@agentv/core 0.11.0 → 0.14.2

package/dist/index.js

@@ -9,7 +9,7 @@ import {
   readTextFile,
   resolveFileReference,
   resolveTargetDefinition
-} from "./chunk-YQBJAT5I.js";
+} from "./chunk-IOCVST3R.js";
 
 // src/evaluation/types.ts
 var TEST_MESSAGE_ROLE_VALUES = ["system", "user", "assistant", "tool"];
@@ -62,48 +62,197 @@ function getHitCount(result) {
 }
 
 // src/evaluation/yaml-parser.ts
+import { readFile as readFile4 } from "node:fs/promises";
+import path6 from "node:path";
+import { parse as parse2 } from "yaml";
+
+// src/evaluation/formatting/segment-formatter.ts
+function extractCodeBlocks(segments) {
+  const CODE_BLOCK_PATTERN = /```[\s\S]*?```/g;
+  const codeBlocks = [];
+  for (const segment of segments) {
+    const typeValue = segment["type"];
+    if (typeof typeValue !== "string" || typeValue !== "text") {
+      continue;
+    }
+    const textValue = segment["value"];
+    if (typeof textValue !== "string") {
+      continue;
+    }
+    const matches = textValue.match(CODE_BLOCK_PATTERN);
+    if (matches) {
+      codeBlocks.push(...matches);
+    }
+  }
+  return codeBlocks;
+}
+function formatFileContents(parts) {
+  const fileCount = parts.filter((p) => p.isFile).length;
+  if (fileCount > 0) {
+    return parts.map((part) => {
+      if (part.isFile && part.displayPath) {
+        return `<file path="${part.displayPath}">
+${part.content}
+</file>`;
+      }
+      return part.content;
+    }).join("\n\n");
+  }
+  return parts.map((p) => p.content).join(" ");
+}
+function formatSegment(segment) {
+  const type = asString(segment.type);
+  if (type === "text") {
+    return asString(segment.value);
+  }
+  if (type === "guideline_ref") {
+    const refPath = asString(segment.path);
+    return refPath ? `<Attached: ${refPath}>` : void 0;
+  }
+  if (type === "file") {
+    const text = asString(segment.text);
+    const filePath = asString(segment.path);
+    if (text && filePath) {
+      return formatFileContents([{ content: text.trim(), isFile: true, displayPath: filePath }]);
+    }
+  }
+  return void 0;
+}
+function hasVisibleContent(segments) {
+  return segments.some((segment) => {
+    const type = asString(segment.type);
+    if (type === "text") {
+      const value = asString(segment.value);
+      return value !== void 0 && value.trim().length > 0;
+    }
+    if (type === "guideline_ref") {
+      return false;
+    }
+    if (type === "file") {
+      const text = asString(segment.text);
+      return text !== void 0 && text.trim().length > 0;
+    }
+    return false;
+  });
+}
+function asString(value) {
+  return typeof value === "string" ? value : void 0;
+}
+
+// src/evaluation/loaders/config-loader.ts
 import micromatch from "micromatch";
+import { readFile } from "node:fs/promises";
+import path2 from "node:path";
+import { parse } from "yaml";
+
+// src/evaluation/loaders/file-resolver.ts
 import { constants } from "node:fs";
-import { access, readFile } from "node:fs/promises";
+import { access } from "node:fs/promises";
 import path from "node:path";
-import { fileURLToPath } from "node:url";
-import { parse } from "yaml";
-var CODE_BLOCK_PATTERN = /```[\s\S]*?```/g;
-var ANSI_YELLOW = "\x1B[33m";
-var ANSI_RESET = "\x1B[0m";
-var SCHEMA_EVAL_V2 = "agentv-eval-v2";
-var SCHEMA_CONFIG_V2 = "agentv-config-v2";
-async function readTestSuiteMetadata(testFilePath) {
+async function fileExists2(absolutePath) {
   try {
-    const absolutePath = path.resolve(testFilePath);
-    const content = await readFile(absolutePath, "utf8");
-    const parsed = parse(content);
-    if (!isJsonObject(parsed)) {
-      return {};
-    }
-    return { target: extractTargetFromSuite(parsed) };
+    await access(absolutePath, constants.F_OK);
+    return true;
   } catch {
-    return {};
+    return false;
   }
 }
-function extractTargetFromSuite(suite) {
-  const execution = suite.execution;
-  if (execution && typeof execution === "object" && !Array.isArray(execution)) {
-    const executionTarget = execution.target;
-    if (typeof executionTarget === "string" && executionTarget.trim().length > 0) {
-      return executionTarget.trim();
+function resolveToAbsolutePath(candidate) {
+  if (candidate instanceof URL) {
+    return new URL(candidate).pathname;
+  }
+  if (typeof candidate === "string") {
+    if (candidate.startsWith("file://")) {
+      return new URL(candidate).pathname;
     }
+    return path.resolve(candidate);
   }
-  const targetValue = suite.target;
-  if (typeof targetValue === "string" && targetValue.trim().length > 0) {
-    return targetValue.trim();
+  throw new TypeError("Unsupported repoRoot value. Expected string or URL.");
+}
+function buildDirectoryChain2(filePath, repoRoot) {
+  const directories = [];
+  const seen = /* @__PURE__ */ new Set();
+  const boundary = path.resolve(repoRoot);
+  let current = path.resolve(path.dirname(filePath));
+  while (current !== void 0) {
+    if (!seen.has(current)) {
+      directories.push(current);
+      seen.add(current);
+    }
+    if (current === boundary) {
+      break;
+    }
+    const parent = path.dirname(current);
+    if (parent === current) {
+      break;
+    }
+    current = parent;
   }
-  return void 0;
+  if (!seen.has(boundary)) {
+    directories.push(boundary);
+  }
+  return directories;
+}
+function buildSearchRoots2(evalPath, repoRoot) {
+  const uniqueRoots = [];
+  const addRoot = (root) => {
+    const normalized = path.resolve(root);
+    if (!uniqueRoots.includes(normalized)) {
+      uniqueRoots.push(normalized);
+    }
+  };
+  let currentDir = path.dirname(evalPath);
+  let reachedBoundary = false;
+  while (!reachedBoundary) {
+    addRoot(currentDir);
+    const parentDir = path.dirname(currentDir);
+    if (currentDir === repoRoot || parentDir === currentDir) {
+      reachedBoundary = true;
+    } else {
+      currentDir = parentDir;
+    }
+  }
+  addRoot(repoRoot);
+  addRoot(process.cwd());
+  return uniqueRoots;
 }
+function trimLeadingSeparators(value) {
+  const trimmed = value.replace(/^[/\\]+/, "");
+  return trimmed.length > 0 ? trimmed : value;
+}
+async function resolveFileReference2(rawValue, searchRoots) {
+  const displayPath = trimLeadingSeparators(rawValue);
+  const potentialPaths = [];
+  if (path.isAbsolute(rawValue)) {
+    potentialPaths.push(path.normalize(rawValue));
+  }
+  for (const base of searchRoots) {
+    potentialPaths.push(path.resolve(base, displayPath));
+  }
+  const attempted = [];
+  const seen = /* @__PURE__ */ new Set();
+  for (const candidate of potentialPaths) {
+    const absoluteCandidate = path.resolve(candidate);
+    if (seen.has(absoluteCandidate)) {
+      continue;
+    }
+    seen.add(absoluteCandidate);
+    attempted.push(absoluteCandidate);
+    if (await fileExists2(absoluteCandidate)) {
+      return { displayPath, resolvedPath: absoluteCandidate, attempted };
+    }
+  }
+  return { displayPath, attempted };
+}
+
+// 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) {
-  const directories = buildDirectoryChain(evalFilePath, repoRoot);
+  const directories = buildDirectoryChain2(evalFilePath, repoRoot);
   for (const directory of directories) {
-    const configPath = path.join(directory, ".agentv", "config.yaml");
+    const configPath = path2.join(directory, ".agentv", "config.yaml");
     if (!await fileExists2(configPath)) {
       continue;
     }
@@ -146,24 +295,134 @@ function isGuidelineFile(filePath, patterns) {
   const patternsToUse = patterns ?? [];
   return micromatch.isMatch(normalized, patternsToUse);
 }
-function extractCodeBlocks(segments) {
-  const codeBlocks = [];
-  for (const segment of segments) {
-    const typeValue = segment["type"];
-    if (typeof typeValue !== "string" || typeValue !== "text") {
+function extractTargetFromSuite(suite) {
+  const execution = suite.execution;
+  if (execution && typeof execution === "object" && !Array.isArray(execution)) {
+    const executionTarget = execution.target;
+    if (typeof executionTarget === "string" && executionTarget.trim().length > 0) {
+      return executionTarget.trim();
+    }
+  }
+  const targetValue = suite.target;
+  if (typeof targetValue === "string" && targetValue.trim().length > 0) {
+    return targetValue.trim();
+  }
+  return void 0;
+}
+function logWarning(message) {
+  console.warn(`${ANSI_YELLOW}Warning: ${message}${ANSI_RESET}`);
+}
+
+// src/evaluation/loaders/evaluator-parser.ts
+import path3 from "node:path";
+var ANSI_YELLOW2 = "\x1B[33m";
+var ANSI_RESET2 = "\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;
+  if (candidateEvaluators === void 0) {
+    return void 0;
+  }
+  if (!Array.isArray(candidateEvaluators)) {
+    logWarning2(`Skipping evaluators for '${evalId}': expected array`);
+    return void 0;
+  }
+  const evaluators = [];
+  for (const rawEvaluator of candidateEvaluators) {
+    if (!isJsonObject2(rawEvaluator)) {
+      logWarning2(`Skipping invalid evaluator entry for '${evalId}' (expected object)`);
       continue;
     }
-    const textValue = segment["value"];
-    if (typeof textValue !== "string") {
+    const name = asString2(rawEvaluator.name);
+    const typeValue = rawEvaluator.type;
+    if (!name || !isEvaluatorKind(typeValue)) {
+      logWarning2(`Skipping evaluator with invalid name/type in '${evalId}'`);
       continue;
     }
-    const matches = textValue.match(CODE_BLOCK_PATTERN);
-    if (matches) {
-      codeBlocks.push(...matches);
+    if (typeValue === "code") {
+      const script = asString2(rawEvaluator.script);
+      if (!script) {
+        logWarning2(`Skipping code evaluator '${name}' in '${evalId}': missing script`);
+        continue;
+      }
+      const cwd = asString2(rawEvaluator.cwd);
+      let resolvedCwd;
+      if (cwd) {
+        const resolved = await resolveFileReference2(cwd, searchRoots);
+        if (resolved.resolvedPath) {
+          resolvedCwd = path3.resolve(resolved.resolvedPath);
+        } else {
+          logWarning2(
+            `Code evaluator '${name}' in '${evalId}': cwd not found (${resolved.displayPath})`,
+            resolved.attempted.length > 0 ? resolved.attempted.map((attempt) => `  Tried: ${attempt}`) : void 0
+          );
+        }
+      } else {
+        resolvedCwd = searchRoots[0];
+      }
+      evaluators.push({
+        name,
+        type: "code",
+        script,
+        cwd,
+        resolvedCwd
+      });
+      continue;
+    }
+    const prompt = asString2(rawEvaluator.prompt);
+    let promptPath;
+    if (prompt) {
+      const resolved = await resolveFileReference2(prompt, searchRoots);
+      if (resolved.resolvedPath) {
+        promptPath = path3.resolve(resolved.resolvedPath);
+      } else {
+        logWarning2(
+          `Inline prompt used for evaluator '${name}' in '${evalId}' (file not found: ${resolved.displayPath})`,
+          resolved.attempted.length > 0 ? resolved.attempted.map((attempt) => `  Tried: ${attempt}`) : void 0
+        );
+      }
     }
+    const _model = asString2(rawEvaluator.model);
+    evaluators.push({
+      name,
+      type: "llm_judge",
+      prompt,
+      promptPath
+    });
   }
-  return codeBlocks;
+  return evaluators.length > 0 ? evaluators : void 0;
 }
+function coerceEvaluator(candidate, contextId) {
+  if (typeof candidate !== "string") {
+    return void 0;
+  }
+  if (isEvaluatorKind(candidate)) {
+    return candidate;
+  }
+  logWarning2(`Unknown evaluator '${candidate}' in ${contextId}, falling back to default`);
+  return void 0;
+}
+function asString2(value) {
+  return typeof value === "string" ? value : void 0;
+}
+function isJsonObject2(value) {
+  return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+function logWarning2(message, details) {
+  if (details && details.length > 0) {
+    const detailBlock = details.join("\n");
+    console.warn(`${ANSI_YELLOW2}Warning: ${message}
+${detailBlock}${ANSI_RESET2}`);
+  } else {
+    console.warn(`${ANSI_YELLOW2}Warning: ${message}${ANSI_RESET2}`);
+  }
+}
+
+// src/evaluation/loaders/message-processor.ts
+import { readFile as readFile2 } from "node:fs/promises";
+import path4 from "node:path";
+var ANSI_YELLOW3 = "\x1B[33m";
+var ANSI_RESET3 = "\x1B[0m";
 async function processMessages(options) {
   const {
     messages,
@@ -189,28 +448,28 @@ async function processMessages(options) {
       if (!isJsonObject(rawSegment)) {
         continue;
       }
-      const segmentType = asString(rawSegment.type);
+      const segmentType = asString3(rawSegment.type);
       if (segmentType === "file") {
-        const rawValue = asString(rawSegment.value);
+        const rawValue = asString3(rawSegment.value);
         if (!rawValue) {
           continue;
         }
-        const { displayPath, resolvedPath, attempted } = await resolveFileReference(
+        const { displayPath, resolvedPath, attempted } = await resolveFileReference2(
           rawValue,
           searchRoots
         );
         if (!resolvedPath) {
           const attempts = attempted.length ? ["  Tried:", ...attempted.map((candidate) => `    ${candidate}`)] : void 0;
           const context = messageType === "input" ? "" : " in expected_messages";
-          logWarning(`File not found${context}: ${displayPath}`, attempts);
+          logWarning3(`File not found${context}: ${displayPath}`, attempts);
           continue;
         }
         try {
-          const fileContent = (await readFile(resolvedPath, "utf8")).replace(/\r\n/g, "\n");
+          const fileContent = (await readFile2(resolvedPath, "utf8")).replace(/\r\n/g, "\n");
           if (messageType === "input" && guidelinePatterns && guidelinePaths) {
-            const relativeToRepo = path.relative(repoRootPath, resolvedPath);
+            const relativeToRepo = path4.relative(repoRootPath, resolvedPath);
             if (isGuidelineFile(relativeToRepo, guidelinePatterns)) {
-              guidelinePaths.push(path.resolve(resolvedPath));
+              guidelinePaths.push(path4.resolve(resolvedPath));
               if (verbose) {
                 console.log(`  [Guideline] Found: ${displayPath}`);
                 console.log(`    Resolved to: ${resolvedPath}`);
@@ -222,7 +481,7 @@ async function processMessages(options) {
             type: "file",
             path: displayPath,
             text: fileContent,
-            resolvedPath: path.resolve(resolvedPath)
+            resolvedPath: path4.resolve(resolvedPath)
           });
           if (verbose) {
             const label = messageType === "input" ? "[File]" : "[Expected Output File]";
@@ -231,7 +490,7 @@ async function processMessages(options) {
           }
         } catch (error) {
           const context = messageType === "input" ? "" : " expected output";
-          logWarning(`Could not read${context} file ${resolvedPath}: ${error.message}`);
+          logWarning3(`Could not read${context} file ${resolvedPath}: ${error.message}`);
         }
         continue;
       }
@@ -245,201 +504,117 @@ async function processMessages(options) {
   }
   return segments;
 }
-async function loadEvalCases(evalFilePath, repoRoot, options) {
-  const verbose = options?.verbose ?? false;
-  const evalIdFilter = options?.evalId;
-  const absoluteTestPath = path.resolve(evalFilePath);
-  if (!await fileExists2(absoluteTestPath)) {
-    throw new Error(`Test file not found: ${evalFilePath}`);
-  }
-  const repoRootPath = resolveToAbsolutePath(repoRoot);
-  const searchRoots = buildSearchRoots(absoluteTestPath, repoRootPath);
-  const config = await loadConfig(absoluteTestPath, repoRootPath);
-  const guidelinePatterns = config?.guideline_patterns;
-  const rawFile = await readFile(absoluteTestPath, "utf8");
-  const parsed = parse(rawFile);
-  if (!isJsonObject(parsed)) {
-    throw new Error(`Invalid test file format: ${evalFilePath}`);
-  }
-  const suite = parsed;
-  const datasetNameFromSuite = asString(suite.dataset)?.trim();
-  const fallbackDataset = path.basename(absoluteTestPath).replace(/\.ya?ml$/i, "") || "eval";
-  const datasetName = datasetNameFromSuite && datasetNameFromSuite.length > 0 ? datasetNameFromSuite : fallbackDataset;
-  const schema = suite.$schema;
-  if (schema !== SCHEMA_EVAL_V2) {
-    const message = typeof schema === "string" ? `Invalid $schema value '${schema}' in ${evalFilePath}. Expected '${SCHEMA_EVAL_V2}'` : `Missing required field '$schema' in ${evalFilePath}.
-Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
-    throw new Error(message);
+async function resolveAssistantContent(content, searchRoots, verbose) {
+  if (typeof content === "string") {
+    return content;
   }
-  const rawTestcases = suite.evalcases;
-  if (!Array.isArray(rawTestcases)) {
-    throw new Error(`Invalid test file format: ${evalFilePath} - missing 'evalcases' field`);
+  if (!content) {
+    return "";
   }
-  const globalEvaluator = coerceEvaluator(suite.evaluator, "global") ?? "llm_judge";
-  const globalExecution = isJsonObject(suite.execution) ? suite.execution : void 0;
-  const globalTarget = asString(globalExecution?.target) ?? asString(suite.target);
-  const results = [];
-  for (const rawEvalcase of rawTestcases) {
-    if (!isJsonObject(rawEvalcase)) {
-      logWarning("Skipping invalid eval case entry (expected object)");
+  const parts = [];
+  for (const entry of content) {
+    if (typeof entry === "string") {
+      parts.push({ content: entry, isFile: false });
       continue;
     }
-    const evalcase = rawEvalcase;
-    const id = asString(evalcase.id);
-    if (evalIdFilter && id !== evalIdFilter) {
+    if (!isJsonObject(entry)) {
       continue;
     }
-    const conversationId = asString(evalcase.conversation_id);
-    const outcome = asString(evalcase.outcome);
-    const inputMessagesValue = evalcase.input_messages;
-    const expectedMessagesValue = evalcase.expected_messages;
-    if (!id || !outcome || !Array.isArray(inputMessagesValue)) {
-      logWarning(`Skipping incomplete eval case: ${id ?? "unknown"}`);
+    const segmentType = asString3(entry.type);
+    if (segmentType === "file") {
+      const rawValue = asString3(entry.value);
+      if (!rawValue) {
+        continue;
+      }
+      const { displayPath, resolvedPath, attempted } = await resolveFileReference2(
+        rawValue,
+        searchRoots
+      );
+      if (!resolvedPath) {
+        const attempts = attempted.length ? ["  Tried:", ...attempted.map((candidate) => `    ${candidate}`)] : void 0;
+        logWarning3(`File not found in expected_messages: ${displayPath}`, attempts);
+        continue;
+      }
+      try {
+        const fileContent = (await readFile2(resolvedPath, "utf8")).replace(/\r\n/g, "\n").trim();
+        parts.push({ content: fileContent, isFile: true, displayPath });
+        if (verbose) {
+          console.log(`  [Expected Assistant File] Found: ${displayPath}`);
+          console.log(`    Resolved to: ${resolvedPath}`);
+        }
+      } catch (error) {
+        logWarning3(`Could not read file ${resolvedPath}: ${error.message}`);
+      }
       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) {
-      logWarning(`No valid expected message found for eval case: ${id}`);
+    const textValue = asString3(entry.text);
+    if (typeof textValue === "string") {
+      parts.push({ content: textValue, isFile: false });
       continue;
     }
-    if (expectedMessages.length > 1) {
-      logWarning(`Multiple expected messages found for eval case: ${id}, using first`);
-    }
-    const guidelinePaths = [];
-    const inputTextParts = [];
-    const inputSegments = await processMessages({
-      messages: inputMessages,
-      searchRoots,
-      repoRootPath,
-      guidelinePatterns,
-      guidelinePaths,
-      textParts: inputTextParts,
-      messageType: "input",
-      verbose
-    });
-    const outputSegments = hasExpectedMessages ? await processMessages({
-      messages: expectedMessages,
-      searchRoots,
-      repoRootPath,
-      guidelinePatterns,
-      messageType: "output",
-      verbose
-    }) : [];
-    const codeSnippets = extractCodeBlocks(inputSegments);
-    const expectedContent = expectedMessages[0]?.content;
-    const referenceAnswer = expectedContent ? await resolveAssistantContent(expectedContent, searchRoots, verbose) : "";
-    const question = inputTextParts.map((part) => part.trim()).filter((part) => part.length > 0).join(" ");
-    const evalCaseEvaluatorKind = coerceEvaluator(evalcase.evaluator, id) ?? globalEvaluator;
-    const evaluators = await parseEvaluators(evalcase, globalExecution, searchRoots, id ?? "unknown");
-    const userFilePaths = [];
-    for (const segment of inputSegments) {
-      if (segment.type === "file" && typeof segment.resolvedPath === "string") {
-        userFilePaths.push(segment.resolvedPath);
-      }
-    }
-    const allFilePaths = [
-      ...guidelinePaths.map((guidelinePath) => path.resolve(guidelinePath)),
-      ...userFilePaths
-    ];
-    const testCase = {
-      id,
-      dataset: datasetName,
-      conversation_id: conversationId,
-      question,
-      input_messages: inputMessages,
-      input_segments: inputSegments,
-      output_segments: outputSegments,
-      reference_answer: referenceAnswer,
-      guideline_paths: guidelinePaths.map((guidelinePath) => path.resolve(guidelinePath)),
-      guideline_patterns: guidelinePatterns,
-      file_paths: allFilePaths,
-      code_snippets: codeSnippets,
-      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");
-      }
+    const valueValue = asString3(entry.value);
+    if (typeof valueValue === "string") {
+      parts.push({ content: valueValue, isFile: false });
+      continue;
     }
-    results.push(testCase);
+    parts.push({ content: JSON.stringify(entry), isFile: false });
   }
-  return results;
+  return formatFileContents(parts);
 }
-function needsRoleMarkers(messages, processedSegmentsByMessage) {
-  if (messages.some((msg) => msg.role === "assistant" || msg.role === "tool")) {
-    return true;
-  }
-  let messagesWithContent = 0;
-  for (const segments of processedSegmentsByMessage) {
-    if (hasVisibleContent(segments)) {
-      messagesWithContent++;
-    }
-  }
-  return messagesWithContent > 1;
+function asString3(value) {
+  return typeof value === "string" ? value : void 0;
 }
-function hasVisibleContent(segments) {
-  return segments.some((segment) => {
-    const type = asString(segment.type);
-    if (type === "text") {
-      const value = asString(segment.value);
-      return value !== void 0 && value.trim().length > 0;
-    }
-    if (type === "guideline_ref") {
-      return false;
-    }
-    if (type === "file") {
-      const text = asString(segment.text);
-      return text !== void 0 && text.trim().length > 0;
-    }
-    return false;
-  });
+function cloneJsonObject(source) {
+  const entries = Object.entries(source).map(([key, value]) => [key, cloneJsonValue(value)]);
+  return Object.fromEntries(entries);
 }
-function formatSegment(segment) {
-  const type = asString(segment.type);
-  if (type === "text") {
-    return asString(segment.value);
+function cloneJsonValue(value) {
+  if (value === null) {
+    return null;
   }
-  if (type === "guideline_ref") {
-    const refPath = asString(segment.path);
-    return refPath ? `<Attached: ${refPath}>` : void 0;
+  if (typeof value === "string" || typeof value === "number" || typeof value === "boolean") {
+    return value;
   }
-  if (type === "file") {
-    const text = asString(segment.text);
-    const filePath = asString(segment.path);
-    if (text && filePath) {
-      return formatFileContents([{ content: text.trim(), isFile: true, displayPath: filePath }]);
-    }
+  if (Array.isArray(value)) {
+    return value.map((item) => cloneJsonValue(item));
+  }
+  if (typeof value === "object") {
+    return cloneJsonObject(value);
+  }
+  return value;
+}
+function logWarning3(message, details) {
+  if (details && details.length > 0) {
+    const detailBlock = details.join("\n");
+    console.warn(`${ANSI_YELLOW3}Warning: ${message}
+${detailBlock}${ANSI_RESET3}`);
+  } else {
+    console.warn(`${ANSI_YELLOW3}Warning: ${message}${ANSI_RESET3}`);
   }
-  return void 0;
 }
+
+// src/evaluation/formatting/prompt-builder.ts
+import { readFile as readFile3 } from "node:fs/promises";
+import path5 from "node:path";
+var ANSI_YELLOW4 = "\x1B[33m";
+var ANSI_RESET4 = "\x1B[0m";
 async function buildPromptInputs(testCase) {
   const guidelineParts = [];
   for (const rawPath of testCase.guideline_paths) {
-    const absolutePath = path.resolve(rawPath);
+    const absolutePath = path5.resolve(rawPath);
     if (!await fileExists2(absolutePath)) {
-      logWarning(`Could not read guideline file ${absolutePath}: file does not exist`);
+      logWarning4(`Could not read guideline file ${absolutePath}: file does not exist`);
       continue;
     }
     try {
-      const content = (await readFile(absolutePath, "utf8")).replace(/\r\n/g, "\n").trim();
+      const content = (await readFile3(absolutePath, "utf8")).replace(/\r\n/g, "\n").trim();
       guidelineParts.push({
         content,
         isFile: true,
-        displayPath: path.basename(absolutePath)
+        displayPath: path5.basename(absolutePath)
       });
     } catch (error) {
-      logWarning(`Could not read guideline file ${absolutePath}: ${error.message}`);
+      logWarning4(`Could not read guideline file ${absolutePath}: ${error.message}`);
     }
   }
   const guidelines = formatFileContents(guidelineParts);
@@ -463,9 +638,9 @@ async function buildPromptInputs(testCase) {
             messageSegments.push({ type: "text", value: segment });
           }
         } else if (isJsonObject(segment)) {
-          const type = asString(segment.type);
+          const type = asString4(segment.type);
           if (type === "file") {
-            const value = asString(segment.value);
+            const value = asString4(segment.value);
             if (!value) continue;
             if (testCase.guideline_patterns && isGuidelineFile(value, testCase.guideline_patterns)) {
               messageSegments.push({ type: "guideline_ref", path: value });
@@ -476,7 +651,7 @@ async function buildPromptInputs(testCase) {
               messageSegments.push({ type: "file", text: fileText, path: value });
             }
           } else if (type === "text") {
-            const textValue = asString(segment.value);
+            const textValue = asString4(segment.value);
             if (textValue && textValue.trim().length > 0) {
               messageSegments.push({ type: "text", value: textValue });
             }
@@ -532,6 +707,18 @@ ${messageContent}`);
   }) : void 0;
   return { question, guidelines, chatPrompt };
 }
+function needsRoleMarkers(messages, processedSegmentsByMessage) {
+  if (messages.some((msg) => msg.role === "assistant" || msg.role === "tool")) {
+    return true;
+  }
+  let messagesWithContent = 0;
+  for (const segments of processedSegmentsByMessage) {
+    if (hasVisibleContent(segments)) {
+      messagesWithContent++;
+    }
+  }
+  return messagesWithContent > 1;
+}
 function buildChatPromptFromSegments(options) {
   const { messages, segmentsByMessage, guidelinePatterns, guidelineContent, systemPrompt } = options;
   if (messages.length === 0) {
@@ -573,13 +760,12 @@ ${guidelineContent.trim()}`);
     const segments = segmentsByMessage[i];
     const contentParts = [];
     let role = message.role;
-    let name;
     if (role === "system") {
       role = "assistant";
       contentParts.push("@[System]:");
     } else if (role === "tool") {
-      role = "function";
-      name = "tool";
+      role = "assistant";
+      contentParts.push("@[Tool]:");
     }
     for (const segment of segments) {
       if (segment.type === "guideline_ref") {
@@ -597,282 +783,398 @@ ${guidelineContent.trim()}`);
     if (contentParts.length === 0) {
       continue;
     }
+    const content = contentParts.join("\n");
     chatPrompt.push({
       role,
-      content: contentParts.join("\n"),
-      ...name ? { name } : {}
+      content
     });
   }
   return chatPrompt.length > 0 ? chatPrompt : void 0;
 }
-async function fileExists2(absolutePath) {
-  try {
-    await access(absolutePath, constants.F_OK);
-    return true;
-  } catch {
-    return false;
-  }
-}
-function resolveToAbsolutePath(candidate) {
-  if (candidate instanceof URL) {
-    return fileURLToPath(candidate);
-  }
-  if (typeof candidate === "string") {
-    if (candidate.startsWith("file://")) {
-      return fileURLToPath(new URL(candidate));
-    }
-    return path.resolve(candidate);
-  }
-  throw new TypeError("Unsupported repoRoot value. Expected string or URL.");
-}
-function asString(value) {
+function asString4(value) {
   return typeof value === "string" ? value : void 0;
 }
-function cloneJsonObject(source) {
-  const entries = Object.entries(source).map(([key, value]) => [key, cloneJsonValue(value)]);
-  return Object.fromEntries(entries);
+function logWarning4(message) {
+  console.warn(`${ANSI_YELLOW4}Warning: ${message}${ANSI_RESET4}`);
 }
-function cloneJsonValue(value) {
-  if (value === null) {
-    return null;
-  }
-  if (typeof value === "string" || typeof value === "number" || typeof value === "boolean") {
-    return value;
-  }
-  if (Array.isArray(value)) {
-    return value.map((item) => cloneJsonValue(item));
+
+// src/evaluation/yaml-parser.ts
+var ANSI_YELLOW5 = "\x1B[33m";
+var ANSI_RESET5 = "\x1B[0m";
+var SCHEMA_EVAL_V2 = "agentv-eval-v2";
+async function readTestSuiteMetadata(testFilePath) {
+  try {
+    const absolutePath = path6.resolve(testFilePath);
+    const content = await readFile4(absolutePath, "utf8");
+    const parsed = parse2(content);
+    if (!isJsonObject(parsed)) {
+      return {};
+    }
+    return { target: extractTargetFromSuite(parsed) };
+  } catch {
+    return {};
   }
-  return cloneJsonObject(value);
 }
-function formatFileContents(parts) {
-  const fileCount = parts.filter((p) => p.isFile).length;
-  if (fileCount > 0) {
-    return parts.map((part) => {
-      if (part.isFile && part.displayPath) {
-        return `<file path="${part.displayPath}">
-${part.content}
-</file>`;
-      }
-      return part.content;
-    }).join("\n\n");
+async function loadEvalCases(evalFilePath, repoRoot, options) {
+  const verbose = options?.verbose ?? false;
+  const evalIdFilter = options?.evalId;
+  const absoluteTestPath = path6.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 readFile4(absoluteTestPath, "utf8");
+  const parsed = parse2(rawFile);
+  if (!isJsonObject(parsed)) {
+    throw new Error(`Invalid test file format: ${evalFilePath}`);
   }
-  return parts.map((p) => p.content).join(" ");
-}
-async function resolveAssistantContent(content, searchRoots, verbose) {
-  if (typeof content === "string") {
-    return content;
+  const suite = parsed;
+  const datasetNameFromSuite = asString5(suite.dataset)?.trim();
+  const fallbackDataset = path6.basename(absoluteTestPath).replace(/\.ya?ml$/i, "") || "eval";
+  const datasetName = datasetNameFromSuite && datasetNameFromSuite.length > 0 ? datasetNameFromSuite : fallbackDataset;
+  const schema = suite.$schema;
+  if (schema !== SCHEMA_EVAL_V2) {
+    const message = typeof schema === "string" ? `Invalid $schema value '${schema}' in ${evalFilePath}. Expected '${SCHEMA_EVAL_V2}'` : `Missing required field '$schema' in ${evalFilePath}.
+Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
+    throw new Error(message);
   }
-  if (!content) {
-    return "";
+  const rawTestcases = suite.evalcases;
+  if (!Array.isArray(rawTestcases)) {
+    throw new Error(`Invalid test file format: ${evalFilePath} - missing 'evalcases' field`);
   }
-  const parts = [];
-  for (const entry of content) {
-    if (typeof entry === "string") {
-      parts.push({ content: entry, isFile: false });
-      continue;
-    }
-    if (!isJsonObject(entry)) {
-      continue;
-    }
-    const segmentType = asString(entry.type);
-    if (segmentType === "file") {
-      const rawValue = asString(entry.value);
-      if (!rawValue) {
-        continue;
-      }
-      const { displayPath, resolvedPath, attempted } = await resolveFileReference(
-        rawValue,
-        searchRoots
-      );
-      if (!resolvedPath) {
-        const attempts = attempted.length ? ["  Tried:", ...attempted.map((candidate) => `    ${candidate}`)] : void 0;
-        logWarning(`File not found in expected_messages: ${displayPath}`, attempts);
-        continue;
-      }
-      try {
-        const fileContent = (await readFile(resolvedPath, "utf8")).replace(/\r\n/g, "\n").trim();
-        parts.push({ content: fileContent, isFile: true, displayPath });
-        if (verbose) {
-          console.log(`  [Expected Assistant File] Found: ${displayPath}`);
-          console.log(`    Resolved to: ${resolvedPath}`);
-        }
-      } catch (error) {
-        logWarning(`Could not read file ${resolvedPath}: ${error.message}`);
-      }
-      continue;
-    }
-    const textValue = asString(entry.text);
-    if (typeof textValue === "string") {
-      parts.push({ content: textValue, isFile: false });
+  const globalEvaluator = coerceEvaluator(suite.evaluator, "global") ?? "llm_judge";
+  const globalExecution = isJsonObject(suite.execution) ? suite.execution : void 0;
+  const _globalTarget = asString5(globalExecution?.target) ?? asString5(suite.target);
+  const results = [];
+  for (const rawEvalcase of rawTestcases) {
+    if (!isJsonObject(rawEvalcase)) {
+      logWarning5("Skipping invalid eval case entry (expected object)");
       continue;
     }
-    const valueValue = asString(entry.value);
-    if (typeof valueValue === "string") {
-      parts.push({ content: valueValue, isFile: false });
+    const evalcase = rawEvalcase;
+    const id = asString5(evalcase.id);
+    if (evalIdFilter && id !== evalIdFilter) {
       continue;
     }
-    parts.push({ content: JSON.stringify(entry), isFile: false });
-  }
-  return formatFileContents(parts);
-}
-async function parseEvaluators(rawEvalCase, globalExecution, searchRoots, evalId) {
-  const execution = rawEvalCase.execution;
-  const candidateEvaluators = isJsonObject(execution) ? execution.evaluators ?? rawEvalCase.evaluators : rawEvalCase.evaluators ?? globalExecution?.evaluators;
-  if (candidateEvaluators === void 0) {
-    return void 0;
-  }
-  if (!Array.isArray(candidateEvaluators)) {
-    logWarning(`Skipping evaluators for '${evalId}': expected array`);
-    return void 0;
-  }
-  const evaluators = [];
-  for (const rawEvaluator of candidateEvaluators) {
-    if (!isJsonObject(rawEvaluator)) {
-      logWarning(`Skipping invalid evaluator entry for '${evalId}' (expected object)`);
+    const conversationId = asString5(evalcase.conversation_id);
+    const outcome = asString5(evalcase.outcome);
+    const inputMessagesValue = evalcase.input_messages;
+    const expectedMessagesValue = evalcase.expected_messages;
+    if (!id || !outcome || !Array.isArray(inputMessagesValue)) {
+      logWarning5(`Skipping incomplete eval case: ${id ?? "unknown"}`);
       continue;
     }
-    const name = asString(rawEvaluator.name);
-    const typeValue = rawEvaluator.type;
-    if (!name || !isEvaluatorKind(typeValue)) {
-      logWarning(`Skipping evaluator with invalid name/type in '${evalId}'`);
+    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) {
+      logWarning5(`No valid expected message found for eval case: ${id}`);
       continue;
     }
-    if (typeValue === "code") {
-      const script = asString(rawEvaluator.script);
-      if (!script) {
-        logWarning(`Skipping code evaluator '${name}' in '${evalId}': missing script`);
-        continue;
-      }
-      const cwd = asString(rawEvaluator.cwd);
-      let resolvedCwd;
-      if (cwd) {
-        const resolved = await resolveFileReference(cwd, searchRoots);
-        if (resolved.resolvedPath) {
-          resolvedCwd = path.resolve(resolved.resolvedPath);
-        } else {
-          logWarning(
-            `Code evaluator '${name}' in '${evalId}': cwd not found (${resolved.displayPath})`,
-            resolved.attempted.length > 0 ? resolved.attempted.map((attempt) => `  Tried: ${attempt}`) : void 0
-          );
-        }
-      } else {
-        resolvedCwd = searchRoots[0];
-      }
-      evaluators.push({
-        name,
-        type: "code",
-        script,
-        cwd,
-        resolvedCwd
-      });
-      continue;
+    if (expectedMessages.length > 1) {
+      logWarning5(`Multiple expected messages found for eval case: ${id}, using first`);
     }
-    const prompt = asString(rawEvaluator.prompt);
-    let promptPath;
-    if (prompt) {
-      const resolved = await resolveFileReference(prompt, searchRoots);
-      if (resolved.resolvedPath) {
-        promptPath = path.resolve(resolved.resolvedPath);
-      } else {
-        logWarning(
-          `Inline prompt used for evaluator '${name}' in '${evalId}' (file not found: ${resolved.displayPath})`,
-          resolved.attempted.length > 0 ? resolved.attempted.map((attempt) => `  Tried: ${attempt}`) : void 0
-        );
+    const guidelinePaths = [];
+    const inputTextParts = [];
+    const inputSegments = await processMessages({
+      messages: inputMessages,
+      searchRoots,
+      repoRootPath,
+      guidelinePatterns,
+      guidelinePaths,
+      textParts: inputTextParts,
+      messageType: "input",
+      verbose
+    });
+    const outputSegments = hasExpectedMessages ? await processMessages({
+      messages: expectedMessages,
+      searchRoots,
+      repoRootPath,
+      guidelinePatterns,
+      messageType: "output",
+      verbose
+    }) : [];
+    const codeSnippets = extractCodeBlocks(inputSegments);
+    const expectedContent = expectedMessages[0]?.content;
+    const referenceAnswer = expectedContent ? await resolveAssistantContent(expectedContent, searchRoots, verbose) : "";
+    const question = inputTextParts.map((part) => part.trim()).filter((part) => part.length > 0).join(" ");
+    const evalCaseEvaluatorKind = coerceEvaluator(evalcase.evaluator, id) ?? globalEvaluator;
+    const evaluators = await parseEvaluators(evalcase, globalExecution, searchRoots, id ?? "unknown");
+    const userFilePaths = [];
+    for (const segment of inputSegments) {
+      if (segment.type === "file" && typeof segment.resolvedPath === "string") {
+        userFilePaths.push(segment.resolvedPath);
       }
     }
-    const model = asString(rawEvaluator.model);
-    evaluators.push({
-      name,
-      type: "llm_judge",
-      prompt,
-      promptPath
-    });
+    const allFilePaths = [
+      ...guidelinePaths.map((guidelinePath) => path6.resolve(guidelinePath)),
+      ...userFilePaths
+    ];
+    const testCase = {
+      id,
+      dataset: datasetName,
+      conversation_id: conversationId,
+      question,
+      input_messages: inputMessages,
+      input_segments: inputSegments,
+      output_segments: outputSegments,
+      reference_answer: referenceAnswer,
+      guideline_paths: guidelinePaths.map((guidelinePath) => path6.resolve(guidelinePath)),
+      guideline_patterns: guidelinePatterns,
+      file_paths: allFilePaths,
+      code_snippets: codeSnippets,
+      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(`    - ${guidelinePath}`);
+        }
+      } else {
+        console.log("  No guidelines found");
+      }
+    }
+    results.push(testCase);
   }
-  return evaluators.length > 0 ? evaluators : void 0;
+  return results;
 }
-function coerceEvaluator(candidate, contextId) {
-  if (typeof candidate !== "string") {
-    return void 0;
-  }
-  if (isEvaluatorKind(candidate)) {
-    return candidate;
-  }
-  logWarning(`Unknown evaluator '${candidate}' in ${contextId}, falling back to default`);
-  return void 0;
+function asString5(value) {
+  return typeof value === "string" ? value : void 0;
 }
-function logWarning(message, details) {
+function logWarning5(message, details) {
   if (details && details.length > 0) {
     const detailBlock = details.join("\n");
-    console.warn(`${ANSI_YELLOW}Warning: ${message}
-${detailBlock}${ANSI_RESET}`);
+    console.warn(`${ANSI_YELLOW5}Warning: ${message}
+${detailBlock}${ANSI_RESET5}`);
   } else {
-    console.warn(`${ANSI_YELLOW}Warning: ${message}${ANSI_RESET}`);
+    console.warn(`${ANSI_YELLOW5}Warning: ${message}${ANSI_RESET5}`);
   }
 }
 
-// src/evaluation/providers/ax.ts
-import { AxAI } from "@ax-llm/ax";
+// src/evaluation/providers/ai-sdk.ts
+import { createAnthropic } from "@ai-sdk/anthropic";
+import { createAzure } from "@ai-sdk/azure";
+import { createGoogleGenerativeAI } from "@ai-sdk/google";
+import { generateText } from "ai";
 var DEFAULT_SYSTEM_PROMPT = "You are a careful assistant. Follow all provided instructions and do not fabricate results.";
+var AzureProvider = class {
+  constructor(targetName, config) {
+    this.config = config;
+    this.id = `azure:${targetName}`;
+    this.targetName = targetName;
+    this.defaults = {
+      temperature: config.temperature,
+      maxOutputTokens: config.maxOutputTokens
+    };
+    this.retryConfig = config.retry;
+    const azure = createAzure(buildAzureOptions(config));
+    this.model = azure(config.deploymentName);
+  }
+  id;
+  kind = "azure";
+  targetName;
+  model;
+  defaults;
+  retryConfig;
+  async invoke(request) {
+    return invokeModel({
+      model: this.model,
+      request,
+      defaults: this.defaults,
+      retryConfig: this.retryConfig
+    });
+  }
+};
+var AnthropicProvider = class {
+  constructor(targetName, config) {
+    this.config = config;
+    this.id = `anthropic:${targetName}`;
+    this.targetName = targetName;
+    this.defaults = {
+      temperature: config.temperature,
+      maxOutputTokens: config.maxOutputTokens,
+      thinkingBudget: config.thinkingBudget
+    };
+    this.retryConfig = config.retry;
+    const anthropic = createAnthropic({
+      apiKey: config.apiKey
+    });
+    this.model = anthropic(config.model);
+  }
+  id;
+  kind = "anthropic";
+  targetName;
+  model;
+  defaults;
+  retryConfig;
+  async invoke(request) {
+    const providerOptions = buildAnthropicProviderOptions(this.defaults);
+    return invokeModel({
+      model: this.model,
+      request,
+      defaults: this.defaults,
+      retryConfig: this.retryConfig,
+      providerOptions
+    });
+  }
+};
+var GeminiProvider = class {
+  constructor(targetName, config) {
+    this.config = config;
+    this.id = `gemini:${targetName}`;
+    this.targetName = targetName;
+    this.defaults = {
+      temperature: config.temperature,
+      maxOutputTokens: config.maxOutputTokens
+    };
+    this.retryConfig = config.retry;
+    const google = createGoogleGenerativeAI({
+      apiKey: config.apiKey
+    });
+    this.model = google(config.model);
+  }
+  id;
+  kind = "gemini";
+  targetName;
+  model;
+  defaults;
+  retryConfig;
+  async invoke(request) {
+    return invokeModel({
+      model: this.model,
+      request,
+      defaults: this.defaults,
+      retryConfig: this.retryConfig
+    });
+  }
+};
+function buildAzureOptions(config) {
+  const options = {
+    apiKey: config.apiKey,
+    apiVersion: config.version,
+    useDeploymentBasedUrls: true
+  };
+  const baseURL = normalizeAzureBaseUrl(config.resourceName);
+  if (baseURL) {
+    options.baseURL = baseURL;
+  } else {
+    options.resourceName = config.resourceName;
+  }
+  return options;
+}
+function normalizeAzureBaseUrl(resourceName) {
+  const trimmed = resourceName.trim();
+  if (!/^https?:\/\//i.test(trimmed)) {
+    return void 0;
+  }
+  const withoutSlash = trimmed.replace(/\/+$/, "");
+  const normalized = withoutSlash.endsWith("/openai") ? withoutSlash : `${withoutSlash}/openai`;
+  return normalized;
+}
+function buildAnthropicProviderOptions(defaults) {
+  if (defaults.thinkingBudget === void 0) {
+    return void 0;
+  }
+  return {
+    anthropic: {
+      thinking: {
+        type: "enabled",
+        budgetTokens: defaults.thinkingBudget
+      }
+    }
+  };
+}
 function buildChatPrompt(request) {
-  if (request.chatPrompt) {
-    const hasSystemMessage = request.chatPrompt.some((message) => message.role === "system");
+  const provided = request.chatPrompt?.length ? request.chatPrompt : void 0;
+  if (provided) {
+    const hasSystemMessage = provided.some((message) => message.role === "system");
     if (hasSystemMessage) {
-      return request.chatPrompt;
+      return provided;
     }
-    const systemContent2 = resolveSystemContent(request);
-    return [{ role: "system", content: systemContent2 }, ...request.chatPrompt];
+    const systemContent2 = resolveSystemContent(request, false);
+    return [{ role: "system", content: systemContent2 }, ...provided];
   }
-  const systemContent = resolveSystemContent(request);
+  const systemContent = resolveSystemContent(request, true);
   const userContent = request.question.trim();
   const prompt = [
-    {
-      role: "system",
-      content: systemContent
-    },
-    {
-      role: "user",
-      content: userContent
-    }
+    { role: "system", content: systemContent },
+    { role: "user", content: userContent }
   ];
   return prompt;
 }
-function resolveSystemContent(request) {
+function resolveSystemContent(request, includeGuidelines) {
   const systemSegments = [];
-  const metadataSystemPrompt = typeof request.metadata?.systemPrompt === "string" ? request.metadata.systemPrompt : void 0;
-  if (metadataSystemPrompt && metadataSystemPrompt.trim().length > 0) {
-    systemSegments.push(metadataSystemPrompt.trim());
+  if (request.systemPrompt && request.systemPrompt.trim().length > 0) {
+    systemSegments.push(request.systemPrompt.trim());
   } else {
     systemSegments.push(DEFAULT_SYSTEM_PROMPT);
   }
-  if (request.guidelines && request.guidelines.trim().length > 0) {
+  if (includeGuidelines && request.guidelines && request.guidelines.trim().length > 0) {
     systemSegments.push(`[[ ## Guidelines ## ]]
 
 ${request.guidelines.trim()}`);
   }
   return systemSegments.join("\n\n");
 }
-function extractModelConfig(request, defaults) {
+function toModelMessages(chatPrompt) {
+  return chatPrompt.map((message) => {
+    if (message.role === "tool" || message.role === "function") {
+      const prefix = message.name ? `@[${message.name}]: ` : "@[Tool]: ";
+      return {
+        role: "assistant",
+        content: `${prefix}${message.content}`
+      };
+    }
+    if (message.role === "assistant" || message.role === "system" || message.role === "user") {
+      return {
+        role: message.role,
+        content: message.content
+      };
+    }
+    return {
+      role: "user",
+      content: message.content
+    };
+  });
+}
+function resolveModelSettings(request, defaults) {
   const temperature = request.temperature ?? defaults.temperature;
-  const maxTokens = request.maxOutputTokens ?? defaults.maxOutputTokens;
-  const config = {};
-  if (temperature !== void 0) {
-    config.temperature = temperature;
-  }
-  if (maxTokens !== void 0) {
-    config.maxTokens = maxTokens;
-  }
-  return Object.keys(config).length > 0 ? config : void 0;
+  const maxOutputTokens = request.maxOutputTokens ?? defaults.maxOutputTokens;
+  return {
+    temperature,
+    maxOutputTokens
+  };
+}
+async function invokeModel(options) {
+  const { model, request, defaults, retryConfig, providerOptions } = options;
+  const chatPrompt = buildChatPrompt(request);
+  const { temperature, maxOutputTokens } = resolveModelSettings(request, defaults);
+  const result = await withRetry(
+    () => generateText({
+      model,
+      messages: toModelMessages(chatPrompt),
+      temperature,
+      maxOutputTokens,
+      maxRetries: 0,
+      abortSignal: request.signal,
+      ...providerOptions ? { providerOptions } : {}
+    }),
+    retryConfig,
+    request.signal
+  );
+  return mapResponse(result);
 }
-function mapResponse(response) {
-  const primary = response.results[0];
-  const text = typeof primary?.content === "string" ? primary.content : "";
-  const reasoning = primary?.thought ?? primary?.thoughtBlock?.data;
-  const usage = toJsonObject(response.modelUsage);
+function mapResponse(result) {
   return {
-    text,
-    reasoning,
-    raw: response,
-    usage
+    text: result.text ?? "",
+    reasoning: result.reasoningText ?? void 0,
+    raw: result,
+    usage: toJsonObject(result.totalUsage ?? result.usage)
   };
 }
 function toJsonObject(value) {
@@ -885,34 +1187,59 @@ function toJsonObject(value) {
     return void 0;
   }
 }
-function ensureChatResponse(result) {
-  if (typeof ReadableStream !== "undefined" && result instanceof ReadableStream) {
-    throw new Error("Streaming responses are not supported for this provider");
+function extractStatus(error) {
+  if (!error || typeof error !== "object") {
+    return void 0;
   }
-  if (!result || typeof result !== "object" || !("results" in result)) {
-    throw new Error("Unexpected response type from AxAI provider");
+  const candidate = error;
+  const directStatus = candidate.status ?? candidate.statusCode;
+  if (typeof directStatus === "number" && Number.isFinite(directStatus)) {
+    return directStatus;
   }
-  return result;
+  const responseStatus = typeof candidate.response === "object" && candidate.response ? candidate.response.status : void 0;
+  if (typeof responseStatus === "number" && Number.isFinite(responseStatus)) {
+    return responseStatus;
+  }
+  const message = typeof candidate.message === "string" ? candidate.message : void 0;
+  if (message) {
+    const match = message.match(/HTTP\s+(\d{3})/i);
+    if (match) {
+      const parsed = Number.parseInt(match[1], 10);
+      if (Number.isFinite(parsed)) {
+        return parsed;
+      }
+    }
+  }
+  return void 0;
 }
-function isRetryableError(error, retryableStatusCodes) {
+function isNetworkError(error) {
   if (!error || typeof error !== "object") {
     return false;
   }
-  if ("status" in error && typeof error.status === "number") {
-    return retryableStatusCodes.includes(error.status);
+  const candidate = error;
+  if (candidate.name === "AbortError") {
+    return false;
   }
-  if ("message" in error && typeof error.message === "string") {
-    const match = error.message.match(/HTTP (\d{3})/);
-    if (match) {
-      const status = Number.parseInt(match[1], 10);
-      return retryableStatusCodes.includes(status);
-    }
+  const code = candidate.code;
+  if (typeof code === "string" && /^E(AI|CONN|HOST|NET|PIPE|TIME|REFUSED|RESET)/i.test(code)) {
+    return true;
   }
-  if ("name" in error && error.name === "AxAIServiceNetworkError") {
+  const message = typeof candidate.message === "string" ? candidate.message : void 0;
+  if (message && /(network|fetch failed|ECONNRESET|ENOTFOUND|EAI_AGAIN|ETIMEDOUT|ECONNREFUSED)/i.test(message)) {
     return true;
   }
   return false;
 }
+function isRetryableError(error, retryableStatusCodes) {
+  const status = extractStatus(error);
+  if (status === 401 || status === 403) {
+    return false;
+  }
+  if (typeof status === "number") {
+    return retryableStatusCodes.includes(status);
+  }
+  return isNetworkError(error);
+}
 function calculateRetryDelay(attempt, config) {
   const delay = Math.min(
     config.maxDelayMs,
@@ -948,152 +1275,16 @@ async function withRetry(fn, retryConfig, signal) {
       }
       const delay = calculateRetryDelay(attempt, config);
       await sleep(delay);
-      if (signal?.aborted) {
-        throw new Error(`Request aborted: ${signal.reason ?? "Unknown reason"}`);
-      }
     }
   }
   throw lastError;
 }
-var AzureProvider = class {
-  constructor(targetName, config) {
-    this.config = config;
-    this.id = `azure:${targetName}`;
-    this.targetName = targetName;
-    this.defaults = {
-      temperature: config.temperature,
-      maxOutputTokens: config.maxOutputTokens
-    };
-    this.retryConfig = config.retry;
-    this.ai = AxAI.create({
-      name: "azure-openai",
-      apiKey: config.apiKey,
-      resourceName: config.resourceName,
-      deploymentName: config.deploymentName,
-      version: config.version,
-      config: {
-        stream: false
-      }
-    });
-  }
-  id;
-  kind = "azure";
-  targetName;
-  ai;
-  defaults;
-  retryConfig;
-  async invoke(request) {
-    const chatPrompt = buildChatPrompt(request);
-    const modelConfig = extractModelConfig(request, this.defaults);
-    const response = await withRetry(
-      async () => await this.ai.chat(
-        {
-          chatPrompt,
-          model: this.config.deploymentName,
-          ...modelConfig ? { modelConfig } : {}
-        },
-        request.signal ? { abortSignal: request.signal } : void 0
-      ),
-      this.retryConfig,
-      request.signal
-    );
-    return mapResponse(ensureChatResponse(response));
-  }
-  getAxAI() {
-    return this.ai;
-  }
-};
-var AnthropicProvider = class {
-  constructor(targetName, config) {
-    this.config = config;
-    this.id = `anthropic:${targetName}`;
-    this.targetName = targetName;
-    this.defaults = {
-      temperature: config.temperature,
-      maxOutputTokens: config.maxOutputTokens,
-      thinkingBudget: config.thinkingBudget
-    };
-    this.retryConfig = config.retry;
-    this.ai = AxAI.create({
-      name: "anthropic",
-      apiKey: config.apiKey
-    });
-  }
-  id;
-  kind = "anthropic";
-  targetName;
-  ai;
-  defaults;
-  retryConfig;
-  async invoke(request) {
-    const chatPrompt = buildChatPrompt(request);
-    const modelConfig = extractModelConfig(request, this.defaults);
-    const response = await withRetry(
-      async () => await this.ai.chat(
-        {
-          chatPrompt,
-          model: this.config.model,
-          ...modelConfig ? { modelConfig } : {}
-        },
-        request.signal ? { abortSignal: request.signal } : void 0
-      ),
-      this.retryConfig,
-      request.signal
-    );
-    return mapResponse(ensureChatResponse(response));
-  }
-  getAxAI() {
-    return this.ai;
-  }
-};
-var GeminiProvider = class {
-  constructor(targetName, config) {
-    this.config = config;
-    this.id = `gemini:${targetName}`;
-    this.targetName = targetName;
-    this.defaults = {
-      temperature: config.temperature,
-      maxOutputTokens: config.maxOutputTokens
-    };
-    this.retryConfig = config.retry;
-    this.ai = AxAI.create({
-      name: "google-gemini",
-      apiKey: config.apiKey
-    });
-  }
-  id;
-  kind = "gemini";
-  targetName;
-  ai;
-  defaults;
-  retryConfig;
-  async invoke(request) {
-    const chatPrompt = buildChatPrompt(request);
-    const modelConfig = extractModelConfig(request, this.defaults);
-    const response = await withRetry(
-      async () => await this.ai.chat(
-        {
-          chatPrompt,
-          model: this.config.model,
-          ...modelConfig ? { modelConfig } : {}
-        },
-        request.signal ? { abortSignal: request.signal } : void 0
-      ),
-      this.retryConfig,
-      request.signal
-    );
-    return mapResponse(ensureChatResponse(response));
-  }
-  getAxAI() {
-    return this.ai;
-  }
-};
 
 // src/evaluation/providers/cli.ts
 import { exec as execWithCallback } from "node:child_process";
 import fs from "node:fs/promises";
 import os from "node:os";
-import path2 from "node:path";
+import path7 from "node:path";
 import { promisify } from "node:util";
 var execAsync = promisify(execWithCallback);
 var DEFAULT_MAX_BUFFER = 10 * 1024 * 1024;
@@ -1135,12 +1326,14 @@ var CliProvider = class {
   supportsBatch = false;
   config;
   runCommand;
+  verbose;
   healthcheckPromise;
   constructor(targetName, config, runner = defaultCommandRunner) {
     this.targetName = targetName;
     this.id = `cli:${targetName}`;
     this.config = config;
     this.runCommand = runner;
+    this.verbose = config.verbose ?? false;
   }
   async invoke(request) {
     if (request.signal?.aborted) {
@@ -1241,6 +1434,11 @@ var CliProvider = class {
         generateOutputFilePath("healthcheck")
       )
     );
+    if (this.verbose) {
+      console.log(
+        `[cli-provider:${this.targetName}] (healthcheck) CLI_EVALS_DIR=${process.env.CLI_EVALS_DIR ?? ""} cwd=${healthcheck.cwd ?? this.config.cwd ?? ""} command=${renderedCommand}`
+      );
+    }
     const result = await this.runCommand(renderedCommand, {
       cwd: healthcheck.cwd ?? this.config.cwd,
       env: process.env,
@@ -1272,7 +1470,7 @@ function normalizeInputFiles(inputFiles) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = path2.resolve(inputFile);
+    const absolutePath = path7.resolve(inputFile);
     if (!unique.has(absolutePath)) {
       unique.set(absolutePath, absolutePath);
     }
@@ -1286,7 +1484,7 @@ function formatFileList(files, template) {
   const formatter = template ?? "{path}";
   return files.map((filePath) => {
     const escapedPath = shellEscape(filePath);
-    const escapedName = shellEscape(path2.basename(filePath));
+    const escapedName = shellEscape(path7.basename(filePath));
     return formatter.replaceAll("{path}", escapedPath).replaceAll("{basename}", escapedName);
   }).join(" ");
 }
@@ -1310,7 +1508,7 @@ function generateOutputFilePath(evalCaseId) {
   const safeEvalId = evalCaseId || "unknown";
   const timestamp = Date.now();
   const random = Math.random().toString(36).substring(2, 9);
-  return path2.join(os.tmpdir(), `agentv-${safeEvalId}-${timestamp}-${random}.json`);
+  return path7.join(os.tmpdir(), `agentv-${safeEvalId}-${timestamp}-${random}.json`);
 }
 function formatTimeoutSuffix(timeoutMs) {
   if (!timeoutMs || timeoutMs <= 0) {
@@ -1326,7 +1524,7 @@ import { randomUUID } from "node:crypto";
 import { constants as constants2, createWriteStream } from "node:fs";
 import { access as access2, mkdtemp, mkdir, rm, writeFile } from "node:fs/promises";
 import { tmpdir } from "node:os";
-import path4 from "node:path";
+import path9 from "node:path";
 import { promisify as promisify2 } from "node:util";
 
 // src/evaluation/providers/codex-log-tracker.ts
@@ -1383,7 +1581,7 @@ function subscribeToCodexLogEntries(listener) {
 }
 
 // src/evaluation/providers/preread.ts
-import path3 from "node:path";
+import path8 from "node:path";
 function buildPromptDocument(request, inputFiles, options) {
   const parts = [];
   const guidelineFiles = collectGuidelineFiles(
@@ -1408,7 +1606,7 @@ function normalizeInputFiles2(inputFiles) {
   }
   const deduped = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = path3.resolve(inputFile);
+    const absolutePath = path8.resolve(inputFile);
     if (!deduped.has(absolutePath)) {
       deduped.set(absolutePath, absolutePath);
     }
@@ -1421,14 +1619,14 @@ function collectGuidelineFiles(inputFiles, guidelinePatterns, overrides) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = path3.resolve(inputFile);
+    const absolutePath = path8.resolve(inputFile);
     if (overrides?.has(absolutePath)) {
       if (!unique.has(absolutePath)) {
         unique.set(absolutePath, absolutePath);
       }
       continue;
     }
-    const normalized = absolutePath.split(path3.sep).join("/");
+    const normalized = absolutePath.split(path8.sep).join("/");
     if (isGuidelineFile(normalized, guidelinePatterns)) {
       if (!unique.has(absolutePath)) {
         unique.set(absolutePath, absolutePath);
@@ -1443,7 +1641,7 @@ function collectInputFiles(inputFiles) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const inputFile of inputFiles) {
-    const absolutePath = path3.resolve(inputFile);
+    const absolutePath = path8.resolve(inputFile);
     if (!unique.has(absolutePath)) {
       unique.set(absolutePath, absolutePath);
     }
@@ -1455,7 +1653,7 @@ function buildMandatoryPrereadBlock(guidelineFiles, inputFiles) {
     return "";
   }
   const buildList = (files) => files.map((absolutePath) => {
-    const fileName = path3.basename(absolutePath);
+    const fileName = path8.basename(absolutePath);
     const fileUri = pathToFileUri(absolutePath);
     return `* [${fileName}](${fileUri})`;
   });
@@ -1475,7 +1673,7 @@ ${buildList(inputFiles).join("\n")}.`);
   return sections.join("\n");
 }
 function pathToFileUri(filePath) {
-  const absolutePath = path3.isAbsolute(filePath) ? filePath : path3.resolve(filePath);
+  const absolutePath = path8.isAbsolute(filePath) ? filePath : path8.resolve(filePath);
   const normalizedPath = absolutePath.replace(/\\/g, "/");
   if (/^[a-zA-Z]:\//.test(normalizedPath)) {
     return `file:///${normalizedPath}`;
@@ -1513,7 +1711,7 @@ var CodexProvider = class {
     const logger = await this.createStreamLogger(request).catch(() => void 0);
     try {
       const promptContent = buildPromptDocument(request, inputFiles);
-      const promptFile = path4.join(workspaceRoot, PROMPT_FILENAME);
+      const promptFile = path9.join(workspaceRoot, PROMPT_FILENAME);
       await writeFile(promptFile, promptContent, "utf8");
       const args = this.buildCodexArgs();
       const cwd = this.resolveCwd(workspaceRoot);
@@ -1563,7 +1761,7 @@ var CodexProvider = class {
     if (!this.config.cwd) {
       return workspaceRoot;
     }
-    return path4.resolve(this.config.cwd);
+    return path9.resolve(this.config.cwd);
   }
   buildCodexArgs() {
     const args = ["--ask-for-approval", "never", "exec", "--json", "--color", "never", "--skip-git-repo-check"];
@@ -1597,7 +1795,7 @@ var CodexProvider = class {
     }
   }
   async createWorkspace() {
-    return await mkdtemp(path4.join(tmpdir(), WORKSPACE_PREFIX));
+    return await mkdtemp(path9.join(tmpdir(), WORKSPACE_PREFIX));
   }
   async cleanupWorkspace(workspaceRoot) {
     try {
@@ -1611,9 +1809,9 @@ var CodexProvider = class {
       return void 0;
     }
     if (this.config.logDir) {
-      return path4.resolve(this.config.logDir);
+      return path9.resolve(this.config.logDir);
     }
-    return path4.join(process.cwd(), ".agentv", "logs", "codex");
+    return path9.join(process.cwd(), ".agentv", "logs", "codex");
   }
   async createStreamLogger(request) {
     const logDir = this.resolveLogDirectory();
@@ -1627,7 +1825,7 @@ var CodexProvider = class {
       console.warn(`Skipping Codex stream logging (could not create ${logDir}): ${message}`);
       return void 0;
     }
-    const filePath = path4.join(logDir, buildLogFilename(request, this.targetName));
+    const filePath = path9.join(logDir, buildLogFilename(request, this.targetName));
     try {
       const logger = await CodexStreamLogger.create({
         filePath,
@@ -1842,7 +2040,7 @@ function tryParseJsonValue(rawLine) {
 async function locateExecutable(candidate) {
   const includesPathSeparator = candidate.includes("/") || candidate.includes("\\");
   if (includesPathSeparator) {
-    const resolved = path4.isAbsolute(candidate) ? candidate : path4.resolve(candidate);
+    const resolved = path9.isAbsolute(candidate) ? candidate : path9.resolve(candidate);
     const executablePath = await ensureWindowsExecutableVariant(resolved);
     await access2(executablePath, constants2.F_OK);
     return executablePath;
@@ -2189,7 +2387,7 @@ var MockProvider = class {
 };
 
 // src/evaluation/providers/vscode.ts
-import path5 from "node:path";
+import path10 from "node:path";
 import { dispatchAgentSession, dispatchBatchAgent, getSubagentRoot, provisionSubagents } from "subagent";
 var VSCodeProvider = class {
   id;
@@ -2302,6 +2500,9 @@ var VSCodeProvider = class {
 };
 function buildPromptDocument2(request, attachments, guidelinePatterns) {
   const parts = [];
+  if (request.systemPrompt && request.systemPrompt.trim().length > 0) {
+    parts.push(request.systemPrompt.trim());
+  }
   const guidelineFiles = collectGuidelineFiles2(attachments, guidelinePatterns);
   const attachmentFiles = collectAttachmentFiles(attachments);
   const nonGuidelineAttachments = attachmentFiles.filter(
@@ -2319,7 +2520,7 @@ function buildMandatoryPrereadBlock2(guidelineFiles, attachmentFiles) {
     return "";
   }
   const buildList = (files) => files.map((absolutePath) => {
-    const fileName = path5.basename(absolutePath);
+    const fileName = path10.basename(absolutePath);
     const fileUri = pathToFileUri2(absolutePath);
     return `* [${fileName}](${fileUri})`;
   });
@@ -2344,8 +2545,8 @@ function collectGuidelineFiles2(attachments, guidelinePatterns) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const attachment of attachments) {
-    const absolutePath = path5.resolve(attachment);
-    const normalized = absolutePath.split(path5.sep).join("/");
+    const absolutePath = path10.resolve(attachment);
+    const normalized = absolutePath.split(path10.sep).join("/");
     if (isGuidelineFile(normalized, guidelinePatterns)) {
       if (!unique.has(absolutePath)) {
         unique.set(absolutePath, absolutePath);
@@ -2360,7 +2561,7 @@ function collectAttachmentFiles(attachments) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const attachment of attachments) {
-    const absolutePath = path5.resolve(attachment);
+    const absolutePath = path10.resolve(attachment);
     if (!unique.has(absolutePath)) {
       unique.set(absolutePath, absolutePath);
     }
@@ -2368,7 +2569,7 @@ function collectAttachmentFiles(attachments) {
   return Array.from(unique.values());
 }
 function pathToFileUri2(filePath) {
-  const absolutePath = path5.isAbsolute(filePath) ? filePath : path5.resolve(filePath);
+  const absolutePath = path10.isAbsolute(filePath) ? filePath : path10.resolve(filePath);
   const normalizedPath = absolutePath.replace(/\\/g, "/");
   if (/^[a-zA-Z]:\//.test(normalizedPath)) {
     return `file:///${normalizedPath}`;
@@ -2381,7 +2582,7 @@ function normalizeAttachments(attachments) {
   }
   const deduped = /* @__PURE__ */ new Set();
   for (const attachment of attachments) {
-    deduped.add(path5.resolve(attachment));
+    deduped.add(path10.resolve(attachment));
   }
   return Array.from(deduped);
 }
@@ -2390,7 +2591,7 @@ function mergeAttachments(all) {
   for (const list of all) {
     if (!list) continue;
     for (const inputFile of list) {
-      deduped.add(path5.resolve(inputFile));
+      deduped.add(path10.resolve(inputFile));
     }
   }
   return deduped.size > 0 ? Array.from(deduped) : void 0;
@@ -2436,9 +2637,9 @@ 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 readFile2 } from "node:fs/promises";
-import path6 from "node:path";
-import { parse as parse2 } from "yaml";
+import { access as access3, readFile as readFile5 } from "node:fs/promises";
+import path11 from "node:path";
+import { parse as parse3 } from "yaml";
 function isRecord(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value);
 }
@@ -2493,12 +2694,12 @@ async function fileExists3(filePath) {
   }
 }
 async function readTargetDefinitions(filePath) {
-  const absolutePath = path6.resolve(filePath);
+  const absolutePath = path11.resolve(filePath);
   if (!await fileExists3(absolutePath)) {
     throw new Error(`targets.yaml not found at ${absolutePath}`);
   }
-  const raw = await readFile2(absolutePath, "utf8");
-  const parsed = parse2(raw);
+  const raw = await readFile5(absolutePath, "utf8");
+  const parsed = parse3(raw);
   if (!isRecord(parsed)) {
     throw new Error(`targets.yaml at ${absolutePath} must be a YAML object with '$schema' and 'targets' fields`);
   }
@@ -2541,18 +2742,34 @@ function resolveAndCreateProvider(definition, env = process.env) {
 }
 
 // src/evaluation/evaluators.ts
-import { randomUUID as randomUUID2 } from "node:crypto";
+var DEFAULT_EVALUATOR_TEMPLATE = `You are an expert evaluator. Your goal is to grade the candidate_answer based on how well it achieves the expected_outcome for the original task.
+
+Use the reference_answer as a gold standard for a high-quality response (if provided). The candidate_answer does not need to match it verbatim, but should capture the key points and follow the same spirit.
+
+Be concise and focused in your evaluation. Provide succinct, specific feedback rather than verbose explanations.
+
+[[ ## expected_outcome ## ]]
+{{expected_outcome}}
+
+[[ ## question ## ]]
+{{question}}
+
+[[ ## reference_answer ## ]]
+{{reference_answer}}
+
+[[ ## candidate_answer ## ]]
+{{candidate_answer}}`;
 var LlmJudgeEvaluator = class {
   kind = "llm_judge";
   resolveJudgeProvider;
   maxOutputTokens;
   temperature;
-  customPrompt;
+  evaluatorTemplate;
   constructor(options) {
     this.resolveJudgeProvider = options.resolveJudgeProvider;
     this.maxOutputTokens = options.maxOutputTokens;
     this.temperature = options.temperature;
-    this.customPrompt = options.customPrompt;
+    this.evaluatorTemplate = options.evaluatorTemplate;
   }
   async evaluate(context) {
     const judgeProvider = await this.resolveJudgeProvider(context);
@@ -2562,26 +2779,21 @@ var LlmJudgeEvaluator = class {
     return this.evaluateWithPrompt(context, judgeProvider);
   }
   async evaluateWithPrompt(context, judgeProvider) {
-    const hasReferenceAnswer = hasNonEmptyReferenceAnswer(context.evalCase);
     const formattedQuestion = context.promptInputs.question && context.promptInputs.question.trim().length > 0 ? context.promptInputs.question : context.evalCase.question;
-    let prompt = buildQualityPrompt(context.evalCase, context.candidate, formattedQuestion);
-    let systemPrompt = context.systemPrompt ?? this.customPrompt ?? buildSystemPrompt(hasReferenceAnswer);
-    if (systemPrompt && hasTemplateVariables(systemPrompt)) {
-      const variables = {
-        input_messages: JSON.stringify(context.evalCase.input_segments, null, 2),
-        output_messages: JSON.stringify(context.evalCase.output_segments, null, 2),
-        candidate_answer: context.candidate,
-        reference_answer: context.evalCase.reference_answer ?? "",
-        expected_outcome: context.evalCase.expected_outcome,
-        question: formattedQuestion
-      };
-      prompt = substituteVariables(systemPrompt, variables);
-      systemPrompt = buildSystemPrompt(hasReferenceAnswer);
-    }
-    const metadata = systemPrompt !== void 0 ? { systemPrompt } : {};
+    const variables = {
+      input_messages: JSON.stringify(context.evalCase.input_segments, null, 2),
+      output_messages: JSON.stringify(context.evalCase.output_segments, null, 2),
+      candidate_answer: context.candidate.trim(),
+      reference_answer: (context.evalCase.reference_answer ?? "").trim(),
+      expected_outcome: context.evalCase.expected_outcome.trim(),
+      question: formattedQuestion.trim()
+    };
+    const systemPrompt = buildOutputSchema();
+    const evaluatorTemplate = context.evaluatorTemplateOverride ?? this.evaluatorTemplate ?? DEFAULT_EVALUATOR_TEMPLATE;
+    const userPrompt = substituteVariables(evaluatorTemplate, variables);
     const response = await judgeProvider.invoke({
-      question: prompt,
-      metadata,
+      question: userPrompt,
+      systemPrompt,
       evalCaseId: context.evalCase.id,
       attempt: context.attempt,
       maxOutputTokens: this.maxOutputTokens,
@@ -2594,11 +2806,9 @@ var LlmJudgeEvaluator = class {
     const reasoning = parsed.reasoning ?? response.reasoning;
     const expectedAspectCount = Math.max(hits.length + misses.length, 1);
     const evaluatorRawRequest = {
-      id: randomUUID2(),
-      provider: judgeProvider.id,
-      prompt,
-      target: context.target.name,
-      ...systemPrompt !== void 0 && { systemPrompt }
+      userPrompt,
+      systemPrompt,
+      target: judgeProvider.targetName
     };
     return {
       score,
@@ -2610,20 +2820,8 @@ var LlmJudgeEvaluator = class {
     };
   }
 };
-function buildSystemPrompt(hasReferenceAnswer) {
-  const basePrompt = [
-    "You are an expert evaluator. Your goal is to grade the candidate_answer based on how well it achieves the expected_outcome for the original task.",
-    ""
-  ];
-  if (hasReferenceAnswer) {
-    basePrompt.push(
-      "Use the reference_answer as a gold standard for a high-quality response. The candidate_answer does not need to match it verbatim, but should capture the key points and follow the same spirit.",
-      ""
-    );
-  }
-  basePrompt.push(
-    "Be concise and focused in your evaluation. Provide succinct, specific feedback rather than verbose explanations.",
-    "",
+function buildOutputSchema() {
+  return [
     "You must respond with a single JSON object matching this schema:",
     "",
     "{",
@@ -2632,30 +2830,7 @@ function buildSystemPrompt(hasReferenceAnswer) {
     '  "misses": [<array of strings, max 4 items, brief specific failures or omissions, empty if none>],',
     '  "reasoning": "<string, concise explanation for the score, 1-2 sentences max>"',
     "}"
-  );
-  return basePrompt.join("\n");
-}
-function buildQualityPrompt(evalCase, candidate, question) {
-  const parts = [
-    "[[ ## expected_outcome ## ]]",
-    evalCase.expected_outcome.trim(),
-    "",
-    "[[ ## question ## ]]",
-    question.trim(),
-    ""
-  ];
-  if (hasNonEmptyReferenceAnswer(evalCase)) {
-    parts.push(
-      "[[ ## reference_answer ## ]]",
-      evalCase.reference_answer.trim(),
-      ""
-    );
-  }
-  parts.push(
-    "[[ ## candidate_answer ## ]]",
-    candidate.trim()
-  );
-  return parts.join("\n");
+  ].join("\n");
 }
 function clampScore(value) {
   if (Number.isNaN(value) || !Number.isFinite(value)) {
@@ -2737,9 +2912,6 @@ function extractJsonBlob(text) {
 function isNonEmptyString(value) {
   return typeof value === "string" && value.trim().length > 0;
 }
-function hasNonEmptyReferenceAnswer(evalCase) {
-  return evalCase.reference_answer !== void 0 && evalCase.reference_answer.trim().length > 0;
-}
 var CodeEvaluator = class {
   kind = "code";
   script;
@@ -2845,19 +3017,16 @@ function parseJsonSafe(payload) {
     return void 0;
   }
 }
-function hasTemplateVariables(text) {
-  return /\$\{[a-zA-Z0-9_]+\}/.test(text);
-}
 function substituteVariables(template, variables) {
-  return template.replace(/\$\{([a-zA-Z0-9_]+)\}/g, (match, varName) => {
+  return template.replace(/\{\{([a-zA-Z0-9_]+)\}\}/g, (match, varName) => {
     return variables[varName] ?? match;
   });
 }
 
 // src/evaluation/orchestrator.ts
-import { createHash, randomUUID as randomUUID3 } from "node:crypto";
+import { createHash, randomUUID as randomUUID2 } from "node:crypto";
 import { mkdir as mkdir2, writeFile as writeFile2 } from "node:fs/promises";
-import path7 from "node:path";
+import path12 from "node:path";
 
 // ../../node_modules/.pnpm/yocto-queue@1.2.1/node_modules/yocto-queue/index.js
 var Node = class {
@@ -3420,6 +3589,7 @@ async function evaluateCandidate(options) {
     }
   }
   return {
+    timestamp: completedAt.toISOString(),
     eval_id: evalCase.id,
     dataset: evalCase.dataset,
     conversation_id: evalCase.conversation_id,
@@ -3427,14 +3597,12 @@ async function evaluateCandidate(options) {
     hits: score.hits,
     misses: score.misses,
     candidate_answer: candidate,
-    expected_aspect_count: score.expectedAspectCount,
     target: target.name,
-    timestamp: completedAt.toISOString(),
     reasoning: score.reasoning,
     raw_aspects: score.rawAspects,
     agent_provider_request: agentProviderRequest,
     lm_provider_request: lmProviderRequest,
-    evaluator_raw_request: evaluatorResults ? void 0 : score.evaluatorRawRequest,
+    evaluator_provider_request: evaluatorResults ? void 0 : score.evaluatorRawRequest,
     evaluator_results: evaluatorResults
   };
 }
@@ -3511,7 +3679,7 @@ async function runEvaluatorList(options) {
           hits: score2.hits,
           misses: score2.misses,
           reasoning: score2.reasoning,
-          evaluator_raw_request: score2.evaluatorRawRequest
+          evaluator_provider_request: score2.evaluatorRawRequest
         });
         continue;
       }
@@ -3538,7 +3706,7 @@ async function runEvaluatorList(options) {
           hits: score2.hits,
           misses: score2.misses,
           reasoning: score2.reasoning,
-          evaluator_raw_request: score2.evaluatorRawRequest
+          evaluator_provider_request: score2.evaluatorRawRequest
         });
         continue;
       }
@@ -3591,7 +3759,7 @@ async function runLlmJudgeEvaluator(options) {
     promptInputs,
     now,
     judgeProvider,
-    systemPrompt: customPrompt,
+    evaluatorTemplateOverride: customPrompt,
     evaluator: config
   });
 }
@@ -3632,8 +3800,8 @@ function buildEvaluatorRegistry(overrides, resolveJudgeProvider) {
 async function dumpPrompt(directory, evalCase, promptInputs) {
   const timestamp = (/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-");
   const filename = `${timestamp}_${sanitizeFilename(evalCase.id)}.json`;
-  const filePath = path7.resolve(directory, filename);
-  await mkdir2(path7.dirname(filePath), { recursive: true });
+  const filePath = path12.resolve(directory, filename);
+  await mkdir2(path12.dirname(filePath), { recursive: true });
   const payload = {
     eval_id: evalCase.id,
     question: promptInputs.question,
@@ -3647,7 +3815,7 @@ function sanitizeFilename(value) {
     return "prompt";
   }
   const sanitized = value.replace(/[^A-Za-z0-9._-]+/g, "_");
-  return sanitized.length > 0 ? sanitized : randomUUID3();
+  return sanitized.length > 0 ? sanitized : randomUUID2();
 }
 async function invokeProvider(provider, options) {
   const { evalCase, promptInputs, attempt, agentTimeoutMs, signal } = options;
@@ -3703,6 +3871,7 @@ function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs,
     }
   }
   return {
+    timestamp: timestamp.toISOString(),
     eval_id: evalCase.id,
     dataset: evalCase.dataset,
     conversation_id: evalCase.conversation_id,
@@ -3710,9 +3879,7 @@ function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs,
     hits: [],
     misses: [`Error: ${message}`],
     candidate_answer: `Error occurred: ${message}`,
-    expected_aspect_count: 0,
     target: targetName,
-    timestamp: timestamp.toISOString(),
     raw_aspects: [],
     agent_provider_request: agentProviderRequest,
     lm_provider_request: lmProviderRequest,