@agentv/core 2.1.0 → 2.2.0

package/dist/index.js

@@ -146,8 +146,8 @@ function mergeExecutionMetrics(summary, metrics) {
 }
 
 // src/evaluation/yaml-parser.ts
-import { readFile as readFile5 } from "node:fs/promises";
-import path6 from "node:path";
+import { readFile as readFile6 } from "node:fs/promises";
+import path7 from "node:path";
 import { parse as parse2 } from "yaml";
 
 // src/evaluation/loaders/config-loader.ts
@@ -257,7 +257,6 @@ async function resolveFileReference2(rawValue, searchRoots) {
 }
 
 // src/evaluation/loaders/config-loader.ts
-var SCHEMA_CONFIG_V2 = "agentv-config-v2";
 var ANSI_YELLOW = "\x1B[33m";
 var ANSI_RESET = "\x1B[0m";
 async function loadConfig(evalFilePath, repoRoot) {
@@ -275,13 +274,6 @@ async function loadConfig(evalFilePath, repoRoot) {
         continue;
       }
       const config = parsed;
-      const schema = config.$schema;
-      if (schema !== SCHEMA_CONFIG_V2) {
-        const message = typeof schema === "string" ? `Invalid $schema value '${schema}' in ${configPath}. Expected '${SCHEMA_CONFIG_V2}'` : `Missing required field '$schema' in ${configPath}.
-Please add '$schema: ${SCHEMA_CONFIG_V2}' at the top of the file.`;
-        logWarning(message);
-        continue;
-      }
       const guidelinePatterns = config.guideline_patterns;
       if (guidelinePatterns !== void 0 && !Array.isArray(guidelinePatterns)) {
         logWarning(`Invalid guideline_patterns in ${configPath}, expected array`);
@@ -390,7 +382,8 @@ var ANSI_YELLOW3 = "\x1B[33m";
 var ANSI_RESET3 = "\x1B[0m";
 async function parseEvaluators(rawEvalCase, globalExecution, searchRoots, evalId) {
   const execution = rawEvalCase.execution;
-  const candidateEvaluators = isJsonObject2(execution) ? execution.evaluators ?? rawEvalCase.evaluators : rawEvalCase.evaluators ?? globalExecution?.evaluators;
+  const executionObject = isJsonObject2(execution) ? execution : void 0;
+  const candidateEvaluators = (executionObject ? executionObject.evaluators : void 0) ?? rawEvalCase.evaluators ?? globalExecution?.evaluators;
   if (candidateEvaluators === void 0) {
     return void 0;
   }
@@ -933,6 +926,11 @@ function isValidFieldAggregationType(value) {
   return typeof value === "string" && VALID_FIELD_AGGREGATION_TYPES.has(value);
 }
 
+// src/evaluation/loaders/jsonl-parser.ts
+import { readFile as readFile4 } from "node:fs/promises";
+import path5 from "node:path";
+import { parse as parseYaml } from "yaml";
+
 // src/evaluation/loaders/message-processor.ts
 import { readFile as readFile3 } from "node:fs/promises";
 import path4 from "node:path";
@@ -1193,28 +1191,271 @@ async function processExpectedMessages(options) {
   return segments;
 }
 
-// src/evaluation/formatting/prompt-builder.ts
-import { readFile as readFile4 } from "node:fs/promises";
-import path5 from "node:path";
+// 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 = path5.extname(filePath).toLowerCase();
+  if (ext === ".jsonl") return "jsonl";
+  if (ext === ".yaml" || ext === ".yml") return "yaml";
+  throw new Error(`Unsupported file format: '${ext}'. Supported formats: .yaml, .yml, .jsonl`);
+}
+async function loadSidecarMetadata(jsonlPath, verbose) {
+  const dir = path5.dirname(jsonlPath);
+  const base = path5.basename(jsonlPath, ".jsonl");
+  const sidecarPath = path5.join(dir, `${base}.yaml`);
+  if (!await fileExists2(sidecarPath)) {
+    if (verbose) {
+      logWarning4(`Sidecar metadata file not found: ${sidecarPath} (using defaults)`);
+    }
+    return {};
+  }
+  try {
+    const content = await readFile4(sidecarPath, "utf8");
+    const parsed = parseYaml(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 = path5.resolve(evalFilePath);
+  const repoRootPath = resolveToAbsolutePath(repoRoot);
+  const searchRoots = buildSearchRoots2(absoluteTestPath, repoRootPath);
+  const config = await loadConfig(absoluteTestPath, repoRootPath);
+  const guidelinePatterns = config?.guideline_patterns;
+  const sidecar = await loadSidecarMetadata(absoluteTestPath, verbose);
+  const rawFile = await readFile4(absoluteTestPath, "utf8");
+  const rawCases = parseJsonlContent(rawFile, evalFilePath);
+  const fallbackDataset = path5.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) => path5.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) => path5.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
+import { readFile as readFile5 } from "node:fs/promises";
+import path6 from "node:path";
+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 = path5.resolve(rawPath);
+    const absolutePath = path6.resolve(rawPath);
     if (!await fileExists2(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 readFile4(absolutePath, "utf8")).replace(/\r\n/g, "\n").trim();
+      const content = (await readFile5(absolutePath, "utf8")).replace(/\r\n/g, "\n").trim();
       guidelineParts.push({
         content,
         isFile: true,
-        displayPath: path5.basename(absolutePath)
+        displayPath: path6.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);
@@ -1238,9 +1479,9 @@ 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 });
@@ -1251,7 +1492,7 @@ 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 });
             }
@@ -1405,21 +1646,21 @@ ${guidelineContent.trim()}`);
   }
   return chatPrompt.length > 0 ? chatPrompt : void 0;
 }
-function asString4(value) {
+function asString5(value) {
   return typeof value === "string" ? value : void 0;
 }
-function logWarning4(message) {
-  console.warn(`${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 = path6.resolve(testFilePath);
-    const content = await readFile5(absolutePath, "utf8");
+    const absolutePath = path7.resolve(testFilePath);
+    const content = await readFile6(absolutePath, "utf8");
     const parsed = parse2(content);
     if (!isJsonObject(parsed)) {
       return {};
@@ -1430,21 +1671,25 @@ 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 = path6.resolve(evalFilePath);
+  const absoluteTestPath = path7.resolve(evalFilePath);
   const repoRootPath = resolveToAbsolutePath(repoRoot);
   const searchRoots = buildSearchRoots2(absoluteTestPath, repoRootPath);
   const config = await loadConfig(absoluteTestPath, repoRootPath);
   const guidelinePatterns = config?.guideline_patterns;
-  const rawFile = await readFile5(absoluteTestPath, "utf8");
+  const rawFile = await readFile6(absoluteTestPath, "utf8");
   const parsed = parse2(rawFile);
   if (!isJsonObject(parsed)) {
     throw new Error(`Invalid test file format: ${evalFilePath}`);
   }
   const suite = parsed;
-  const datasetNameFromSuite = asString5(suite.dataset)?.trim();
-  const fallbackDataset = path6.basename(absoluteTestPath).replace(/\.ya?ml$/i, "") || "eval";
+  const datasetNameFromSuite = asString6(suite.dataset)?.trim();
+  const fallbackDataset = path7.basename(absoluteTestPath).replace(/\.ya?ml$/i, "") || "eval";
   const datasetName = datasetNameFromSuite && datasetNameFromSuite.length > 0 ? datasetNameFromSuite : fallbackDataset;
   const rawTestcases = suite.evalcases;
   if (!Array.isArray(rawTestcases)) {
@@ -1452,24 +1697,24 @@ 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;
@@ -1480,7 +1725,7 @@ 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: ${id}`);
       continue;
     }
     const guidelinePaths = [];
@@ -1521,7 +1766,7 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
       evaluators = await parseEvaluators(evalcase, globalExecution, searchRoots, id ?? "unknown");
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
-      logError(`Skipping eval case '${id}': ${message}`);
+      logError2(`Skipping eval case '${id}': ${message}`);
       continue;
     }
     const inlineRubrics = evalcase.rubrics;
@@ -1536,8 +1781,8 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
           };
         }
         return {
-          id: asString5(rubric.id) ?? `rubric-${index + 1}`,
-          description: asString5(rubric.description) ?? "",
+          id: asString6(rubric.id) ?? `rubric-${index + 1}`,
+          description: asString6(rubric.description) ?? "",
           weight: typeof rubric.weight === "number" ? rubric.weight : 1,
           required: typeof rubric.required === "boolean" ? rubric.required : true
         };
@@ -1558,7 +1803,7 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
       }
     }
     const allFilePaths = [
-      ...guidelinePaths.map((guidelinePath) => path6.resolve(guidelinePath)),
+      ...guidelinePaths.map((guidelinePath) => path7.resolve(guidelinePath)),
       ...userFilePaths
     ];
     const testCase = {
@@ -1570,7 +1815,7 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
       input_segments: inputSegments,
       expected_messages: outputSegments,
       reference_answer: referenceAnswer,
-      guideline_paths: guidelinePaths.map((guidelinePath) => path6.resolve(guidelinePath)),
+      guideline_paths: guidelinePaths.map((guidelinePath) => path7.resolve(guidelinePath)),
       guideline_patterns: guidelinePatterns,
       file_paths: allFilePaths,
       expected_outcome: outcome,
@@ -1593,25 +1838,25 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
   }
   return results;
 }
-function asString5(value) {
+function asString6(value) {
   return typeof value === "string" ? value : void 0;
 }
-function logWarning5(message, details) {
+function logWarning6(message, details) {
   if (details && details.length > 0) {
     const detailBlock = details.join("\n");
-    console.warn(`${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}`);
   }
 }
 
@@ -1954,7 +2199,7 @@ import { randomUUID } from "node:crypto";
 import { createWriteStream } from "node:fs";
 import { mkdir, mkdtemp, rm, writeFile } from "node:fs/promises";
 import { tmpdir } from "node:os";
-import path8 from "node:path";
+import path9 from "node:path";
 
 // src/evaluation/providers/claude-code-log-tracker.ts
 var GLOBAL_LOGS_KEY = Symbol.for("agentv.claudeCodeLogs");
@@ -2010,7 +2255,7 @@ function subscribeToClaudeCodeLogEntries(listener) {
 }
 
 // src/evaluation/providers/preread.ts
-import path7 from "node:path";
+import path8 from "node:path";
 function buildPromptDocument(request, inputFiles, options) {
   const parts = [];
   const guidelineFiles = collectGuidelineFiles(
@@ -2033,7 +2278,7 @@ function normalizeInputFiles(inputFiles) {
   }
   const deduped = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = path7.resolve(inputFile);
+    const absolutePath = path8.resolve(inputFile);
     if (!deduped.has(absolutePath)) {
       deduped.set(absolutePath, absolutePath);
     }
@@ -2046,14 +2291,14 @@ function collectGuidelineFiles(inputFiles, guidelinePatterns, overrides) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = path7.resolve(inputFile);
+    const absolutePath = path8.resolve(inputFile);
     if (overrides?.has(absolutePath)) {
       if (!unique.has(absolutePath)) {
         unique.set(absolutePath, absolutePath);
       }
       continue;
     }
-    const normalized = absolutePath.split(path7.sep).join("/");
+    const normalized = absolutePath.split(path8.sep).join("/");
     if (isGuidelineFile(normalized, guidelinePatterns)) {
       if (!unique.has(absolutePath)) {
         unique.set(absolutePath, absolutePath);
@@ -2068,7 +2313,7 @@ function collectInputFiles(inputFiles) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = path7.resolve(inputFile);
+    const absolutePath = path8.resolve(inputFile);
     if (!unique.has(absolutePath)) {
       unique.set(absolutePath, absolutePath);
     }
@@ -2080,7 +2325,7 @@ function buildMandatoryPrereadBlock(guidelineFiles, inputFiles) {
     return "";
   }
   const buildList = (files) => files.map((absolutePath) => {
-    const fileName = path7.basename(absolutePath);
+    const fileName = path8.basename(absolutePath);
     const fileUri = pathToFileUri(absolutePath);
     return `* [${fileName}](${fileUri})`;
   });
@@ -2100,7 +2345,7 @@ ${buildList(inputFiles).join("\n")}.`);
   return sections.join("\n");
 }
 function pathToFileUri(filePath) {
-  const absolutePath = path7.isAbsolute(filePath) ? filePath : path7.resolve(filePath);
+  const absolutePath = path8.isAbsolute(filePath) ? filePath : path8.resolve(filePath);
   const normalizedPath = absolutePath.replace(/\\/g, "/");
   if (/^[a-zA-Z]:\//.test(normalizedPath)) {
     return `file:///${normalizedPath}`;
@@ -2137,7 +2382,7 @@ var ClaudeCodeProvider = class {
     const workspaceRoot = await this.createWorkspace();
     const logger = await this.createStreamLogger(request).catch(() => void 0);
     try {
-      const promptFile = path8.join(workspaceRoot, PROMPT_FILENAME);
+      const promptFile = path9.join(workspaceRoot, PROMPT_FILENAME);
       await writeFile(promptFile, request.question, "utf8");
       const args = this.buildClaudeCodeArgs(request.question, inputFiles);
       const cwd = this.resolveCwd();
@@ -2185,7 +2430,7 @@ var ClaudeCodeProvider = class {
     if (!this.config.cwd) {
       return process.cwd();
     }
-    return path8.resolve(this.config.cwd);
+    return path9.resolve(this.config.cwd);
   }
   buildClaudeCodeArgs(prompt, inputFiles) {
     const args = [];
@@ -2242,7 +2487,7 @@ ${filesContext}`;
     }
   }
   async createWorkspace() {
-    return await mkdtemp(path8.join(tmpdir(), WORKSPACE_PREFIX));
+    return await mkdtemp(path9.join(tmpdir(), WORKSPACE_PREFIX));
   }
   async cleanupWorkspace(workspaceRoot) {
     try {
@@ -2256,9 +2501,9 @@ ${filesContext}`;
       return void 0;
     }
     if (this.config.logDir) {
-      return path8.resolve(this.config.logDir);
+      return path9.resolve(this.config.logDir);
     }
-    return path8.join(process.cwd(), ".agentv", "logs", "claude-code");
+    return path9.join(process.cwd(), ".agentv", "logs", "claude-code");
   }
   async createStreamLogger(request) {
     const logDir = this.resolveLogDirectory();
@@ -2272,7 +2517,7 @@ ${filesContext}`;
       console.warn(`Skipping Claude Code stream logging (could not create ${logDir}): ${message}`);
       return void 0;
     }
-    const filePath = path8.join(logDir, buildLogFilename(request, this.targetName));
+    const filePath = path9.join(logDir, buildLogFilename(request, this.targetName));
     try {
       const logger = await ClaudeCodeStreamLogger.create({
         filePath,
@@ -2677,10 +2922,10 @@ function escapeShellArg(arg) {
 }
 async function defaultClaudeCodeRunner(options) {
   const tempId = randomUUID();
-  const stdoutFile = path8.join(tmpdir(), `agentv-cc-${tempId}-stdout`);
-  const stderrFile = path8.join(tmpdir(), `agentv-cc-${tempId}-stderr`);
-  const exitFile = path8.join(tmpdir(), `agentv-cc-${tempId}-exit`);
-  const pidFile = path8.join(tmpdir(), `agentv-cc-${tempId}-pid`);
+  const stdoutFile = path9.join(tmpdir(), `agentv-cc-${tempId}-stdout`);
+  const stderrFile = path9.join(tmpdir(), `agentv-cc-${tempId}-stderr`);
+  const exitFile = path9.join(tmpdir(), `agentv-cc-${tempId}-exit`);
+  const pidFile = path9.join(tmpdir(), `agentv-cc-${tempId}-pid`);
   try {
     return await runClaudeCodeWithTempFiles(options, stdoutFile, stderrFile, exitFile, pidFile);
   } finally {
@@ -2720,8 +2965,8 @@ async function runClaudeCodeWithTempFiles(options, stdoutFile, stderrFile, exitF
   let lastStdoutSize = 0;
   const readFileIfExists = async (filePath) => {
     try {
-      const { readFile: readFile7 } = await import("node:fs/promises");
-      return await readFile7(filePath, "utf8");
+      const { readFile: readFile8 } = await import("node:fs/promises");
+      return await readFile8(filePath, "utf8");
     } catch {
       return "";
     }
@@ -2796,7 +3041,7 @@ async function runClaudeCodeWithTempFiles(options, stdoutFile, stderrFile, exitF
 import { exec as execWithCallback } from "node:child_process";
 import fs from "node:fs/promises";
 import os from "node:os";
-import path9 from "node:path";
+import path10 from "node:path";
 import { promisify } from "node:util";
 import { z } from "zod";
 var ToolCallSchema = z.object({
@@ -3253,7 +3498,7 @@ function normalizeInputFiles2(inputFiles) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = path9.resolve(inputFile);
+    const absolutePath = path10.resolve(inputFile);
     if (!unique.has(absolutePath)) {
       unique.set(absolutePath, absolutePath);
     }
@@ -3267,7 +3512,7 @@ function formatFileList(files, template) {
   const formatter = template ?? "{path}";
   return files.map((filePath) => {
     const escapedPath = shellEscape(filePath);
-    const escapedName = shellEscape(path9.basename(filePath));
+    const escapedName = shellEscape(path10.basename(filePath));
     return formatter.replaceAll("{path}", escapedPath).replaceAll("{basename}", escapedName);
   }).join(" ");
 }
@@ -3291,7 +3536,7 @@ function generateOutputFilePath(evalCaseId, extension = ".json") {
   const safeEvalId = evalCaseId || "unknown";
   const timestamp = Date.now();
   const random = Math.random().toString(36).substring(2, 9);
-  return path9.join(os.tmpdir(), `agentv-${safeEvalId}-${timestamp}-${random}${extension}`);
+  return path10.join(os.tmpdir(), `agentv-${safeEvalId}-${timestamp}-${random}${extension}`);
 }
 function formatTimeoutSuffix2(timeoutMs) {
   if (!timeoutMs || timeoutMs <= 0) {
@@ -3307,7 +3552,7 @@ import { randomUUID as randomUUID2 } from "node:crypto";
 import { constants as constants2, createWriteStream as createWriteStream2 } from "node:fs";
 import { access as access2, mkdir as mkdir2, mkdtemp as mkdtemp2, rm as rm2, writeFile as writeFile2 } from "node:fs/promises";
 import { tmpdir as tmpdir2 } from "node:os";
-import path10 from "node:path";
+import path11 from "node:path";
 import { promisify as promisify2 } from "node:util";
 
 // src/evaluation/providers/codex-log-tracker.ts
@@ -3402,7 +3647,7 @@ var CodexProvider = class {
       const promptContent = `${systemPrompt}
 
 ${basePrompt}`;
-      const promptFile = path10.join(workspaceRoot, PROMPT_FILENAME2);
+      const promptFile = path11.join(workspaceRoot, PROMPT_FILENAME2);
       await writeFile2(promptFile, promptContent, "utf8");
       const args = this.buildCodexArgs();
       const cwd = this.resolveCwd(workspaceRoot);
@@ -3452,7 +3697,7 @@ ${basePrompt}`;
     if (!this.config.cwd) {
       return workspaceRoot;
     }
-    return path10.resolve(this.config.cwd);
+    return path11.resolve(this.config.cwd);
   }
   buildCodexArgs() {
     const args = [
@@ -3494,7 +3739,7 @@ ${basePrompt}`;
     }
   }
   async createWorkspace() {
-    return await mkdtemp2(path10.join(tmpdir2(), WORKSPACE_PREFIX2));
+    return await mkdtemp2(path11.join(tmpdir2(), WORKSPACE_PREFIX2));
   }
   async cleanupWorkspace(workspaceRoot) {
     try {
@@ -3508,9 +3753,9 @@ ${basePrompt}`;
       return void 0;
     }
     if (this.config.logDir) {
-      return path10.resolve(this.config.logDir);
+      return path11.resolve(this.config.logDir);
     }
-    return path10.join(process.cwd(), ".agentv", "logs", "codex");
+    return path11.join(process.cwd(), ".agentv", "logs", "codex");
   }
   async createStreamLogger(request) {
     const logDir = this.resolveLogDirectory();
@@ -3524,7 +3769,7 @@ ${basePrompt}`;
       console.warn(`Skipping Codex stream logging (could not create ${logDir}): ${message}`);
       return void 0;
     }
-    const filePath = path10.join(logDir, buildLogFilename2(request, this.targetName));
+    const filePath = path11.join(logDir, buildLogFilename2(request, this.targetName));
     try {
       const logger = await CodexStreamLogger.create({
         filePath,
@@ -3739,7 +3984,7 @@ function tryParseJsonValue2(rawLine) {
 async function locateExecutable(candidate) {
   const includesPathSeparator = candidate.includes("/") || candidate.includes("\\");
   if (includesPathSeparator) {
-    const resolved = path10.isAbsolute(candidate) ? candidate : path10.resolve(candidate);
+    const resolved = path11.isAbsolute(candidate) ? candidate : path11.resolve(candidate);
     const executablePath = await ensureWindowsExecutableVariant(resolved);
     await access2(executablePath, constants2.F_OK);
     return executablePath;
@@ -4252,7 +4497,7 @@ import { randomUUID as randomUUID3 } from "node:crypto";
 import { createWriteStream as createWriteStream3 } from "node:fs";
 import { mkdir as mkdir3, mkdtemp as mkdtemp3, rm as rm3, writeFile as writeFile3 } from "node:fs/promises";
 import { tmpdir as tmpdir3 } from "node:os";
-import path11 from "node:path";
+import path12 from "node:path";
 
 // src/evaluation/providers/pi-log-tracker.ts
 var GLOBAL_LOGS_KEY3 = Symbol.for("agentv.piLogs");
@@ -4336,7 +4581,7 @@ var PiCodingAgentProvider = class {
     const workspaceRoot = await this.createWorkspace();
     const logger = await this.createStreamLogger(request).catch(() => void 0);
     try {
-      const promptFile = path11.join(workspaceRoot, PROMPT_FILENAME3);
+      const promptFile = path12.join(workspaceRoot, PROMPT_FILENAME3);
       await writeFile3(promptFile, request.question, "utf8");
       const args = this.buildPiArgs(request.question, inputFiles);
       const cwd = this.resolveCwd(workspaceRoot);
@@ -4378,7 +4623,7 @@ var PiCodingAgentProvider = class {
     if (!this.config.cwd) {
       return workspaceRoot;
     }
-    return path11.resolve(this.config.cwd);
+    return path12.resolve(this.config.cwd);
   }
   buildPiArgs(prompt, inputFiles) {
     const args = [];
@@ -4467,7 +4712,7 @@ ${prompt}`;
     return env;
   }
   async createWorkspace() {
-    return await mkdtemp3(path11.join(tmpdir3(), WORKSPACE_PREFIX3));
+    return await mkdtemp3(path12.join(tmpdir3(), WORKSPACE_PREFIX3));
   }
   async cleanupWorkspace(workspaceRoot) {
     try {
@@ -4477,9 +4722,9 @@ ${prompt}`;
   }
   resolveLogDirectory() {
     if (this.config.logDir) {
-      return path11.resolve(this.config.logDir);
+      return path12.resolve(this.config.logDir);
     }
-    return path11.join(process.cwd(), ".agentv", "logs", "pi-coding-agent");
+    return path12.join(process.cwd(), ".agentv", "logs", "pi-coding-agent");
   }
   async createStreamLogger(request) {
     const logDir = this.resolveLogDirectory();
@@ -4493,7 +4738,7 @@ ${prompt}`;
       console.warn(`Skipping Pi stream logging (could not create ${logDir}): ${message}`);
       return void 0;
     }
-    const filePath = path11.join(logDir, buildLogFilename3(request, this.targetName));
+    const filePath = path12.join(logDir, buildLogFilename3(request, this.targetName));
     try {
       const logger = await PiStreamLogger.create({
         filePath,
@@ -4926,7 +5171,7 @@ async function defaultPiRunner(options) {
 }
 
 // src/evaluation/providers/vscode.ts
-import path12 from "node:path";
+import path13 from "node:path";
 import {
   dispatchAgentSession,
   dispatchBatchAgent,
@@ -5101,7 +5346,7 @@ function buildMandatoryPrereadBlock2(guidelineFiles, attachmentFiles) {
     return "";
   }
   const buildList = (files) => files.map((absolutePath) => {
-    const fileName = path12.basename(absolutePath);
+    const fileName = path13.basename(absolutePath);
     const fileUri = pathToFileUri2(absolutePath);
     return `* [${fileName}](${fileUri})`;
   });
@@ -5126,8 +5371,8 @@ function collectGuidelineFiles2(attachments, guidelinePatterns) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const attachment of attachments) {
-    const absolutePath = path12.resolve(attachment);
-    const normalized = absolutePath.split(path12.sep).join("/");
+    const absolutePath = path13.resolve(attachment);
+    const normalized = absolutePath.split(path13.sep).join("/");
     if (isGuidelineFile(normalized, guidelinePatterns)) {
       if (!unique.has(absolutePath)) {
         unique.set(absolutePath, absolutePath);
@@ -5142,7 +5387,7 @@ function collectAttachmentFiles(attachments) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const attachment of attachments) {
-    const absolutePath = path12.resolve(attachment);
+    const absolutePath = path13.resolve(attachment);
     if (!unique.has(absolutePath)) {
       unique.set(absolutePath, absolutePath);
     }
@@ -5150,7 +5395,7 @@ function collectAttachmentFiles(attachments) {
   return Array.from(unique.values());
 }
 function pathToFileUri2(filePath) {
-  const absolutePath = path12.isAbsolute(filePath) ? filePath : path12.resolve(filePath);
+  const absolutePath = path13.isAbsolute(filePath) ? filePath : path13.resolve(filePath);
   const normalizedPath = absolutePath.replace(/\\/g, "/");
   if (/^[a-zA-Z]:\//.test(normalizedPath)) {
     return `file:///${normalizedPath}`;
@@ -5163,7 +5408,7 @@ function normalizeAttachments(attachments) {
   }
   const deduped = /* @__PURE__ */ new Set();
   for (const attachment of attachments) {
-    deduped.add(path12.resolve(attachment));
+    deduped.add(path13.resolve(attachment));
   }
   return Array.from(deduped);
 }
@@ -5172,7 +5417,7 @@ function mergeAttachments(all) {
   for (const list of all) {
     if (!list) continue;
     for (const inputFile of list) {
-      deduped.add(path12.resolve(inputFile));
+      deduped.add(path13.resolve(inputFile));
     }
   }
   return deduped.size > 0 ? Array.from(deduped) : void 0;
@@ -5220,8 +5465,8 @@ total unlocked subagents available: ${result.created.length + result.skippedExis
 
 // src/evaluation/providers/targets-file.ts
 import { constants as constants3 } from "node:fs";
-import { access as access3, readFile as readFile6 } from "node:fs/promises";
-import path13 from "node:path";
+import { access as access3, readFile as readFile7 } from "node:fs/promises";
+import path14 from "node:path";
 import { parse as parse3 } from "yaml";
 function isRecord(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value);
@@ -5258,11 +5503,11 @@ async function fileExists3(filePath) {
   }
 }
 async function readTargetDefinitions(filePath) {
-  const absolutePath = path13.resolve(filePath);
+  const absolutePath = path14.resolve(filePath);
   if (!await fileExists3(absolutePath)) {
     throw new Error(`targets.yaml not found at ${absolutePath}`);
   }
-  const raw = await readFile6(absolutePath, "utf8");
+  const raw = await readFile7(absolutePath, "utf8");
   const parsed = parse3(raw);
   if (!isRecord(parsed)) {
     throw new Error(`targets.yaml at ${absolutePath} must be a YAML object with a 'targets' field`);
@@ -5469,15 +5714,15 @@ async function execFileWithStdinNode(argv, stdinPayload, options) {
   });
 }
 async function execShellWithStdin(command, stdinPayload, options = {}) {
-  const { mkdir: mkdir4, readFile: readFile7, rm: rm4, writeFile: writeFile4 } = await import("node:fs/promises");
+  const { mkdir: mkdir4, readFile: readFile8, rm: rm4, writeFile: writeFile4 } = await import("node:fs/promises");
   const { tmpdir: tmpdir4 } = await import("node:os");
-  const path15 = await import("node:path");
+  const path16 = await import("node:path");
   const { randomUUID: randomUUID4 } = await import("node:crypto");
-  const dir = path15.join(tmpdir4(), `agentv-exec-${randomUUID4()}`);
+  const dir = path16.join(tmpdir4(), `agentv-exec-${randomUUID4()}`);
   await mkdir4(dir, { recursive: true });
-  const stdinPath = path15.join(dir, "stdin.txt");
-  const stdoutPath = path15.join(dir, "stdout.txt");
-  const stderrPath = path15.join(dir, "stderr.txt");
+  const stdinPath = path16.join(dir, "stdin.txt");
+  const stdoutPath = path16.join(dir, "stdout.txt");
+  const stderrPath = path16.join(dir, "stderr.txt");
   await writeFile4(stdinPath, stdinPayload, "utf8");
   const wrappedCommand = process.platform === "win32" ? `(${command}) < ${shellEscapePath(stdinPath)} > ${shellEscapePath(stdoutPath)} 2> ${shellEscapePath(stderrPath)}` : `(${command}) < ${shellEscapePath(stdinPath)} > ${shellEscapePath(stdoutPath)} 2> ${shellEscapePath(stderrPath)}`;
   const { spawn: spawn4 } = await import("node:child_process");
@@ -5507,8 +5752,8 @@ async function execShellWithStdin(command, stdinPayload, options = {}) {
         resolve(code ?? 0);
       });
     });
-    const stdout = (await readFile7(stdoutPath, "utf8")).replace(/\r\n/g, "\n");
-    const stderr = (await readFile7(stderrPath, "utf8")).replace(/\r\n/g, "\n");
+    const stdout = (await readFile8(stdoutPath, "utf8")).replace(/\r\n/g, "\n");
+    const stderr = (await readFile8(stderrPath, "utf8")).replace(/\r\n/g, "\n");
     return { stdout, stderr, exitCode };
   } finally {
     await rm4(dir, { recursive: true, force: true });
@@ -5780,7 +6025,7 @@ var CodeEvaluator = class {
       outputMessages: context.outputMessages ?? null,
       guidelineFiles: context.evalCase.guideline_paths,
       inputFiles: context.evalCase.file_paths.filter(
-        (path15) => !context.evalCase.guideline_paths.includes(path15)
+        (path16) => !context.evalCase.guideline_paths.includes(path16)
       ),
       inputMessages: context.evalCase.input_messages,
       traceSummary: context.traceSummary ?? null,
@@ -6539,115 +6784,115 @@ var FieldAccuracyEvaluator = class {
    * Evaluate a single field against the expected value.
    */
   evaluateField(fieldConfig, candidateData, expectedData) {
-    const { path: path15, match, required = true, weight = 1 } = fieldConfig;
-    const candidateValue = resolvePath(candidateData, path15);
-    const expectedValue = resolvePath(expectedData, path15);
+    const { path: path16, match, required = true, weight = 1 } = fieldConfig;
+    const candidateValue = resolvePath(candidateData, path16);
+    const expectedValue = resolvePath(expectedData, path16);
     if (expectedValue === void 0) {
       return {
-        path: path15,
+        path: path16,
         score: 1,
         // No expected value means no comparison needed
         weight,
         hit: true,
-        message: `${path15}: no expected value`
+        message: `${path16}: no expected value`
       };
     }
     if (candidateValue === void 0) {
       if (required) {
         return {
-          path: path15,
+          path: path16,
           score: 0,
           weight,
           hit: false,
-          message: `${path15} (required, missing)`
+          message: `${path16} (required, missing)`
         };
       }
       return {
-        path: path15,
+        path: path16,
         score: 1,
         // Don't penalize missing optional fields
         weight: 0,
         // Zero weight means it won't affect the score
         hit: true,
-        message: `${path15}: optional field missing`
+        message: `${path16}: optional field missing`
       };
     }
     switch (match) {
       case "exact":
-        return this.compareExact(path15, candidateValue, expectedValue, weight);
+        return this.compareExact(path16, candidateValue, expectedValue, weight);
       case "numeric_tolerance":
         return this.compareNumericTolerance(
-          path15,
+          path16,
           candidateValue,
           expectedValue,
           fieldConfig,
           weight
         );
       case "date":
-        return this.compareDate(path15, candidateValue, expectedValue, fieldConfig, weight);
+        return this.compareDate(path16, candidateValue, expectedValue, fieldConfig, weight);
       default:
         return {
-          path: path15,
+          path: path16,
           score: 0,
           weight,
           hit: false,
-          message: `${path15}: unknown match type "${match}"`
+          message: `${path16}: unknown match type "${match}"`
         };
     }
   }
   /**
    * Exact equality comparison.
    */
-  compareExact(path15, candidateValue, expectedValue, weight) {
+  compareExact(path16, candidateValue, expectedValue, weight) {
     if (deepEqual(candidateValue, expectedValue)) {
       return {
-        path: path15,
+        path: path16,
         score: 1,
         weight,
         hit: true,
-        message: path15
+        message: path16
       };
     }
     if (typeof candidateValue !== typeof expectedValue) {
       return {
-        path: path15,
+        path: path16,
         score: 0,
         weight,
         hit: false,
-        message: `${path15} (type mismatch: got ${typeof candidateValue}, expected ${typeof expectedValue})`
+        message: `${path16} (type mismatch: got ${typeof candidateValue}, expected ${typeof expectedValue})`
       };
     }
     return {
-      path: path15,
+      path: path16,
       score: 0,
       weight,
       hit: false,
-      message: `${path15} (value mismatch)`
+      message: `${path16} (value mismatch)`
     };
   }
   /**
    * Numeric comparison with absolute or relative tolerance.
    */
-  compareNumericTolerance(path15, candidateValue, expectedValue, fieldConfig, weight) {
+  compareNumericTolerance(path16, candidateValue, expectedValue, fieldConfig, weight) {
     const { tolerance = 0, relative = false } = fieldConfig;
     const candidateNum = toNumber(candidateValue);
     const expectedNum = toNumber(expectedValue);
     if (candidateNum === null || expectedNum === null) {
       return {
-        path: path15,
+        path: path16,
         score: 0,
         weight,
         hit: false,
-        message: `${path15} (non-numeric value)`
+        message: `${path16} (non-numeric value)`
       };
     }
     if (!Number.isFinite(candidateNum) || !Number.isFinite(expectedNum)) {
       return {
-        path: path15,
+        path: path16,
         score: 0,
         weight,
         hit: false,
-        message: `${path15} (invalid numeric value)`
+        message: `${path16} (invalid numeric value)`
       };
     }
     const diff = Math.abs(candidateNum - expectedNum);
@@ -6660,61 +6905,61 @@ var FieldAccuracyEvaluator = class {
     }
     if (withinTolerance) {
       return {
-        path: path15,
+        path: path16,
         score: 1,
         weight,
         hit: true,
-        message: `${path15} (within tolerance: diff=${diff.toFixed(2)})`
+        message: `${path16} (within tolerance: diff=${diff.toFixed(2)})`
       };
     }
     return {
-      path: path15,
+      path: path16,
       score: 0,
       weight,
       hit: false,
-      message: `${path15} (outside tolerance: diff=${diff.toFixed(2)}, tolerance=${tolerance})`
+      message: `${path16} (outside tolerance: diff=${diff.toFixed(2)}, tolerance=${tolerance})`
     };
   }
   /**
    * Date comparison with format normalization.
    */
-  compareDate(path15, candidateValue, expectedValue, fieldConfig, weight) {
+  compareDate(path16, candidateValue, expectedValue, fieldConfig, weight) {
     const formats = fieldConfig.formats ?? DEFAULT_DATE_FORMATS;
     const candidateDate = parseDate(String(candidateValue), formats);
     const expectedDate = parseDate(String(expectedValue), formats);
     if (candidateDate === null) {
       return {
-        path: path15,
+        path: path16,
         score: 0,
         weight,
         hit: false,
-        message: `${path15} (unparseable candidate date)`
+        message: `${path16} (unparseable candidate date)`
       };
     }
     if (expectedDate === null) {
       return {
-        path: path15,
+        path: path16,
         score: 0,
         weight,
         hit: false,
-        message: `${path15} (unparseable expected date)`
+        message: `${path16} (unparseable expected date)`
       };
     }
     if (candidateDate.getFullYear() === expectedDate.getFullYear() && candidateDate.getMonth() === expectedDate.getMonth() && candidateDate.getDate() === expectedDate.getDate()) {
       return {
-        path: path15,
+        path: path16,
         score: 1,
         weight,
         hit: true,
-        message: path15
+        message: path16
       };
     }
     return {
-      path: path15,
+      path: path16,
       score: 0,
       weight,
       hit: false,
-      message: `${path15} (date mismatch: got ${formatDateISO(candidateDate)}, expected ${formatDateISO(expectedDate)})`
+      message: `${path16} (date mismatch: got ${formatDateISO(candidateDate)}, expected ${formatDateISO(expectedDate)})`
     };
   }
   /**
@@ -6754,11 +6999,11 @@ var FieldAccuracyEvaluator = class {
     };
   }
 };
-function resolvePath(obj, path15) {
-  if (!path15 || !obj) {
+function resolvePath(obj, path16) {
+  if (!path16 || !obj) {
     return void 0;
   }
-  const parts = path15.split(/\.|\[|\]/).filter((p) => p.length > 0);
+  const parts = path16.split(/\.|\[|\]/).filter((p) => p.length > 0);
   let current = obj;
   for (const part of parts) {
     if (current === null || current === void 0) {
@@ -7194,7 +7439,7 @@ var ToolTrajectoryEvaluator = class {
 
 // src/evaluation/orchestrator.ts
 import { createHash } from "node:crypto";
-import path14 from "node:path";
+import path15 from "node:path";
 
 // ../../node_modules/.bun/yocto-queue@1.2.2/node_modules/yocto-queue/index.js
 var Node = class {
@@ -7993,7 +8238,7 @@ async function runEvaluatorList(options) {
         });
       }
       if (evaluator.type === "composite") {
-        const evalFileDir = evalCase.guideline_paths[0] ? path14.dirname(evalCase.guideline_paths[0]) : process.cwd();
+        const evalFileDir = evalCase.guideline_paths[0] ? path15.dirname(evalCase.guideline_paths[0]) : process.cwd();
         const createEvaluator = (memberConfig) => {
           switch (memberConfig.type) {
             case "llm_judge":
@@ -8567,6 +8812,7 @@ export {
   createAgentKernel,
   createProvider,
   deepEqual,
+  detectFormat,
   ensureVSCodeSubagents,
   executeScript,
   explorationRatio,