@agentv/core 2.1.1 → 2.2.0

package/dist/index.cjs

@@ -53,6 +53,7 @@ __export(index_exports, {
   createAgentKernel: () => createAgentKernel,
   createProvider: () => createProvider,
   deepEqual: () => deepEqual,
+  detectFormat: () => detectFormat,
   ensureVSCodeSubagents: () => ensureVSCodeSubagents,
   executeScript: () => executeScript,
   explorationRatio: () => explorationRatio,
@@ -226,9 +227,9 @@ function mergeExecutionMetrics(summary, metrics) {
 }
 
 // src/evaluation/yaml-parser.ts
-var import_promises6 = require("fs/promises");
-var import_node_path6 = __toESM(require("path"), 1);
-var import_yaml2 = require("yaml");
+var import_promises7 = require("fs/promises");
+var import_node_path7 = __toESM(require("path"), 1);
+var import_yaml3 = require("yaml");
 
 // src/evaluation/loaders/config-loader.ts
 var import_promises2 = require("fs/promises");
@@ -1006,6 +1007,11 @@ function isValidFieldAggregationType(value) {
   return typeof value === "string" && VALID_FIELD_AGGREGATION_TYPES.has(value);
 }
 
+// src/evaluation/loaders/jsonl-parser.ts
+var import_promises5 = require("fs/promises");
+var import_node_path5 = __toESM(require("path"), 1);
+var import_yaml2 = require("yaml");
+
 // src/evaluation/loaders/message-processor.ts
 var import_promises4 = require("fs/promises");
 var import_node_path4 = __toESM(require("path"), 1);
@@ -1266,28 +1272,271 @@ async function processExpectedMessages(options) {
   return segments;
 }
 
-// src/evaluation/formatting/prompt-builder.ts
-var import_promises5 = require("fs/promises");
-var import_node_path5 = __toESM(require("path"), 1);
+// src/evaluation/loaders/jsonl-parser.ts
 var ANSI_YELLOW5 = "\x1B[33m";
+var ANSI_RED = "\x1B[31m";
 var ANSI_RESET5 = "\x1B[0m";
+function detectFormat(filePath) {
+  const ext = import_node_path5.default.extname(filePath).toLowerCase();
+  if (ext === ".jsonl") return "jsonl";
+  if (ext === ".yaml" || ext === ".yml") return "yaml";
+  throw new Error(`Unsupported file format: '${ext}'. Supported formats: .yaml, .yml, .jsonl`);
+}
+async function loadSidecarMetadata(jsonlPath, verbose) {
+  const dir = import_node_path5.default.dirname(jsonlPath);
+  const base = import_node_path5.default.basename(jsonlPath, ".jsonl");
+  const sidecarPath = import_node_path5.default.join(dir, `${base}.yaml`);
+  if (!await fileExists(sidecarPath)) {
+    if (verbose) {
+      logWarning4(`Sidecar metadata file not found: ${sidecarPath} (using defaults)`);
+    }
+    return {};
+  }
+  try {
+    const content = await (0, import_promises5.readFile)(sidecarPath, "utf8");
+    const parsed = (0, import_yaml2.parse)(content);
+    if (!isJsonObject(parsed)) {
+      logWarning4(`Invalid sidecar metadata format in ${sidecarPath}`);
+      return {};
+    }
+    return {
+      description: asString4(parsed.description),
+      dataset: asString4(parsed.dataset),
+      execution: isJsonObject(parsed.execution) ? parsed.execution : void 0,
+      evaluator: parsed.evaluator
+    };
+  } catch (error) {
+    logWarning4(`Could not read sidecar metadata from ${sidecarPath}: ${error.message}`);
+    return {};
+  }
+}
+function parseJsonlContent(content, filePath) {
+  const lines = content.split("\n");
+  const cases = [];
+  for (let i = 0; i < lines.length; i++) {
+    const line = lines[i].trim();
+    if (line === "") continue;
+    try {
+      const parsed = JSON.parse(line);
+      if (!isJsonObject(parsed)) {
+        throw new Error("Expected JSON object");
+      }
+      cases.push(parsed);
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error);
+      throw new Error(`Line ${i + 1}: Invalid JSON - ${message}
+  File: ${filePath}`);
+    }
+  }
+  return cases;
+}
+async function loadEvalCasesFromJsonl(evalFilePath, repoRoot, options) {
+  const verbose = options?.verbose ?? false;
+  const evalIdFilter = options?.evalId;
+  const absoluteTestPath = import_node_path5.default.resolve(evalFilePath);
+  const repoRootPath = resolveToAbsolutePath(repoRoot);
+  const searchRoots = buildSearchRoots(absoluteTestPath, repoRootPath);
+  const config = await loadConfig(absoluteTestPath, repoRootPath);
+  const guidelinePatterns = config?.guideline_patterns;
+  const sidecar = await loadSidecarMetadata(absoluteTestPath, verbose);
+  const rawFile = await (0, import_promises5.readFile)(absoluteTestPath, "utf8");
+  const rawCases = parseJsonlContent(rawFile, evalFilePath);
+  const fallbackDataset = import_node_path5.default.basename(absoluteTestPath, ".jsonl") || "eval";
+  const datasetName = sidecar.dataset && sidecar.dataset.trim().length > 0 ? sidecar.dataset : fallbackDataset;
+  const globalEvaluator = coerceEvaluator(sidecar.evaluator, "sidecar") ?? "llm_judge";
+  const globalExecution = sidecar.execution;
+  if (verbose) {
+    console.log(`
+[JSONL Dataset: ${evalFilePath}]`);
+    console.log(`  Cases: ${rawCases.length}`);
+    console.log(`  Dataset name: ${datasetName}`);
+    if (sidecar.description) {
+      console.log(`  Description: ${sidecar.description}`);
+    }
+  }
+  const results = [];
+  for (let lineIndex = 0; lineIndex < rawCases.length; lineIndex++) {
+    const evalcase = rawCases[lineIndex];
+    const lineNumber = lineIndex + 1;
+    const id = asString4(evalcase.id);
+    if (evalIdFilter && id !== evalIdFilter) {
+      continue;
+    }
+    const conversationId = asString4(evalcase.conversation_id);
+    const outcome = asString4(evalcase.expected_outcome) ?? asString4(evalcase.outcome);
+    const inputMessagesValue = evalcase.input_messages;
+    const expectedMessagesValue = evalcase.expected_messages;
+    if (!id || !outcome || !Array.isArray(inputMessagesValue)) {
+      logError(
+        `Skipping incomplete eval case at line ${lineNumber}: ${id ?? "unknown"}. Missing required fields: id, expected_outcome, and/or input_messages`
+      );
+      continue;
+    }
+    const hasExpectedMessages = Array.isArray(expectedMessagesValue) && expectedMessagesValue.length > 0;
+    const inputMessages = inputMessagesValue.filter(
+      (msg) => isTestMessage(msg)
+    );
+    const expectedMessages = hasExpectedMessages ? expectedMessagesValue.filter((msg) => isTestMessage(msg)) : [];
+    if (hasExpectedMessages && expectedMessages.length === 0) {
+      logError(`Line ${lineNumber}: No valid expected message found for eval case: ${id}`);
+      continue;
+    }
+    const guidelinePaths = [];
+    const inputTextParts = [];
+    const inputSegments = await processMessages({
+      messages: inputMessages,
+      searchRoots,
+      repoRootPath,
+      guidelinePatterns,
+      guidelinePaths,
+      textParts: inputTextParts,
+      messageType: "input",
+      verbose
+    });
+    const outputSegments = hasExpectedMessages ? await processExpectedMessages({
+      messages: expectedMessages,
+      searchRoots,
+      repoRootPath,
+      verbose
+    }) : [];
+    let referenceAnswer = "";
+    if (outputSegments.length > 0) {
+      const lastMessage = outputSegments[outputSegments.length - 1];
+      const content = lastMessage.content;
+      const toolCalls = lastMessage.tool_calls;
+      if (typeof content === "string") {
+        referenceAnswer = content;
+      } else if (content !== void 0 && content !== null) {
+        referenceAnswer = JSON.stringify(content, null, 2);
+      } else if (toolCalls !== void 0 && toolCalls !== null) {
+        referenceAnswer = JSON.stringify(toolCalls, null, 2);
+      }
+    }
+    const question = inputTextParts.map((part) => part.trim()).filter((part) => part.length > 0).join(" ");
+    const caseExecution = isJsonObject(evalcase.execution) ? evalcase.execution : void 0;
+    const mergedExecution = caseExecution ?? globalExecution;
+    const evalCaseEvaluatorKind = coerceEvaluator(evalcase.evaluator, id) ?? globalEvaluator;
+    let evaluators;
+    try {
+      evaluators = await parseEvaluators(evalcase, mergedExecution, searchRoots, id ?? "unknown");
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error);
+      logError(`Skipping eval case '${id}' at line ${lineNumber}: ${message}`);
+      continue;
+    }
+    const inlineRubrics = evalcase.rubrics;
+    if (inlineRubrics !== void 0 && Array.isArray(inlineRubrics)) {
+      const rubricItems = inlineRubrics.filter((r) => isJsonObject(r) || typeof r === "string").map((rubric, index) => {
+        if (typeof rubric === "string") {
+          return {
+            id: `rubric-${index + 1}`,
+            description: rubric,
+            weight: 1,
+            required: true
+          };
+        }
+        return {
+          id: asString4(rubric.id) ?? `rubric-${index + 1}`,
+          description: asString4(rubric.description) ?? "",
+          weight: typeof rubric.weight === "number" ? rubric.weight : 1,
+          required: typeof rubric.required === "boolean" ? rubric.required : true
+        };
+      }).filter((r) => r.description.length > 0);
+      if (rubricItems.length > 0) {
+        const rubricEvaluator = {
+          name: "rubric",
+          type: "llm_judge",
+          rubrics: rubricItems
+        };
+        evaluators = evaluators ? [rubricEvaluator, ...evaluators] : [rubricEvaluator];
+      }
+    }
+    const userFilePaths = [];
+    for (const segment of inputSegments) {
+      if (segment.type === "file" && typeof segment.resolvedPath === "string") {
+        userFilePaths.push(segment.resolvedPath);
+      }
+    }
+    const allFilePaths = [
+      ...guidelinePaths.map((guidelinePath) => import_node_path5.default.resolve(guidelinePath)),
+      ...userFilePaths
+    ];
+    const testCase = {
+      id,
+      dataset: datasetName,
+      conversation_id: conversationId,
+      question,
+      input_messages: inputMessages,
+      input_segments: inputSegments,
+      expected_messages: outputSegments,
+      reference_answer: referenceAnswer,
+      guideline_paths: guidelinePaths.map((guidelinePath) => import_node_path5.default.resolve(guidelinePath)),
+      guideline_patterns: guidelinePatterns,
+      file_paths: allFilePaths,
+      expected_outcome: outcome,
+      evaluator: evalCaseEvaluatorKind,
+      evaluators
+    };
+    if (verbose) {
+      console.log(`
+[Eval Case: ${id}]`);
+      if (testCase.guideline_paths.length > 0) {
+        console.log(`  Guidelines used: ${testCase.guideline_paths.length}`);
+        for (const guidelinePath of testCase.guideline_paths) {
+          console.log(`    - ${guidelinePath}`);
+        }
+      } else {
+        console.log("  No guidelines found");
+      }
+    }
+    results.push(testCase);
+  }
+  return results;
+}
+function asString4(value) {
+  return typeof value === "string" ? value : void 0;
+}
+function logWarning4(message, details) {
+  if (details && details.length > 0) {
+    const detailBlock = details.join("\n");
+    console.warn(`${ANSI_YELLOW5}Warning: ${message}
+${detailBlock}${ANSI_RESET5}`);
+  } else {
+    console.warn(`${ANSI_YELLOW5}Warning: ${message}${ANSI_RESET5}`);
+  }
+}
+function logError(message, details) {
+  if (details && details.length > 0) {
+    const detailBlock = details.join("\n");
+    console.error(`${ANSI_RED}Error: ${message}
+${detailBlock}${ANSI_RESET5}`);
+  } else {
+    console.error(`${ANSI_RED}Error: ${message}${ANSI_RESET5}`);
+  }
+}
+
+// src/evaluation/formatting/prompt-builder.ts
+var import_promises6 = require("fs/promises");
+var import_node_path6 = __toESM(require("path"), 1);
+var ANSI_YELLOW6 = "\x1B[33m";
+var ANSI_RESET6 = "\x1B[0m";
 async function buildPromptInputs(testCase, mode = "lm") {
   const guidelineParts = [];
   for (const rawPath of testCase.guideline_paths) {
-    const absolutePath = import_node_path5.default.resolve(rawPath);
+    const absolutePath = import_node_path6.default.resolve(rawPath);
     if (!await fileExists(absolutePath)) {
-      logWarning4(`Could not read guideline file ${absolutePath}: file does not exist`);
+      logWarning5(`Could not read guideline file ${absolutePath}: file does not exist`);
       continue;
     }
     try {
-      const content = (await (0, import_promises5.readFile)(absolutePath, "utf8")).replace(/\r\n/g, "\n").trim();
+      const content = (await (0, import_promises6.readFile)(absolutePath, "utf8")).replace(/\r\n/g, "\n").trim();
       guidelineParts.push({
         content,
         isFile: true,
-        displayPath: import_node_path5.default.basename(absolutePath)
+        displayPath: import_node_path6.default.basename(absolutePath)
       });
     } catch (error) {
-      logWarning4(`Could not read guideline file ${absolutePath}: ${error.message}`);
+      logWarning5(`Could not read guideline file ${absolutePath}: ${error.message}`);
     }
   }
   const guidelines = formatFileContents(guidelineParts);
@@ -1311,9 +1560,9 @@ async function buildPromptInputs(testCase, mode = "lm") {
             messageSegments.push({ type: "text", value: segment });
           }
         } else if (isJsonObject(segment)) {
-          const type = asString4(segment.type);
+          const type = asString5(segment.type);
           if (type === "file") {
-            const value = asString4(segment.value);
+            const value = asString5(segment.value);
             if (!value) continue;
             if (testCase.guideline_patterns && isGuidelineFile(value, testCase.guideline_patterns)) {
               messageSegments.push({ type: "guideline_ref", path: value });
@@ -1324,7 +1573,7 @@ async function buildPromptInputs(testCase, mode = "lm") {
               messageSegments.push({ type: "file", text: fileText, path: value });
             }
           } else if (type === "text") {
-            const textValue = asString4(segment.value);
+            const textValue = asString5(segment.value);
             if (textValue && textValue.trim().length > 0) {
               messageSegments.push({ type: "text", value: textValue });
             }
@@ -1478,22 +1727,22 @@ ${guidelineContent.trim()}`);
   }
   return chatPrompt.length > 0 ? chatPrompt : void 0;
 }
-function asString4(value) {
+function asString5(value) {
   return typeof value === "string" ? value : void 0;
 }
-function logWarning4(message) {
-  console.warn(`${ANSI_YELLOW5}Warning: ${message}${ANSI_RESET5}`);
+function logWarning5(message) {
+  console.warn(`${ANSI_YELLOW6}Warning: ${message}${ANSI_RESET6}`);
 }
 
 // src/evaluation/yaml-parser.ts
-var ANSI_YELLOW6 = "\x1B[33m";
-var ANSI_RED = "\x1B[31m";
-var ANSI_RESET6 = "\x1B[0m";
+var ANSI_YELLOW7 = "\x1B[33m";
+var ANSI_RED2 = "\x1B[31m";
+var ANSI_RESET7 = "\x1B[0m";
 async function readTestSuiteMetadata(testFilePath) {
   try {
-    const absolutePath = import_node_path6.default.resolve(testFilePath);
-    const content = await (0, import_promises6.readFile)(absolutePath, "utf8");
-    const parsed = (0, import_yaml2.parse)(content);
+    const absolutePath = import_node_path7.default.resolve(testFilePath);
+    const content = await (0, import_promises7.readFile)(absolutePath, "utf8");
+    const parsed = (0, import_yaml3.parse)(content);
     if (!isJsonObject(parsed)) {
       return {};
     }
@@ -1503,21 +1752,25 @@ async function readTestSuiteMetadata(testFilePath) {
   }
 }
 async function loadEvalCases(evalFilePath, repoRoot, options) {
+  const format = detectFormat(evalFilePath);
+  if (format === "jsonl") {
+    return loadEvalCasesFromJsonl(evalFilePath, repoRoot, options);
+  }
   const verbose = options?.verbose ?? false;
   const evalIdFilter = options?.evalId;
-  const absoluteTestPath = import_node_path6.default.resolve(evalFilePath);
+  const absoluteTestPath = import_node_path7.default.resolve(evalFilePath);
   const repoRootPath = resolveToAbsolutePath(repoRoot);
   const searchRoots = buildSearchRoots(absoluteTestPath, repoRootPath);
   const config = await loadConfig(absoluteTestPath, repoRootPath);
   const guidelinePatterns = config?.guideline_patterns;
-  const rawFile = await (0, import_promises6.readFile)(absoluteTestPath, "utf8");
-  const parsed = (0, import_yaml2.parse)(rawFile);
+  const rawFile = await (0, import_promises7.readFile)(absoluteTestPath, "utf8");
+  const parsed = (0, import_yaml3.parse)(rawFile);
   if (!isJsonObject(parsed)) {
     throw new Error(`Invalid test file format: ${evalFilePath}`);
   }
   const suite = parsed;
-  const datasetNameFromSuite = asString5(suite.dataset)?.trim();
-  const fallbackDataset = import_node_path6.default.basename(absoluteTestPath).replace(/\.ya?ml$/i, "") || "eval";
+  const datasetNameFromSuite = asString6(suite.dataset)?.trim();
+  const fallbackDataset = import_node_path7.default.basename(absoluteTestPath).replace(/\.ya?ml$/i, "") || "eval";
   const datasetName = datasetNameFromSuite && datasetNameFromSuite.length > 0 ? datasetNameFromSuite : fallbackDataset;
   const rawTestcases = suite.evalcases;
   if (!Array.isArray(rawTestcases)) {
@@ -1525,24 +1778,24 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
   }
   const globalEvaluator = coerceEvaluator(suite.evaluator, "global") ?? "llm_judge";
   const globalExecution = isJsonObject(suite.execution) ? suite.execution : void 0;
-  const _globalTarget = asString5(globalExecution?.target) ?? asString5(suite.target);
+  const _globalTarget = asString6(globalExecution?.target) ?? asString6(suite.target);
   const results = [];
   for (const rawEvalcase of rawTestcases) {
     if (!isJsonObject(rawEvalcase)) {
-      logWarning5("Skipping invalid eval case entry (expected object)");
+      logWarning6("Skipping invalid eval case entry (expected object)");
       continue;
     }
     const evalcase = rawEvalcase;
-    const id = asString5(evalcase.id);
+    const id = asString6(evalcase.id);
     if (evalIdFilter && id !== evalIdFilter) {
       continue;
     }
-    const conversationId = asString5(evalcase.conversation_id);
-    const outcome = asString5(evalcase.expected_outcome) ?? asString5(evalcase.outcome);
+    const conversationId = asString6(evalcase.conversation_id);
+    const outcome = asString6(evalcase.expected_outcome) ?? asString6(evalcase.outcome);
     const inputMessagesValue = evalcase.input_messages;
     const expectedMessagesValue = evalcase.expected_messages;
     if (!id || !outcome || !Array.isArray(inputMessagesValue)) {
-      logError(
+      logError2(
         `Skipping incomplete eval case: ${id ?? "unknown"}. Missing required fields: id, outcome, and/or input_messages`
       );
       continue;
@@ -1553,7 +1806,7 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
     );
     const expectedMessages = hasExpectedMessages ? expectedMessagesValue.filter((msg) => isTestMessage(msg)) : [];
     if (hasExpectedMessages && expectedMessages.length === 0) {
-      logError(`No valid expected message found for eval case: ${id}`);
+      logError2(`No valid expected message found for eval case: ${id}`);
       continue;
     }
     const guidelinePaths = [];
@@ -1594,7 +1847,7 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
       evaluators = await parseEvaluators(evalcase, globalExecution, searchRoots, id ?? "unknown");
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
-      logError(`Skipping eval case '${id}': ${message}`);
+      logError2(`Skipping eval case '${id}': ${message}`);
       continue;
     }
     const inlineRubrics = evalcase.rubrics;
@@ -1609,8 +1862,8 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
           };
         }
         return {
-          id: asString5(rubric.id) ?? `rubric-${index + 1}`,
-          description: asString5(rubric.description) ?? "",
+          id: asString6(rubric.id) ?? `rubric-${index + 1}`,
+          description: asString6(rubric.description) ?? "",
           weight: typeof rubric.weight === "number" ? rubric.weight : 1,
           required: typeof rubric.required === "boolean" ? rubric.required : true
         };
@@ -1631,7 +1884,7 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
       }
     }
     const allFilePaths = [
-      ...guidelinePaths.map((guidelinePath) => import_node_path6.default.resolve(guidelinePath)),
+      ...guidelinePaths.map((guidelinePath) => import_node_path7.default.resolve(guidelinePath)),
       ...userFilePaths
     ];
     const testCase = {
@@ -1643,7 +1896,7 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
       input_segments: inputSegments,
       expected_messages: outputSegments,
       reference_answer: referenceAnswer,
-      guideline_paths: guidelinePaths.map((guidelinePath) => import_node_path6.default.resolve(guidelinePath)),
+      guideline_paths: guidelinePaths.map((guidelinePath) => import_node_path7.default.resolve(guidelinePath)),
       guideline_patterns: guidelinePatterns,
       file_paths: allFilePaths,
       expected_outcome: outcome,
@@ -1666,35 +1919,35 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
   }
   return results;
 }
-function asString5(value) {
+function asString6(value) {
   return typeof value === "string" ? value : void 0;
 }
-function logWarning5(message, details) {
+function logWarning6(message, details) {
   if (details && details.length > 0) {
     const detailBlock = details.join("\n");
-    console.warn(`${ANSI_YELLOW6}Warning: ${message}
-${detailBlock}${ANSI_RESET6}`);
+    console.warn(`${ANSI_YELLOW7}Warning: ${message}
+${detailBlock}${ANSI_RESET7}`);
   } else {
-    console.warn(`${ANSI_YELLOW6}Warning: ${message}${ANSI_RESET6}`);
+    console.warn(`${ANSI_YELLOW7}Warning: ${message}${ANSI_RESET7}`);
   }
 }
-function logError(message, details) {
+function logError2(message, details) {
   if (details && details.length > 0) {
     const detailBlock = details.join("\n");
-    console.error(`${ANSI_RED}Error: ${message}
-${detailBlock}${ANSI_RESET6}`);
+    console.error(`${ANSI_RED2}Error: ${message}
+${detailBlock}${ANSI_RESET7}`);
   } else {
-    console.error(`${ANSI_RED}Error: ${message}${ANSI_RESET6}`);
+    console.error(`${ANSI_RED2}Error: ${message}${ANSI_RESET7}`);
   }
 }
 
 // src/evaluation/file-utils.ts
 var import_node_fs2 = require("fs");
-var import_promises7 = require("fs/promises");
-var import_node_path7 = __toESM(require("path"), 1);
+var import_promises8 = require("fs/promises");
+var import_node_path8 = __toESM(require("path"), 1);
 async function fileExists2(filePath) {
   try {
-    await (0, import_promises7.access)(filePath, import_node_fs2.constants.F_OK);
+    await (0, import_promises8.access)(filePath, import_node_fs2.constants.F_OK);
     return true;
   } catch {
     return false;
@@ -1704,22 +1957,22 @@ function normalizeLineEndings(content) {
   return content.replace(/\r\n/g, "\n");
 }
 async function readTextFile(filePath) {
-  const content = await (0, import_promises7.readFile)(filePath, "utf8");
+  const content = await (0, import_promises8.readFile)(filePath, "utf8");
   return normalizeLineEndings(content);
 }
 async function readJsonFile(filePath) {
-  const content = await (0, import_promises7.readFile)(filePath, "utf8");
+  const content = await (0, import_promises8.readFile)(filePath, "utf8");
   return JSON.parse(content);
 }
 async function findGitRoot(startPath) {
-  let currentDir = import_node_path7.default.dirname(import_node_path7.default.resolve(startPath));
-  const root = import_node_path7.default.parse(currentDir).root;
+  let currentDir = import_node_path8.default.dirname(import_node_path8.default.resolve(startPath));
+  const root = import_node_path8.default.parse(currentDir).root;
   while (currentDir !== root) {
-    const gitPath = import_node_path7.default.join(currentDir, ".git");
+    const gitPath = import_node_path8.default.join(currentDir, ".git");
     if (await fileExists2(gitPath)) {
       return currentDir;
     }
-    const parentDir = import_node_path7.default.dirname(currentDir);
+    const parentDir = import_node_path8.default.dirname(currentDir);
     if (parentDir === currentDir) {
       break;
     }
@@ -1730,8 +1983,8 @@ async function findGitRoot(startPath) {
 function buildDirectoryChain2(filePath, repoRoot) {
   const directories = [];
   const seen = /* @__PURE__ */ new Set();
-  const boundary = import_node_path7.default.resolve(repoRoot);
-  let current = import_node_path7.default.resolve(import_node_path7.default.dirname(filePath));
+  const boundary = import_node_path8.default.resolve(repoRoot);
+  let current = import_node_path8.default.resolve(import_node_path8.default.dirname(filePath));
   while (current !== void 0) {
     if (!seen.has(current)) {
       directories.push(current);
@@ -1740,7 +1993,7 @@ function buildDirectoryChain2(filePath, repoRoot) {
     if (current === boundary) {
       break;
     }
-    const parent = import_node_path7.default.dirname(current);
+    const parent = import_node_path8.default.dirname(current);
     if (parent === current) {
       break;
     }
@@ -1754,16 +2007,16 @@ function buildDirectoryChain2(filePath, repoRoot) {
 function buildSearchRoots2(evalPath, repoRoot) {
   const uniqueRoots = [];
   const addRoot = (root) => {
-    const normalized = import_node_path7.default.resolve(root);
+    const normalized = import_node_path8.default.resolve(root);
     if (!uniqueRoots.includes(normalized)) {
       uniqueRoots.push(normalized);
     }
   };
-  let currentDir = import_node_path7.default.dirname(evalPath);
+  let currentDir = import_node_path8.default.dirname(evalPath);
   let reachedBoundary = false;
   while (!reachedBoundary) {
     addRoot(currentDir);
-    const parentDir = import_node_path7.default.dirname(currentDir);
+    const parentDir = import_node_path8.default.dirname(currentDir);
     if (currentDir === repoRoot || parentDir === currentDir) {
       reachedBoundary = true;
     } else {
@@ -1781,16 +2034,16 @@ function trimLeadingSeparators2(value) {
 async function resolveFileReference2(rawValue, searchRoots) {
   const displayPath = trimLeadingSeparators2(rawValue);
   const potentialPaths = [];
-  if (import_node_path7.default.isAbsolute(rawValue)) {
-    potentialPaths.push(import_node_path7.default.normalize(rawValue));
+  if (import_node_path8.default.isAbsolute(rawValue)) {
+    potentialPaths.push(import_node_path8.default.normalize(rawValue));
   }
   for (const base of searchRoots) {
-    potentialPaths.push(import_node_path7.default.resolve(base, displayPath));
+    potentialPaths.push(import_node_path8.default.resolve(base, displayPath));
   }
   const attempted = [];
   const seen = /* @__PURE__ */ new Set();
   for (const candidate of potentialPaths) {
-    const absoluteCandidate = import_node_path7.default.resolve(candidate);
+    const absoluteCandidate = import_node_path8.default.resolve(candidate);
     if (seen.has(absoluteCandidate)) {
       continue;
     }
@@ -2140,9 +2393,9 @@ async function withRetry(fn, retryConfig, signal) {
 var import_node_child_process = require("child_process");
 var import_node_crypto = require("crypto");
 var import_node_fs3 = require("fs");
-var import_promises8 = require("fs/promises");
+var import_promises9 = require("fs/promises");
 var import_node_os = require("os");
-var import_node_path9 = __toESM(require("path"), 1);
+var import_node_path10 = __toESM(require("path"), 1);
 
 // src/evaluation/providers/claude-code-log-tracker.ts
 var GLOBAL_LOGS_KEY = Symbol.for("agentv.claudeCodeLogs");
@@ -2198,7 +2451,7 @@ function subscribeToClaudeCodeLogEntries(listener) {
 }
 
 // src/evaluation/providers/preread.ts
-var import_node_path8 = __toESM(require("path"), 1);
+var import_node_path9 = __toESM(require("path"), 1);
 function buildPromptDocument(request, inputFiles, options) {
   const parts = [];
   const guidelineFiles = collectGuidelineFiles(
@@ -2221,7 +2474,7 @@ function normalizeInputFiles(inputFiles) {
   }
   const deduped = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = import_node_path8.default.resolve(inputFile);
+    const absolutePath = import_node_path9.default.resolve(inputFile);
     if (!deduped.has(absolutePath)) {
       deduped.set(absolutePath, absolutePath);
     }
@@ -2234,14 +2487,14 @@ function collectGuidelineFiles(inputFiles, guidelinePatterns, overrides) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = import_node_path8.default.resolve(inputFile);
+    const absolutePath = import_node_path9.default.resolve(inputFile);
     if (overrides?.has(absolutePath)) {
       if (!unique.has(absolutePath)) {
         unique.set(absolutePath, absolutePath);
       }
       continue;
     }
-    const normalized = absolutePath.split(import_node_path8.default.sep).join("/");
+    const normalized = absolutePath.split(import_node_path9.default.sep).join("/");
     if (isGuidelineFile(normalized, guidelinePatterns)) {
       if (!unique.has(absolutePath)) {
         unique.set(absolutePath, absolutePath);
@@ -2256,7 +2509,7 @@ function collectInputFiles(inputFiles) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = import_node_path8.default.resolve(inputFile);
+    const absolutePath = import_node_path9.default.resolve(inputFile);
     if (!unique.has(absolutePath)) {
       unique.set(absolutePath, absolutePath);
     }
@@ -2268,7 +2521,7 @@ function buildMandatoryPrereadBlock(guidelineFiles, inputFiles) {
     return "";
   }
   const buildList = (files) => files.map((absolutePath) => {
-    const fileName = import_node_path8.default.basename(absolutePath);
+    const fileName = import_node_path9.default.basename(absolutePath);
     const fileUri = pathToFileUri(absolutePath);
     return `* [${fileName}](${fileUri})`;
   });
@@ -2288,7 +2541,7 @@ ${buildList(inputFiles).join("\n")}.`);
   return sections.join("\n");
 }
 function pathToFileUri(filePath) {
-  const absolutePath = import_node_path8.default.isAbsolute(filePath) ? filePath : import_node_path8.default.resolve(filePath);
+  const absolutePath = import_node_path9.default.isAbsolute(filePath) ? filePath : import_node_path9.default.resolve(filePath);
   const normalizedPath = absolutePath.replace(/\\/g, "/");
   if (/^[a-zA-Z]:\//.test(normalizedPath)) {
     return `file:///${normalizedPath}`;
@@ -2325,8 +2578,8 @@ var ClaudeCodeProvider = class {
     const workspaceRoot = await this.createWorkspace();
     const logger = await this.createStreamLogger(request).catch(() => void 0);
     try {
-      const promptFile = import_node_path9.default.join(workspaceRoot, PROMPT_FILENAME);
-      await (0, import_promises8.writeFile)(promptFile, request.question, "utf8");
+      const promptFile = import_node_path10.default.join(workspaceRoot, PROMPT_FILENAME);
+      await (0, import_promises9.writeFile)(promptFile, request.question, "utf8");
       const args = this.buildClaudeCodeArgs(request.question, inputFiles);
       const cwd = this.resolveCwd();
       const result = await this.executeClaudeCode(args, cwd, request.signal, logger);
@@ -2373,7 +2626,7 @@ var ClaudeCodeProvider = class {
     if (!this.config.cwd) {
       return process.cwd();
     }
-    return import_node_path9.default.resolve(this.config.cwd);
+    return import_node_path10.default.resolve(this.config.cwd);
   }
   buildClaudeCodeArgs(prompt, inputFiles) {
     const args = [];
@@ -2430,11 +2683,11 @@ ${filesContext}`;
     }
   }
   async createWorkspace() {
-    return await (0, import_promises8.mkdtemp)(import_node_path9.default.join((0, import_node_os.tmpdir)(), WORKSPACE_PREFIX));
+    return await (0, import_promises9.mkdtemp)(import_node_path10.default.join((0, import_node_os.tmpdir)(), WORKSPACE_PREFIX));
   }
   async cleanupWorkspace(workspaceRoot) {
     try {
-      await (0, import_promises8.rm)(workspaceRoot, { recursive: true, force: true });
+      await (0, import_promises9.rm)(workspaceRoot, { recursive: true, force: true });
     } catch {
     }
   }
@@ -2444,9 +2697,9 @@ ${filesContext}`;
       return void 0;
     }
     if (this.config.logDir) {
-      return import_node_path9.default.resolve(this.config.logDir);
+      return import_node_path10.default.resolve(this.config.logDir);
     }
-    return import_node_path9.default.join(process.cwd(), ".agentv", "logs", "claude-code");
+    return import_node_path10.default.join(process.cwd(), ".agentv", "logs", "claude-code");
   }
   async createStreamLogger(request) {
     const logDir = this.resolveLogDirectory();
@@ -2454,13 +2707,13 @@ ${filesContext}`;
       return void 0;
     }
     try {
-      await (0, import_promises8.mkdir)(logDir, { recursive: true });
+      await (0, import_promises9.mkdir)(logDir, { recursive: true });
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
       console.warn(`Skipping Claude Code stream logging (could not create ${logDir}): ${message}`);
       return void 0;
     }
-    const filePath = import_node_path9.default.join(logDir, buildLogFilename(request, this.targetName));
+    const filePath = import_node_path10.default.join(logDir, buildLogFilename(request, this.targetName));
     try {
       const logger = await ClaudeCodeStreamLogger.create({
         filePath,
@@ -2865,16 +3118,16 @@ function escapeShellArg(arg) {
 }
 async function defaultClaudeCodeRunner(options) {
   const tempId = (0, import_node_crypto.randomUUID)();
-  const stdoutFile = import_node_path9.default.join((0, import_node_os.tmpdir)(), `agentv-cc-${tempId}-stdout`);
-  const stderrFile = import_node_path9.default.join((0, import_node_os.tmpdir)(), `agentv-cc-${tempId}-stderr`);
-  const exitFile = import_node_path9.default.join((0, import_node_os.tmpdir)(), `agentv-cc-${tempId}-exit`);
-  const pidFile = import_node_path9.default.join((0, import_node_os.tmpdir)(), `agentv-cc-${tempId}-pid`);
+  const stdoutFile = import_node_path10.default.join((0, import_node_os.tmpdir)(), `agentv-cc-${tempId}-stdout`);
+  const stderrFile = import_node_path10.default.join((0, import_node_os.tmpdir)(), `agentv-cc-${tempId}-stderr`);
+  const exitFile = import_node_path10.default.join((0, import_node_os.tmpdir)(), `agentv-cc-${tempId}-exit`);
+  const pidFile = import_node_path10.default.join((0, import_node_os.tmpdir)(), `agentv-cc-${tempId}-pid`);
   try {
     return await runClaudeCodeWithTempFiles(options, stdoutFile, stderrFile, exitFile, pidFile);
   } finally {
     for (const file of [stdoutFile, stderrFile, exitFile, pidFile]) {
       try {
-        await (0, import_promises8.rm)(file, { force: true });
+        await (0, import_promises9.rm)(file, { force: true });
       } catch {
       }
     }
@@ -2908,8 +3161,8 @@ async function runClaudeCodeWithTempFiles(options, stdoutFile, stderrFile, exitF
   let lastStdoutSize = 0;
   const readFileIfExists = async (filePath) => {
     try {
-      const { readFile: readFile8 } = await import("fs/promises");
-      return await readFile8(filePath, "utf8");
+      const { readFile: readFile9 } = await import("fs/promises");
+      return await readFile9(filePath, "utf8");
     } catch {
       return "";
     }
@@ -2982,9 +3235,9 @@ async function runClaudeCodeWithTempFiles(options, stdoutFile, stderrFile, exitF
 
 // src/evaluation/providers/cli.ts
 var import_node_child_process2 = require("child_process");
-var import_promises9 = __toESM(require("fs/promises"), 1);
+var import_promises10 = __toESM(require("fs/promises"), 1);
 var import_node_os2 = __toESM(require("os"), 1);
-var import_node_path10 = __toESM(require("path"), 1);
+var import_node_path11 = __toESM(require("path"), 1);
 var import_node_util = require("util");
 var import_zod = require("zod");
 var ToolCallSchema = import_zod.z.object({
@@ -3353,7 +3606,7 @@ var CliProvider = class {
       throw new Error(`Failed to read output file '${filePath}': ${errorMsg}`);
     } finally {
       if (!this.keepTempFiles) {
-        await import_promises9.default.unlink(filePath).catch(() => {
+        await import_promises10.default.unlink(filePath).catch(() => {
         });
       }
     }
@@ -3441,7 +3694,7 @@ function normalizeInputFiles2(inputFiles) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = import_node_path10.default.resolve(inputFile);
+    const absolutePath = import_node_path11.default.resolve(inputFile);
     if (!unique.has(absolutePath)) {
       unique.set(absolutePath, absolutePath);
     }
@@ -3455,7 +3708,7 @@ function formatFileList(files, template) {
   const formatter = template ?? "{path}";
   return files.map((filePath) => {
     const escapedPath = shellEscape(filePath);
-    const escapedName = shellEscape(import_node_path10.default.basename(filePath));
+    const escapedName = shellEscape(import_node_path11.default.basename(filePath));
     return formatter.replaceAll("{path}", escapedPath).replaceAll("{basename}", escapedName);
   }).join(" ");
 }
@@ -3479,7 +3732,7 @@ function generateOutputFilePath(evalCaseId, extension = ".json") {
   const safeEvalId = evalCaseId || "unknown";
   const timestamp = Date.now();
   const random = Math.random().toString(36).substring(2, 9);
-  return import_node_path10.default.join(import_node_os2.default.tmpdir(), `agentv-${safeEvalId}-${timestamp}-${random}${extension}`);
+  return import_node_path11.default.join(import_node_os2.default.tmpdir(), `agentv-${safeEvalId}-${timestamp}-${random}${extension}`);
 }
 function formatTimeoutSuffix2(timeoutMs) {
   if (!timeoutMs || timeoutMs <= 0) {
@@ -3493,9 +3746,9 @@ function formatTimeoutSuffix2(timeoutMs) {
 var import_node_child_process3 = require("child_process");
 var import_node_crypto2 = require("crypto");
 var import_node_fs4 = require("fs");
-var import_promises10 = require("fs/promises");
+var import_promises11 = require("fs/promises");
 var import_node_os3 = require("os");
-var import_node_path11 = __toESM(require("path"), 1);
+var import_node_path12 = __toESM(require("path"), 1);
 var import_node_util2 = require("util");
 
 // src/evaluation/providers/codex-log-tracker.ts
@@ -3590,8 +3843,8 @@ var CodexProvider = class {
       const promptContent = `${systemPrompt}
 
 ${basePrompt}`;
-      const promptFile = import_node_path11.default.join(workspaceRoot, PROMPT_FILENAME2);
-      await (0, import_promises10.writeFile)(promptFile, promptContent, "utf8");
+      const promptFile = import_node_path12.default.join(workspaceRoot, PROMPT_FILENAME2);
+      await (0, import_promises11.writeFile)(promptFile, promptContent, "utf8");
       const args = this.buildCodexArgs();
       const cwd = this.resolveCwd(workspaceRoot);
       const result = await this.executeCodex(args, cwd, promptContent, request.signal, logger);
@@ -3640,7 +3893,7 @@ ${basePrompt}`;
     if (!this.config.cwd) {
       return workspaceRoot;
     }
-    return import_node_path11.default.resolve(this.config.cwd);
+    return import_node_path12.default.resolve(this.config.cwd);
   }
   buildCodexArgs() {
     const args = [
@@ -3682,11 +3935,11 @@ ${basePrompt}`;
     }
   }
   async createWorkspace() {
-    return await (0, import_promises10.mkdtemp)(import_node_path11.default.join((0, import_node_os3.tmpdir)(), WORKSPACE_PREFIX2));
+    return await (0, import_promises11.mkdtemp)(import_node_path12.default.join((0, import_node_os3.tmpdir)(), WORKSPACE_PREFIX2));
   }
   async cleanupWorkspace(workspaceRoot) {
     try {
-      await (0, import_promises10.rm)(workspaceRoot, { recursive: true, force: true });
+      await (0, import_promises11.rm)(workspaceRoot, { recursive: true, force: true });
     } catch {
     }
   }
@@ -3696,9 +3949,9 @@ ${basePrompt}`;
       return void 0;
     }
     if (this.config.logDir) {
-      return import_node_path11.default.resolve(this.config.logDir);
+      return import_node_path12.default.resolve(this.config.logDir);
     }
-    return import_node_path11.default.join(process.cwd(), ".agentv", "logs", "codex");
+    return import_node_path12.default.join(process.cwd(), ".agentv", "logs", "codex");
   }
   async createStreamLogger(request) {
     const logDir = this.resolveLogDirectory();
@@ -3706,13 +3959,13 @@ ${basePrompt}`;
       return void 0;
     }
     try {
-      await (0, import_promises10.mkdir)(logDir, { recursive: true });
+      await (0, import_promises11.mkdir)(logDir, { recursive: true });
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
       console.warn(`Skipping Codex stream logging (could not create ${logDir}): ${message}`);
       return void 0;
     }
-    const filePath = import_node_path11.default.join(logDir, buildLogFilename2(request, this.targetName));
+    const filePath = import_node_path12.default.join(logDir, buildLogFilename2(request, this.targetName));
     try {
       const logger = await CodexStreamLogger.create({
         filePath,
@@ -3927,9 +4180,9 @@ function tryParseJsonValue2(rawLine) {
 async function locateExecutable(candidate) {
   const includesPathSeparator = candidate.includes("/") || candidate.includes("\\");
   if (includesPathSeparator) {
-    const resolved = import_node_path11.default.isAbsolute(candidate) ? candidate : import_node_path11.default.resolve(candidate);
+    const resolved = import_node_path12.default.isAbsolute(candidate) ? candidate : import_node_path12.default.resolve(candidate);
     const executablePath = await ensureWindowsExecutableVariant(resolved);
-    await (0, import_promises10.access)(executablePath, import_node_fs4.constants.F_OK);
+    await (0, import_promises11.access)(executablePath, import_node_fs4.constants.F_OK);
     return executablePath;
   }
   const locator = process.platform === "win32" ? "where" : "which";
@@ -3939,7 +4192,7 @@ async function locateExecutable(candidate) {
     const preferred = selectExecutableCandidate(lines);
     if (preferred) {
       const executablePath = await ensureWindowsExecutableVariant(preferred);
-      await (0, import_promises10.access)(executablePath, import_node_fs4.constants.F_OK);
+      await (0, import_promises11.access)(executablePath, import_node_fs4.constants.F_OK);
       return executablePath;
     }
   } catch {
@@ -3973,7 +4226,7 @@ async function ensureWindowsExecutableVariant(candidate) {
   for (const ext of extensions) {
     const withExtension = `${candidate}${ext}`;
     try {
-      await (0, import_promises10.access)(withExtension, import_node_fs4.constants.F_OK);
+      await (0, import_promises11.access)(withExtension, import_node_fs4.constants.F_OK);
       return withExtension;
     } catch {
     }
@@ -4438,9 +4691,9 @@ function extractToolCalls2(content) {
 var import_node_child_process4 = require("child_process");
 var import_node_crypto3 = require("crypto");
 var import_node_fs5 = require("fs");
-var import_promises11 = require("fs/promises");
+var import_promises12 = require("fs/promises");
 var import_node_os4 = require("os");
-var import_node_path12 = __toESM(require("path"), 1);
+var import_node_path13 = __toESM(require("path"), 1);
 
 // src/evaluation/providers/pi-log-tracker.ts
 var GLOBAL_LOGS_KEY3 = Symbol.for("agentv.piLogs");
@@ -4524,8 +4777,8 @@ var PiCodingAgentProvider = class {
     const workspaceRoot = await this.createWorkspace();
     const logger = await this.createStreamLogger(request).catch(() => void 0);
     try {
-      const promptFile = import_node_path12.default.join(workspaceRoot, PROMPT_FILENAME3);
-      await (0, import_promises11.writeFile)(promptFile, request.question, "utf8");
+      const promptFile = import_node_path13.default.join(workspaceRoot, PROMPT_FILENAME3);
+      await (0, import_promises12.writeFile)(promptFile, request.question, "utf8");
       const args = this.buildPiArgs(request.question, inputFiles);
       const cwd = this.resolveCwd(workspaceRoot);
       const result = await this.executePi(args, cwd, request.signal, logger);
@@ -4566,7 +4819,7 @@ var PiCodingAgentProvider = class {
     if (!this.config.cwd) {
       return workspaceRoot;
     }
-    return import_node_path12.default.resolve(this.config.cwd);
+    return import_node_path13.default.resolve(this.config.cwd);
   }
   buildPiArgs(prompt, inputFiles) {
     const args = [];
@@ -4655,19 +4908,19 @@ ${prompt}`;
     return env;
   }
   async createWorkspace() {
-    return await (0, import_promises11.mkdtemp)(import_node_path12.default.join((0, import_node_os4.tmpdir)(), WORKSPACE_PREFIX3));
+    return await (0, import_promises12.mkdtemp)(import_node_path13.default.join((0, import_node_os4.tmpdir)(), WORKSPACE_PREFIX3));
   }
   async cleanupWorkspace(workspaceRoot) {
     try {
-      await (0, import_promises11.rm)(workspaceRoot, { recursive: true, force: true });
+      await (0, import_promises12.rm)(workspaceRoot, { recursive: true, force: true });
     } catch {
     }
   }
   resolveLogDirectory() {
     if (this.config.logDir) {
-      return import_node_path12.default.resolve(this.config.logDir);
+      return import_node_path13.default.resolve(this.config.logDir);
     }
-    return import_node_path12.default.join(process.cwd(), ".agentv", "logs", "pi-coding-agent");
+    return import_node_path13.default.join(process.cwd(), ".agentv", "logs", "pi-coding-agent");
   }
   async createStreamLogger(request) {
     const logDir = this.resolveLogDirectory();
@@ -4675,13 +4928,13 @@ ${prompt}`;
       return void 0;
     }
     try {
-      await (0, import_promises11.mkdir)(logDir, { recursive: true });
+      await (0, import_promises12.mkdir)(logDir, { recursive: true });
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
       console.warn(`Skipping Pi stream logging (could not create ${logDir}): ${message}`);
       return void 0;
     }
-    const filePath = import_node_path12.default.join(logDir, buildLogFilename3(request, this.targetName));
+    const filePath = import_node_path13.default.join(logDir, buildLogFilename3(request, this.targetName));
     try {
       const logger = await PiStreamLogger.create({
         filePath,
@@ -5114,7 +5367,7 @@ async function defaultPiRunner(options) {
 }
 
 // src/evaluation/providers/targets.ts
-var import_node_path13 = __toESM(require("path"), 1);
+var import_node_path14 = __toESM(require("path"), 1);
 var import_zod2 = require("zod");
 var CliHealthcheckHttpInputSchema = import_zod2.z.object({
   type: import_zod2.z.literal("http"),
@@ -5220,11 +5473,11 @@ function normalizeCliHealthcheck(input, env, targetName, evalFilePath) {
     allowLiteral: true,
     optionalEnv: true
   });
-  if (cwd && evalFilePath && !import_node_path13.default.isAbsolute(cwd)) {
-    cwd = import_node_path13.default.resolve(import_node_path13.default.dirname(import_node_path13.default.resolve(evalFilePath)), cwd);
+  if (cwd && evalFilePath && !import_node_path14.default.isAbsolute(cwd)) {
+    cwd = import_node_path14.default.resolve(import_node_path14.default.dirname(import_node_path14.default.resolve(evalFilePath)), cwd);
   }
   if (!cwd && evalFilePath) {
-    cwd = import_node_path13.default.dirname(import_node_path13.default.resolve(evalFilePath));
+    cwd = import_node_path14.default.dirname(import_node_path14.default.resolve(evalFilePath));
   }
   return {
     type: "command",
@@ -5251,11 +5504,11 @@ function normalizeCliTargetInput(input, env, evalFilePath) {
     allowLiteral: true,
     optionalEnv: true
   });
-  if (cwd && evalFilePath && !import_node_path13.default.isAbsolute(cwd)) {
-    cwd = import_node_path13.default.resolve(import_node_path13.default.dirname(import_node_path13.default.resolve(evalFilePath)), cwd);
+  if (cwd && evalFilePath && !import_node_path14.default.isAbsolute(cwd)) {
+    cwd = import_node_path14.default.resolve(import_node_path14.default.dirname(import_node_path14.default.resolve(evalFilePath)), cwd);
   }
   if (!cwd && evalFilePath) {
-    cwd = import_node_path13.default.dirname(import_node_path13.default.resolve(evalFilePath));
+    cwd = import_node_path14.default.dirname(import_node_path14.default.resolve(evalFilePath));
   }
   const timeoutSeconds = input.timeout_seconds ?? input.timeoutSeconds;
   const timeoutMs = timeoutSeconds !== void 0 ? Math.floor(timeoutSeconds * 1e3) : void 0;
@@ -5760,8 +6013,8 @@ function resolveCliConfig(target, env, evalFilePath) {
   const parseResult = CliTargetInputSchema.safeParse(target, { errorMap: cliErrorMap });
   if (!parseResult.success) {
     const firstError = parseResult.error.errors[0];
-    const path17 = firstError?.path.join(".") || "";
-    const prefix = path17 ? `${target.name} ${path17}: ` : `${target.name}: `;
+    const path18 = firstError?.path.join(".") || "";
+    const prefix = path18 ? `${target.name} ${path18}: ` : `${target.name}: `;
     throw new Error(`${prefix}${firstError?.message}`);
   }
   const normalized = normalizeCliTargetInput(parseResult.data, env, evalFilePath);
@@ -5949,7 +6202,7 @@ function resolveOptionalNumberArray(source, description) {
 }
 
 // src/evaluation/providers/vscode.ts
-var import_node_path14 = __toESM(require("path"), 1);
+var import_node_path15 = __toESM(require("path"), 1);
 var import_subagent = require("subagent");
 
 // src/evaluation/providers/vscode-templates.ts
@@ -6119,7 +6372,7 @@ function buildMandatoryPrereadBlock2(guidelineFiles, attachmentFiles) {
     return "";
   }
   const buildList = (files) => files.map((absolutePath) => {
-    const fileName = import_node_path14.default.basename(absolutePath);
+    const fileName = import_node_path15.default.basename(absolutePath);
     const fileUri = pathToFileUri2(absolutePath);
     return `* [${fileName}](${fileUri})`;
   });
@@ -6144,8 +6397,8 @@ function collectGuidelineFiles2(attachments, guidelinePatterns) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const attachment of attachments) {
-    const absolutePath = import_node_path14.default.resolve(attachment);
-    const normalized = absolutePath.split(import_node_path14.default.sep).join("/");
+    const absolutePath = import_node_path15.default.resolve(attachment);
+    const normalized = absolutePath.split(import_node_path15.default.sep).join("/");
     if (isGuidelineFile(normalized, guidelinePatterns)) {
       if (!unique.has(absolutePath)) {
         unique.set(absolutePath, absolutePath);
@@ -6160,7 +6413,7 @@ function collectAttachmentFiles(attachments) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const attachment of attachments) {
-    const absolutePath = import_node_path14.default.resolve(attachment);
+    const absolutePath = import_node_path15.default.resolve(attachment);
     if (!unique.has(absolutePath)) {
       unique.set(absolutePath, absolutePath);
     }
@@ -6168,7 +6421,7 @@ function collectAttachmentFiles(attachments) {
   return Array.from(unique.values());
 }
 function pathToFileUri2(filePath) {
-  const absolutePath = import_node_path14.default.isAbsolute(filePath) ? filePath : import_node_path14.default.resolve(filePath);
+  const absolutePath = import_node_path15.default.isAbsolute(filePath) ? filePath : import_node_path15.default.resolve(filePath);
   const normalizedPath = absolutePath.replace(/\\/g, "/");
   if (/^[a-zA-Z]:\//.test(normalizedPath)) {
     return `file:///${normalizedPath}`;
@@ -6181,7 +6434,7 @@ function normalizeAttachments(attachments) {
   }
   const deduped = /* @__PURE__ */ new Set();
   for (const attachment of attachments) {
-    deduped.add(import_node_path14.default.resolve(attachment));
+    deduped.add(import_node_path15.default.resolve(attachment));
   }
   return Array.from(deduped);
 }
@@ -6190,7 +6443,7 @@ function mergeAttachments(all) {
   for (const list of all) {
     if (!list) continue;
     for (const inputFile of list) {
-      deduped.add(import_node_path14.default.resolve(inputFile));
+      deduped.add(import_node_path15.default.resolve(inputFile));
     }
   }
   return deduped.size > 0 ? Array.from(deduped) : void 0;
@@ -6238,9 +6491,9 @@ total unlocked subagents available: ${result.created.length + result.skippedExis
 
 // src/evaluation/providers/targets-file.ts
 var import_node_fs6 = require("fs");
-var import_promises12 = require("fs/promises");
-var import_node_path15 = __toESM(require("path"), 1);
-var import_yaml3 = require("yaml");
+var import_promises13 = require("fs/promises");
+var import_node_path16 = __toESM(require("path"), 1);
+var import_yaml4 = require("yaml");
 function isRecord(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value);
 }
@@ -6269,19 +6522,19 @@ function assertTargetDefinition(value, index, filePath) {
 }
 async function fileExists3(filePath) {
   try {
-    await (0, import_promises12.access)(filePath, import_node_fs6.constants.F_OK);
+    await (0, import_promises13.access)(filePath, import_node_fs6.constants.F_OK);
     return true;
   } catch {
     return false;
   }
 }
 async function readTargetDefinitions(filePath) {
-  const absolutePath = import_node_path15.default.resolve(filePath);
+  const absolutePath = import_node_path16.default.resolve(filePath);
   if (!await fileExists3(absolutePath)) {
     throw new Error(`targets.yaml not found at ${absolutePath}`);
   }
-  const raw = await (0, import_promises12.readFile)(absolutePath, "utf8");
-  const parsed = (0, import_yaml3.parse)(raw);
+  const raw = await (0, import_promises13.readFile)(absolutePath, "utf8");
+  const parsed = (0, import_yaml4.parse)(raw);
   if (!isRecord(parsed)) {
     throw new Error(`targets.yaml at ${absolutePath} must be a YAML object with a 'targets' field`);
   }
@@ -6487,15 +6740,15 @@ async function execFileWithStdinNode(argv, stdinPayload, options) {
   });
 }
 async function execShellWithStdin(command, stdinPayload, options = {}) {
-  const { mkdir: mkdir4, readFile: readFile8, rm: rm4, writeFile: writeFile4 } = await import("fs/promises");
+  const { mkdir: mkdir4, readFile: readFile9, rm: rm4, writeFile: writeFile4 } = await import("fs/promises");
   const { tmpdir: tmpdir4 } = await import("os");
-  const path17 = await import("path");
+  const path18 = await import("path");
   const { randomUUID: randomUUID4 } = await import("crypto");
-  const dir = path17.join(tmpdir4(), `agentv-exec-${randomUUID4()}`);
+  const dir = path18.join(tmpdir4(), `agentv-exec-${randomUUID4()}`);
   await mkdir4(dir, { recursive: true });
-  const stdinPath = path17.join(dir, "stdin.txt");
-  const stdoutPath = path17.join(dir, "stdout.txt");
-  const stderrPath = path17.join(dir, "stderr.txt");
+  const stdinPath = path18.join(dir, "stdin.txt");
+  const stdoutPath = path18.join(dir, "stdout.txt");
+  const stderrPath = path18.join(dir, "stderr.txt");
   await writeFile4(stdinPath, stdinPayload, "utf8");
   const wrappedCommand = process.platform === "win32" ? `(${command}) < ${shellEscapePath(stdinPath)} > ${shellEscapePath(stdoutPath)} 2> ${shellEscapePath(stderrPath)}` : `(${command}) < ${shellEscapePath(stdinPath)} > ${shellEscapePath(stdoutPath)} 2> ${shellEscapePath(stderrPath)}`;
   const { spawn: spawn4 } = await import("child_process");
@@ -6525,8 +6778,8 @@ async function execShellWithStdin(command, stdinPayload, options = {}) {
         resolve(code ?? 0);
       });
     });
-    const stdout = (await readFile8(stdoutPath, "utf8")).replace(/\r\n/g, "\n");
-    const stderr = (await readFile8(stderrPath, "utf8")).replace(/\r\n/g, "\n");
+    const stdout = (await readFile9(stdoutPath, "utf8")).replace(/\r\n/g, "\n");
+    const stderr = (await readFile9(stderrPath, "utf8")).replace(/\r\n/g, "\n");
     return { stdout, stderr, exitCode };
   } finally {
     await rm4(dir, { recursive: true, force: true });
@@ -6798,7 +7051,7 @@ var CodeEvaluator = class {
       outputMessages: context.outputMessages ?? null,
       guidelineFiles: context.evalCase.guideline_paths,
       inputFiles: context.evalCase.file_paths.filter(
-        (path17) => !context.evalCase.guideline_paths.includes(path17)
+        (path18) => !context.evalCase.guideline_paths.includes(path18)
       ),
       inputMessages: context.evalCase.input_messages,
       traceSummary: context.traceSummary ?? null,
@@ -7584,115 +7837,115 @@ var FieldAccuracyEvaluator = class {
    * Evaluate a single field against the expected value.
    */
   evaluateField(fieldConfig, candidateData, expectedData) {
-    const { path: path17, match, required = true, weight = 1 } = fieldConfig;
-    const candidateValue = resolvePath(candidateData, path17);
-    const expectedValue = resolvePath(expectedData, path17);
+    const { path: path18, match, required = true, weight = 1 } = fieldConfig;
+    const candidateValue = resolvePath(candidateData, path18);
+    const expectedValue = resolvePath(expectedData, path18);
     if (expectedValue === void 0) {
       return {
-        path: path17,
+        path: path18,
         score: 1,
         // No expected value means no comparison needed
         weight,
         hit: true,
-        message: `${path17}: no expected value`
+        message: `${path18}: no expected value`
       };
     }
     if (candidateValue === void 0) {
       if (required) {
         return {
-          path: path17,
+          path: path18,
           score: 0,
           weight,
           hit: false,
-          message: `${path17} (required, missing)`
+          message: `${path18} (required, missing)`
         };
       }
       return {
-        path: path17,
+        path: path18,
         score: 1,
         // Don't penalize missing optional fields
         weight: 0,
         // Zero weight means it won't affect the score
         hit: true,
-        message: `${path17}: optional field missing`
+        message: `${path18}: optional field missing`
       };
     }
     switch (match) {
       case "exact":
-        return this.compareExact(path17, candidateValue, expectedValue, weight);
+        return this.compareExact(path18, candidateValue, expectedValue, weight);
       case "numeric_tolerance":
         return this.compareNumericTolerance(
-          path17,
+          path18,
           candidateValue,
           expectedValue,
           fieldConfig,
           weight
         );
       case "date":
-        return this.compareDate(path17, candidateValue, expectedValue, fieldConfig, weight);
+        return this.compareDate(path18, candidateValue, expectedValue, fieldConfig, weight);
       default:
         return {
-          path: path17,
+          path: path18,
           score: 0,
           weight,
           hit: false,
-          message: `${path17}: unknown match type "${match}"`
+          message: `${path18}: unknown match type "${match}"`
         };
     }
   }
   /**
    * Exact equality comparison.
    */
-  compareExact(path17, candidateValue, expectedValue, weight) {
+  compareExact(path18, candidateValue, expectedValue, weight) {
     if (deepEqual(candidateValue, expectedValue)) {
       return {
-        path: path17,
+        path: path18,
         score: 1,
         weight,
         hit: true,
-        message: path17
+        message: path18
       };
     }
     if (typeof candidateValue !== typeof expectedValue) {
       return {
-        path: path17,
+        path: path18,
         score: 0,
         weight,
         hit: false,
-        message: `${path17} (type mismatch: got ${typeof candidateValue}, expected ${typeof expectedValue})`
+        message: `${path18} (type mismatch: got ${typeof candidateValue}, expected ${typeof expectedValue})`
       };
     }
     return {
-      path: path17,
+      path: path18,
       score: 0,
       weight,
       hit: false,
-      message: `${path17} (value mismatch)`
+      message: `${path18} (value mismatch)`
     };
   }
   /**
    * Numeric comparison with absolute or relative tolerance.
    */
-  compareNumericTolerance(path17, candidateValue, expectedValue, fieldConfig, weight) {
+  compareNumericTolerance(path18, candidateValue, expectedValue, fieldConfig, weight) {
     const { tolerance = 0, relative = false } = fieldConfig;
     const candidateNum = toNumber(candidateValue);
     const expectedNum = toNumber(expectedValue);
     if (candidateNum === null || expectedNum === null) {
       return {
-        path: path17,
+        path: path18,
         score: 0,
         weight,
         hit: false,
-        message: `${path17} (non-numeric value)`
+        message: `${path18} (non-numeric value)`
       };
     }
     if (!Number.isFinite(candidateNum) || !Number.isFinite(expectedNum)) {
       return {
-        path: path17,
+        path: path18,
         score: 0,
         weight,
         hit: false,
-        message: `${path17} (invalid numeric value)`
+        message: `${path18} (invalid numeric value)`
       };
     }
     const diff = Math.abs(candidateNum - expectedNum);
@@ -7705,61 +7958,61 @@ var FieldAccuracyEvaluator = class {
     }
     if (withinTolerance) {
       return {
-        path: path17,
+        path: path18,
         score: 1,
         weight,
         hit: true,
-        message: `${path17} (within tolerance: diff=${diff.toFixed(2)})`
+        message: `${path18} (within tolerance: diff=${diff.toFixed(2)})`
       };
     }
     return {
-      path: path17,
+      path: path18,
       score: 0,
       weight,
       hit: false,
-      message: `${path17} (outside tolerance: diff=${diff.toFixed(2)}, tolerance=${tolerance})`
+      message: `${path18} (outside tolerance: diff=${diff.toFixed(2)}, tolerance=${tolerance})`
     };
   }
   /**
    * Date comparison with format normalization.
    */
-  compareDate(path17, candidateValue, expectedValue, fieldConfig, weight) {
+  compareDate(path18, candidateValue, expectedValue, fieldConfig, weight) {
     const formats = fieldConfig.formats ?? DEFAULT_DATE_FORMATS;
     const candidateDate = parseDate(String(candidateValue), formats);
     const expectedDate = parseDate(String(expectedValue), formats);
     if (candidateDate === null) {
       return {
-        path: path17,
+        path: path18,
         score: 0,
         weight,
         hit: false,
-        message: `${path17} (unparseable candidate date)`
+        message: `${path18} (unparseable candidate date)`
       };
     }
     if (expectedDate === null) {
       return {
-        path: path17,
+        path: path18,
         score: 0,
         weight,
         hit: false,
-        message: `${path17} (unparseable expected date)`
+        message: `${path18} (unparseable expected date)`
       };
     }
     if (candidateDate.getFullYear() === expectedDate.getFullYear() && candidateDate.getMonth() === expectedDate.getMonth() && candidateDate.getDate() === expectedDate.getDate()) {
       return {
-        path: path17,
+        path: path18,
         score: 1,
         weight,
         hit: true,
-        message: path17
+        message: path18
       };
     }
     return {
-      path: path17,
+      path: path18,
       score: 0,
       weight,
       hit: false,
-      message: `${path17} (date mismatch: got ${formatDateISO(candidateDate)}, expected ${formatDateISO(expectedDate)})`
+      message: `${path18} (date mismatch: got ${formatDateISO(candidateDate)}, expected ${formatDateISO(expectedDate)})`
     };
   }
   /**
@@ -7799,11 +8052,11 @@ var FieldAccuracyEvaluator = class {
     };
   }
 };
-function resolvePath(obj, path17) {
-  if (!path17 || !obj) {
+function resolvePath(obj, path18) {
+  if (!path18 || !obj) {
     return void 0;
   }
-  const parts = path17.split(/\.|\[|\]/).filter((p) => p.length > 0);
+  const parts = path18.split(/\.|\[|\]/).filter((p) => p.length > 0);
   let current = obj;
   for (const part of parts) {
     if (current === null || current === void 0) {
@@ -8239,7 +8492,7 @@ var ToolTrajectoryEvaluator = class {
 
 // src/evaluation/orchestrator.ts
 var import_node_crypto5 = require("crypto");
-var import_node_path16 = __toESM(require("path"), 1);
+var import_node_path17 = __toESM(require("path"), 1);
 
 // ../../node_modules/.bun/yocto-queue@1.2.2/node_modules/yocto-queue/index.js
 var Node = class {
@@ -9038,7 +9291,7 @@ async function runEvaluatorList(options) {
         });
       }
       if (evaluator.type === "composite") {
-        const evalFileDir = evalCase.guideline_paths[0] ? import_node_path16.default.dirname(evalCase.guideline_paths[0]) : process.cwd();
+        const evalFileDir = evalCase.guideline_paths[0] ? import_node_path17.default.dirname(evalCase.guideline_paths[0]) : process.cwd();
         const createEvaluator = (memberConfig) => {
           switch (memberConfig.type) {
             case "llm_judge":
@@ -9613,6 +9866,7 @@ function createAgentKernel() {
   createAgentKernel,
   createProvider,
   deepEqual,
+  detectFormat,
   ensureVSCodeSubagents,
   executeScript,
   explorationRatio,