadEvalCases(evalFilePath, repoRoot, options) {
   const verbose = options?.verbose ?? false;
+  const evalIdFilter = options?.evalId;
   const absoluteTestPath = path.resolve(evalFilePath);
   if (!await fileExists2(absoluteTestPath)) {
     throw new Error(`Test file not found: ${evalFilePath}`);
@@ -249,62 +250,39 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
   const results = [];
   for (const rawEvalcase of rawTestcases) {
     if (!isJsonObject(rawEvalcase)) {
-      logWarning("Skipping invalid test case entry (expected object)");
+      logWarning("Skipping invalid eval case entry (expected object)");
       continue;
     }
     const evalcase = rawEvalcase;
     const id = asString(evalcase.id);
+    if (evalIdFilter && id !== evalIdFilter) {
+      continue;
+    }
     const conversationId = asString(evalcase.conversation_id);
     const outcome = asString(evalcase.outcome);
     const inputMessagesValue = evalcase.input_messages;
     const expectedMessagesValue = evalcase.expected_messages;
     if (!id || !outcome || !Array.isArray(inputMessagesValue)) {
-      logWarning(`Skipping incomplete test case: ${id ?? "unknown"}`);
+      logWarning(`Skipping incomplete eval case: ${id ?? "unknown"}`);
       continue;
     }
     if (!Array.isArray(expectedMessagesValue)) {
-      logWarning(`Test case '${id}' missing expected_messages array`);
+      logWarning(`Eval case '${id}' missing expected_messages array`);
       continue;
     }
     const inputMessages = inputMessagesValue.filter((msg) => isTestMessage(msg));
     const expectedMessages = expectedMessagesValue.filter((msg) => isTestMessage(msg));
-    const assistantMessages = expectedMessages.filter((message) => message.role === "assistant");
-    const userMessages = inputMessages.filter((message) => message.role === "user");
-    const systemMessages = inputMessages.filter((message) => message.role === "system");
-    if (assistantMessages.length === 0) {
-      logWarning(`No assistant message found for test case: ${id}`);
+    if (expectedMessages.length === 0) {
+      logWarning(`No expected message found for eval case: ${id}`);
       continue;
     }
-    if (assistantMessages.length > 1) {
-      logWarning(`Multiple assistant messages found for test case: ${id}, using first`);
-    }
-    if (systemMessages.length > 1) {
-      logWarning(`Multiple system messages found for test case: ${id}, using first`);
-    }
-    let systemMessageContent;
-    if (systemMessages.length > 0) {
-      const content = systemMessages[0]?.content;
-      if (typeof content === "string") {
-        systemMessageContent = content;
-      } else if (Array.isArray(content)) {
-        const textParts = [];
-        for (const segment of content) {
-          if (isJsonObject(segment)) {
-            const value = segment.value;
-            if (typeof value === "string") {
-              textParts.push(value);
-            }
-          }
-        }
-        if (textParts.length > 0) {
-          systemMessageContent = textParts.join("\n\n");
-        }
-      }
+    if (expectedMessages.length > 1) {
+      logWarning(`Multiple expected messages found for eval case: ${id}, using first`);
     }
     const guidelinePaths = [];
     const inputTextParts = [];
     const inputSegments = await processMessages({
-      messages: userMessages,
+      messages: inputMessages,
       searchRoots,
       repoRootPath,
       guidelinePatterns,
@@ -314,7 +292,7 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
       verbose
     });
     const outputSegments = await processMessages({
-      messages: assistantMessages,
+      messages: expectedMessages,
       searchRoots,
       repoRootPath,
       guidelinePatterns,
@@ -322,10 +300,10 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
       verbose
     });
     const codeSnippets = extractCodeBlocks(inputSegments);
-    const assistantContent = assistantMessages[0]?.content;
-    const referenceAnswer = await resolveAssistantContent(assistantContent, searchRoots, verbose);
+    const expectedContent = expectedMessages[0]?.content;
+    const referenceAnswer = await resolveAssistantContent(expectedContent, searchRoots, verbose);
     const question = inputTextParts.map((part) => part.trim()).filter((part) => part.length > 0).join(" ");
-    const testCaseEvaluatorKind = coerceEvaluator(evalcase.evaluator, id) ?? globalEvaluator;
+    const evalCaseEvaluatorKind = coerceEvaluator(evalcase.evaluator, id) ?? globalEvaluator;
     const evaluators = await parseEvaluators(evalcase, searchRoots, id ?? "unknown");
     const userFilePaths = [];
     for (const segment of inputSegments) {
@@ -344,19 +322,18 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
       question,
       input_segments: inputSegments,
       output_segments: outputSegments,
-      system_message: systemMessageContent,
       reference_answer: referenceAnswer,
       guideline_paths: guidelinePaths.map((guidelinePath) => path.resolve(guidelinePath)),
       guideline_patterns: guidelinePatterns,
       file_paths: allFilePaths,
       code_snippets: codeSnippets,
       expected_outcome: outcome,
-      evaluator: testCaseEvaluatorKind,
+      evaluator: evalCaseEvaluatorKind,
       evaluators
     };
     if (verbose) {
       console.log(`
-[Test Case: ${id}]`);
+[Eval Case: ${id}]`);
       if (testCase.guideline_paths.length > 0) {
         console.log(`  Guidelines used: ${testCase.guideline_paths.length}`);
         for (const guidelinePath of testCase.guideline_paths) {
@@ -415,7 +392,7 @@ ${body}`);
   }
   const question = questionParts.map((part) => part.trim()).filter((part) => part.length > 0).join("\n\n");
   const guidelines = guidelineContents.map((part) => part.trim()).filter((part) => part.length > 0).join("\n\n");
-  return { question, guidelines, systemMessage: testCase.system_message };
+  return { question, guidelines };
 }
 async function fileExists2(absolutePath) {
   try {
@@ -2918,7 +2895,6 @@ var CodeEvaluator = class {
         expected_outcome: context.evalCase.expected_outcome,
         reference_answer: context.evalCase.reference_answer,
         candidate_answer: context.candidate,
-        system_message: context.promptInputs.systemMessage ?? "",
         guideline_paths: context.evalCase.guideline_paths,
         input_files: context.evalCase.file_paths,
         input_segments: context.evalCase.input_segments
@@ -3160,7 +3136,7 @@ function validateConcurrency(concurrency) {
 // src/evaluation/orchestrator.ts
 async function runEvaluation(options) {
   const {
-    testFilePath,
+    testFilePath: evalFilePath,
     repoRoot,
     target,
     targets,
@@ -3179,11 +3155,11 @@ async function runEvaluation(options) {
     onProgress
   } = options;
   const load = loadEvalCases;
-  const evalCases = await load(testFilePath, repoRoot, { verbose });
+  const evalCases = await load(evalFilePath, repoRoot, { verbose, evalId });
   const filteredEvalCases = filterEvalCases(evalCases, evalId);
   if (filteredEvalCases.length === 0) {
     if (evalId) {
-      throw new Error(`Test case with id '${evalId}' not found in ${testFilePath}`);
+      throw new Error(`Eval case with id '${evalId}' not found in ${evalFilePath}`);
     }
     return [];
   }
@@ -3562,8 +3538,7 @@ async function evaluateCandidate(options) {
   const rawRequest = {
     question: promptInputs.question,
     ...isAgentProvider(provider) ? {} : { guidelines: promptInputs.guidelines },
-    guideline_paths: evalCase.guideline_paths,
-    system_message: promptInputs.systemMessage ?? ""
+    guideline_paths: evalCase.guideline_paths
   };
   return {
     eval_id: evalCase.id,
@@ -3827,7 +3802,6 @@ function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs,
     question: promptInputs.question,
     ...isAgentProvider(provider) ? {} : { guidelines: promptInputs.guidelines },
     guideline_paths: evalCase.guideline_paths,
-    system_message: promptInputs.systemMessage ?? "",
     error: message
   };
   return {