agentv 0.2.3 → 0.2.8

package/dist/{chunk-S3RN2GSO.js → chunk-RLBRJX7V.js}

@@ -585,7 +585,7 @@ var require_utc = __commonJS({
 import { Command } from "commander";
 import { readFileSync as readFileSync2 } from "node:fs";
 
-// ../../packages/core/dist/chunk-5REK5RSI.js
+// ../../packages/core/dist/chunk-XXNQA4EW.js
 import { constants } from "node:fs";
 import { access } from "node:fs/promises";
 import path from "node:path";
@@ -613,6 +613,30 @@ async function findGitRoot(startPath) {
   }
   return null;
 }
+function buildDirectoryChain(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;
+  }
+  if (!seen.has(boundary)) {
+    directories.push(boundary);
+  }
+  return directories;
+}
 function buildSearchRoots(evalPath, repoRoot) {
   const uniqueRoots = [];
   const addRoot = (root2) => {
@@ -664,8 +688,32 @@ async function resolveFileReference(rawValue, searchRoots) {
   }
   return { displayPath, attempted };
 }
+var KNOWN_PROVIDERS = [
+  "azure",
+  "anthropic",
+  "gemini",
+  "mock",
+  "vscode",
+  "vscode-insiders"
+];
+var PROVIDER_ALIASES = [
+  "azure-openai",
+  // alias for "azure"
+  "google",
+  // alias for "gemini"
+  "google-gemini",
+  // alias for "gemini"
+  "openai",
+  // legacy/future support
+  "bedrock",
+  // legacy/future support
+  "vertex"
+  // legacy/future support
+];
+var TARGETS_SCHEMA_V2 = "agentv-targets-v2";
 
 // ../../packages/core/dist/index.js
+import micromatch from "micromatch";
 import { constants as constants3 } from "node:fs";
 import { access as access3, readFile as readFile2 } from "node:fs/promises";
 import path7 from "node:path";
@@ -9025,17 +9073,16 @@ var coerce = {
 var NEVER = INVALID;
 
 // ../../packages/core/dist/index.js
-import { mkdtemp, readFile as readFile22, rm as rm2, writeFile as writeFile3 } from "node:fs/promises";
-import { tmpdir } from "node:os";
+import { readFile as readFile22 } from "node:fs/promises";
 import path22 from "node:path";
 
-// ../../node_modules/.pnpm/subagent@0.4.1/node_modules/subagent/dist/vscode/agentDispatch.js
+// ../../node_modules/.pnpm/subagent@0.4.2/node_modules/subagent/dist/vscode/agentDispatch.js
 import { exec, spawn } from "child_process";
 import { copyFile, mkdir as mkdir2, readdir as readdir2, readFile, stat as stat2, writeFile } from "fs/promises";
 import path5 from "path";
 import { promisify } from "util";
 
-// ../../node_modules/.pnpm/subagent@0.4.1/node_modules/subagent/dist/vscode/constants.js
+// ../../node_modules/.pnpm/subagent@0.4.2/node_modules/subagent/dist/vscode/constants.js
 import os from "os";
 import path2 from "path";
 var DEFAULT_LOCK_NAME = "subagent.lock";
@@ -9047,7 +9094,7 @@ var DEFAULT_SUBAGENT_ROOT = getDefaultSubagentRoot();
 var DEFAULT_WAKEUP_FILENAME = "wakeup.chatmode.md";
 var DEFAULT_ALIVE_FILENAME = ".alive";
 
-// ../../node_modules/.pnpm/subagent@0.4.1/node_modules/subagent/dist/utils/fs.js
+// ../../node_modules/.pnpm/subagent@0.4.2/node_modules/subagent/dist/utils/fs.js
 import { constants as constants2 } from "fs";
 import { access as access2, mkdir, readdir, rm, stat } from "fs/promises";
 import path3 from "path";
@@ -9080,14 +9127,14 @@ async function removeIfExists(target) {
   }
 }
 
-// ../../node_modules/.pnpm/subagent@0.4.1/node_modules/subagent/dist/utils/time.js
+// ../../node_modules/.pnpm/subagent@0.4.2/node_modules/subagent/dist/utils/time.js
 function sleep(ms2) {
   return new Promise((resolve) => {
     setTimeout(resolve, ms2);
   });
 }
 
-// ../../node_modules/.pnpm/subagent@0.4.1/node_modules/subagent/dist/utils/workspace.js
+// ../../node_modules/.pnpm/subagent@0.4.2/node_modules/subagent/dist/utils/workspace.js
 import path4 from "path";
 
 // ../../node_modules/.pnpm/json5@2.2.3/node_modules/json5/dist/index.mjs
@@ -10175,7 +10222,7 @@ var JSON5 = {
 var lib = JSON5;
 var dist_default = lib;
 
-// ../../node_modules/.pnpm/subagent@0.4.1/node_modules/subagent/dist/utils/workspace.js
+// ../../node_modules/.pnpm/subagent@0.4.2/node_modules/subagent/dist/utils/workspace.js
 function transformWorkspacePaths(workspaceContent, templateDir) {
   let workspace;
   try {
@@ -10248,7 +10295,7 @@ function transformWorkspacePaths(workspaceContent, templateDir) {
   return JSON.stringify(transformedWorkspace, null, 2);
 }
 
-// ../../node_modules/.pnpm/subagent@0.4.1/node_modules/subagent/dist/vscode/agentDispatch.js
+// ../../node_modules/.pnpm/subagent@0.4.2/node_modules/subagent/dist/vscode/agentDispatch.js
 var execAsync = promisify(exec);
 var DEFAULT_WORKSPACE_TEMPLATE = {
   folders: [
@@ -10430,9 +10477,9 @@ async function prepareSubagentDirectory(subagentDir, promptFile, chatId, workspa
   return 0;
 }
 function createRequestPrompt(userQuery, responseFileTmp, responseFileFinal, subagentName, vscodeCmd) {
-  const escapedUserQuery = userQuery.replace(/`/g, "\\`");
   return `[[ ## task ## ]]
-${escapedUserQuery}
+
+${userQuery}
 
 [[ ## system_instructions ## ]]
 
@@ -10589,7 +10636,7 @@ async function dispatchAgentSession(options) {
   }
 }
 
-// ../../node_modules/.pnpm/subagent@0.4.1/node_modules/subagent/dist/vscode/provision.js
+// ../../node_modules/.pnpm/subagent@0.4.2/node_modules/subagent/dist/vscode/provision.js
 import { writeFile as writeFile2 } from "fs/promises";
 import path6 from "path";
 var DEFAULT_WORKSPACE_TEMPLATE2 = {
@@ -10762,9 +10809,52 @@ var CODE_BLOCK_PATTERN = /```[\s\S]*?```/g;
 var ANSI_YELLOW = "\x1B[33m";
 var ANSI_RESET = "\x1B[0m";
 var SCHEMA_EVAL_V2 = "agentv-eval-v2";
-function isGuidelineFile(filePath) {
+var SCHEMA_CONFIG_V2 = "agentv-config-v2";
+async function loadConfig(evalFilePath, repoRoot) {
+  const directories = buildDirectoryChain(evalFilePath, repoRoot);
+  for (const directory of directories) {
+    const configPath = path7.join(directory, ".agentv", "config.yaml");
+    if (!await fileExists2(configPath)) {
+      continue;
+    }
+    try {
+      const rawConfig = await readFile2(configPath, "utf8");
+      const parsed = parse3(rawConfig);
+      if (!isJsonObject(parsed)) {
+        logWarning(`Invalid .agentv/config.yaml format at ${configPath}`);
+        continue;
+      }
+      const config = parsed;
+      const schema = config.$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`);
+        continue;
+      }
+      if (Array.isArray(guidelinePatterns) && !guidelinePatterns.every((p) => typeof p === "string")) {
+        logWarning(`Invalid guideline_patterns in ${configPath}, all entries must be strings`);
+        continue;
+      }
+      return {
+        guideline_patterns: guidelinePatterns
+      };
+    } catch (error) {
+      logWarning(`Could not read .agentv/config.yaml at ${configPath}: ${error.message}`);
+      continue;
+    }
+  }
+  return null;
+}
+function isGuidelineFile(filePath, patterns) {
   const normalized = filePath.split("\\").join("/");
-  return normalized.endsWith(".instructions.md") || normalized.includes("/instructions/") || normalized.endsWith(".prompt.md") || normalized.includes("/prompts/");
+  const patternsToUse = patterns ?? [];
+  return micromatch.isMatch(normalized, patternsToUse);
 }
 function extractCodeBlocks(segments) {
   const codeBlocks = [];
@@ -10784,43 +10874,45 @@ function extractCodeBlocks(segments) {
   }
   return codeBlocks;
 }
-async function loadTestCases(testFilePath, repoRoot, options) {
+async function loadEvalCases(evalFilePath, repoRoot, options) {
   const verbose = options?.verbose ?? false;
-  const absoluteTestPath = path7.resolve(testFilePath);
+  const absoluteTestPath = path7.resolve(evalFilePath);
   if (!await fileExists2(absoluteTestPath)) {
-    throw new Error(`Test file not found: ${testFilePath}`);
+    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 readFile2(absoluteTestPath, "utf8");
   const parsed = parse3(rawFile);
   if (!isJsonObject(parsed)) {
-    throw new Error(`Invalid test file format: ${testFilePath}`);
+    throw new Error(`Invalid test file format: ${evalFilePath}`);
   }
   const suite = parsed;
   const schema = suite.$schema;
   if (schema !== SCHEMA_EVAL_V2) {
-    const message = typeof schema === "string" ? `Invalid $schema value '${schema}' in ${testFilePath}. Expected '${SCHEMA_EVAL_V2}'` : `Missing required field '$schema' in ${testFilePath}.
+    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);
   }
   const rawTestcases = suite.evalcases;
   if (!Array.isArray(rawTestcases)) {
-    throw new Error(`Invalid test file format: ${testFilePath} - missing 'evalcases' field`);
+    throw new Error(`Invalid test file format: ${evalFilePath} - missing 'evalcases' field`);
   }
   const globalGrader = coerceGrader(suite.grader) ?? "llm_judge";
   const results = [];
-  for (const rawTestcase of rawTestcases) {
-    if (!isJsonObject(rawTestcase)) {
+  for (const rawEvalcase of rawTestcases) {
+    if (!isJsonObject(rawEvalcase)) {
       logWarning("Skipping invalid test case entry (expected object)");
       continue;
     }
-    const testcase = rawTestcase;
-    const id = asString(testcase.id);
-    const conversationId = asString(testcase.conversation_id);
-    const outcome = asString(testcase.outcome);
-    const inputMessagesValue = testcase.input_messages;
-    const expectedMessagesValue = testcase.expected_messages;
+    const evalcase = rawEvalcase;
+    const id = asString(evalcase.id);
+    const conversationId = asString(evalcase.conversation_id);
+    const outcome = asString(evalcase.outcome);
+    const inputMessagesValue = evalcase.input_messages;
+    const expectedMessagesValue = evalcase.expected_messages;
     if (!id || !outcome || !Array.isArray(inputMessagesValue)) {
       logWarning(`Skipping incomplete test case: ${id ?? "unknown"}`);
       continue;
@@ -10833,6 +10925,7 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
     const expectedMessages = expectedMessagesValue.filter((msg) => isTestMessage(msg));
     const assistantMessages = expectedMessages.filter((message) => message.role === "assistant");
     const userMessages = inputMessages.filter((message) => message.role === "user");
+    const systemMessages = inputMessages.filter((message) => message.role === "system");
     if (assistantMessages.length === 0) {
       logWarning(`No assistant message found for test case: ${id}`);
       continue;
@@ -10840,6 +10933,29 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
     if (assistantMessages.length > 1) {
       logWarning(`Multiple assistant messages found for test case: ${id}, using first`);
     }
+    if (systemMessages.length > 1) {
+      logWarning(`Multiple system messages found for test case: ${id}, using first`);
+    }
+    let systemMessageContent;
+    if (systemMessages.length > 0) {
+      const content = systemMessages[0]?.content;
+      if (typeof content === "string") {
+        systemMessageContent = content;
+      } else if (Array.isArray(content)) {
+        const textParts = [];
+        for (const segment of content) {
+          if (isJsonObject(segment)) {
+            const value = segment.value;
+            if (typeof value === "string") {
+              textParts.push(value);
+            }
+          }
+        }
+        if (textParts.length > 0) {
+          systemMessageContent = textParts.join("\n\n");
+        }
+      }
+    }
     const userSegments = [];
     const guidelinePaths = [];
     const userTextParts = [];
@@ -10871,7 +10987,8 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
           }
           try {
             const fileContent = (await readFile2(resolvedPath, "utf8")).replace(/\r\n/g, "\n");
-            if (isGuidelineFile(displayPath)) {
+            const relativeToRepo = path7.relative(repoRootPath, resolvedPath);
+            if (isGuidelineFile(relativeToRepo, guidelinePatterns)) {
               guidelinePaths.push(path7.resolve(resolvedPath));
               if (verbose) {
                 console.log(`  [Guideline] Found: ${displayPath}`);
@@ -10881,7 +10998,8 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
               userSegments.push({
                 type: "file",
                 path: displayPath,
-                text: fileContent
+                text: fileContent,
+                resolvedPath: path7.resolve(resolvedPath)
               });
               if (verbose) {
                 console.log(`  [File] Found: ${displayPath}`);
@@ -10903,16 +11021,29 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
     }
     const codeSnippets = extractCodeBlocks(userSegments);
     const assistantContent = assistantMessages[0]?.content;
-    const expectedAssistantRaw = normalizeAssistantContent(assistantContent);
+    const expectedAssistantRaw = await resolveAssistantContent(assistantContent, searchRoots, verbose);
     const userTextPrompt = userTextParts.map((part) => part.trim()).filter((part) => part.length > 0).join(" ");
-    const testCaseGrader = coerceGrader(testcase.grader) ?? globalGrader;
+    const testCaseGrader = coerceGrader(evalcase.grader) ?? globalGrader;
+    const userFilePaths = [];
+    for (const segment of userSegments) {
+      if (segment.type === "file" && typeof segment.resolvedPath === "string") {
+        userFilePaths.push(segment.resolvedPath);
+      }
+    }
+    const allFilePaths = [
+      ...guidelinePaths.map((guidelinePath) => path7.resolve(guidelinePath)),
+      ...userFilePaths
+    ];
     const testCase = {
       id,
       conversation_id: conversationId,
       task: userTextPrompt,
       user_segments: userSegments,
+      system_message: systemMessageContent,
       expected_assistant_raw: expectedAssistantRaw,
       guideline_paths: guidelinePaths.map((guidelinePath) => path7.resolve(guidelinePath)),
+      guideline_patterns: guidelinePatterns,
+      file_paths: allFilePaths,
       code_snippets: codeSnippets,
       outcome,
       grader: testCaseGrader
@@ -10978,7 +11109,7 @@ ${body}`);
   }
   const request = requestParts.map((part) => part.trim()).filter((part) => part.length > 0).join("\n\n");
   const guidelines = guidelineContents.map((part) => part.trim()).filter((part) => part.length > 0).join("\n\n");
-  return { request, guidelines };
+  return { request, guidelines, systemMessage: testCase.system_message };
 }
 async function fileExists2(absolutePath) {
   try {
@@ -11019,7 +11150,7 @@ function cloneJsonValue(value) {
   }
   return cloneJsonObject(value);
 }
-function normalizeAssistantContent(content) {
+async function resolveAssistantContent(content, searchRoots, verbose) {
   if (typeof content === "string") {
     return content;
   }
@@ -11032,12 +11163,42 @@ function normalizeAssistantContent(content) {
       parts.push(entry);
       continue;
     }
-    const textValue = asString(entry["text"]);
+    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 readFile2(resolvedPath, "utf8")).replace(/\r\n/g, "\n");
+        parts.push(fileContent);
+        if (verbose) {
+          console.log(`  [Expected Assistant File] Found: ${displayPath}`);
+          console.log(`    Resolved to: ${resolvedPath}`);
+        }
+      } catch (error) {
+        logWarning(`Could not read file ${resolvedPath}: ${error.message}`);
+      }
+      continue;
+    }
+    const textValue = asString(entry.text);
     if (typeof textValue === "string") {
       parts.push(textValue);
       continue;
     }
-    const valueValue = asString(entry["value"]);
+    const valueValue = asString(entry.value);
     if (typeof valueValue === "string") {
       parts.push(valueValue);
       continue;
@@ -11071,15 +11232,18 @@ function buildChatPrompt(request) {
     return request.chatPrompt;
   }
   const systemSegments = [];
-  if (request.guidelines && request.guidelines.trim().length > 0) {
-    systemSegments.push(`Guidelines:
-${request.guidelines.trim()}`);
-  }
   const metadataSystemPrompt = typeof request.metadata?.systemPrompt === "string" ? request.metadata.systemPrompt : void 0;
   if (metadataSystemPrompt && metadataSystemPrompt.trim().length > 0) {
     systemSegments.push(metadataSystemPrompt.trim());
+  } else {
+    systemSegments.push(DEFAULT_SYSTEM_PROMPT);
   }
-  const systemContent = systemSegments.length > 0 ? systemSegments.join("\n\n") : DEFAULT_SYSTEM_PROMPT;
+  if (request.guidelines && request.guidelines.trim().length > 0) {
+    systemSegments.push(`[[ ## Guidelines ## ]]
+
+${request.guidelines.trim()}`);
+  }
+  const systemContent = systemSegments.join("\n\n");
   const userContent = request.prompt.trim();
   const prompt = [
     {
@@ -11528,7 +11692,6 @@ function resolveOptionalBoolean(source2) {
 function isLikelyEnvReference(value) {
   return /^[A-Z0-9_]+$/.test(value);
 }
-var PROMPT_FILE_PREFIX = "bbeval-vscode-";
 var VSCodeProvider = class {
   id;
   kind;
@@ -11545,128 +11708,89 @@ var VSCodeProvider = class {
       throw new Error("VS Code provider request was aborted before dispatch");
     }
     const attachments = normalizeAttachments(request.attachments);
-    const promptContent = buildPromptDocument(request, attachments);
-    const directory = await mkdtemp(path22.join(tmpdir(), PROMPT_FILE_PREFIX));
-    const promptPath = path22.join(directory, `${request.testCaseId ?? "request"}.prompt.md`);
-    try {
-      await writeFile3(promptPath, promptContent, "utf8");
-      const session = await dispatchAgentSession({
-        userQuery: composeUserQuery(request),
-        promptFile: promptPath,
-        extraAttachments: attachments,
-        wait: this.config.waitForResponse,
-        dryRun: this.config.dryRun,
-        vscodeCmd: this.config.command,
-        subagentRoot: this.config.subagentRoot,
-        workspaceTemplate: this.config.workspaceTemplate,
-        silent: true
-      });
-      if (session.exitCode !== 0 || !session.responseFile) {
-        const failure = session.error ?? "VS Code subagent did not produce a response";
-        throw new Error(failure);
-      }
-      if (this.config.dryRun) {
-        return {
-          text: "",
-          raw: {
-            session,
-            promptFile: promptPath,
-            attachments
-          }
-        };
-      }
-      const responseText = await readFile22(session.responseFile, "utf8");
+    const promptContent = buildPromptDocument(request, attachments, request.guideline_patterns);
+    const session = await dispatchAgentSession({
+      userQuery: promptContent,
+      // Use full prompt content instead of just request.prompt
+      extraAttachments: attachments,
+      wait: this.config.waitForResponse,
+      dryRun: this.config.dryRun,
+      vscodeCmd: this.config.command,
+      subagentRoot: this.config.subagentRoot,
+      workspaceTemplate: this.config.workspaceTemplate,
+      silent: true
+    });
+    if (session.exitCode !== 0 || !session.responseFile) {
+      const failure = session.error ?? "VS Code subagent did not produce a response";
+      throw new Error(failure);
+    }
+    if (this.config.dryRun) {
       return {
-        text: responseText,
+        text: "",
         raw: {
           session,
-          promptFile: promptPath,
           attachments
         }
       };
-    } finally {
-      await rm2(directory, { recursive: true, force: true });
     }
+    const responseText = await readFile22(session.responseFile, "utf8");
+    return {
+      text: responseText,
+      raw: {
+        session,
+        attachments
+      }
+    };
   }
 };
-function buildPromptDocument(request, attachments) {
+function buildPromptDocument(request, attachments, guidelinePatterns) {
   const parts = [];
-  const instructionFiles = collectInstructionFiles(attachments);
-  if (instructionFiles.length > 0) {
-    parts.push(buildMandatoryPrereadBlock(instructionFiles));
-  }
-  parts.push(`# BbEval Request`);
-  if (request.testCaseId) {
-    parts.push(`- Test Case: ${request.testCaseId}`);
-  }
-  if (request.metadata?.target) {
-    parts.push(`- Target: ${String(request.metadata.target)}`);
-  }
-  parts.push("\n## Task\n", request.prompt.trim());
-  if (request.guidelines && request.guidelines.trim().length > 0) {
-    parts.push("\n## Guidelines\n", request.guidelines.trim());
-  }
-  if (attachments && attachments.length > 0) {
-    const attachmentList = attachments.map((item) => `- ${item}`).join("\n");
-    parts.push("\n## Attachments\n", attachmentList);
+  const guidelineFiles = collectGuidelineFiles(attachments, guidelinePatterns);
+  if (guidelineFiles.length > 0) {
+    parts.push("\n", buildMandatoryPrereadBlock(guidelineFiles));
   }
+  parts.push("\n[[ ## user_query ## ]]\n", request.prompt.trim());
   return parts.join("\n").trim();
 }
-function buildMandatoryPrereadBlock(instructionFiles) {
-  if (instructionFiles.length === 0) {
+function buildMandatoryPrereadBlock(guidelineFiles) {
+  if (guidelineFiles.length === 0) {
     return "";
   }
   const fileList = [];
-  const tokenList = [];
   let counter = 0;
-  for (const absolutePath of instructionFiles) {
+  for (const absolutePath of guidelineFiles) {
     counter += 1;
     const fileName = path22.basename(absolutePath);
     const fileUri = pathToFileUri(absolutePath);
-    fileList.push(`[${fileName}](${fileUri})`);
-    tokenList.push(`INSTRUCTIONS_READ: \`${fileName}\` i=${counter} SHA256=<hex>`);
+    fileList.push(`* [${fileName}](${fileUri})`);
   }
-  const filesText = fileList.join(", ");
-  const tokensText = tokenList.join("\n");
+  const filesText = fileList.join("\n");
   const instruction = [
-    `Read all instruction files: ${filesText}.`,
-    `After reading each file, compute its SHA256 hash using this PowerShell command:`,
-    "`Get-FileHash -Algorithm SHA256 -LiteralPath '<file-path>' | Select-Object -ExpandProperty Hash`.",
-    `Then include, at the top of your reply, these exact tokens on separate lines:
+    `Read all guideline files:
+${filesText}.
 `,
-    tokensText,
-    `
-Replace \`<hex>\` with the actual SHA256 hash value computed from the PowerShell command.`,
     `If any file is missing, fail with ERROR: missing-file <filename> and stop.
 `,
-    `Then fetch all documentation required by the instructions before proceeding with your task.`
-  ].join(" ");
-  return `[[ ## mandatory_pre_read ## ]]
-
-${instruction}
-
-`;
+    `Then apply system_instructions on the user query below.`
+  ].join("");
+  return `${instruction}`;
 }
-function collectInstructionFiles(attachments) {
+function collectGuidelineFiles(attachments, guidelinePatterns) {
   if (!attachments || attachments.length === 0) {
     return [];
   }
   const unique = /* @__PURE__ */ new Map();
   for (const attachment of attachments) {
-    if (!isInstructionPath(attachment)) {
-      continue;
-    }
     const absolutePath = path22.resolve(attachment);
-    if (!unique.has(absolutePath)) {
-      unique.set(absolutePath, absolutePath);
+    const normalized = absolutePath.split(path22.sep).join("/");
+    if (isGuidelineFile(normalized, guidelinePatterns)) {
+      if (!unique.has(absolutePath)) {
+        unique.set(absolutePath, absolutePath);
+      }
     }
   }
   return Array.from(unique.values());
 }
-function isInstructionPath(filePath) {
-  const normalized = filePath.split(path22.sep).join("/");
-  return normalized.endsWith(".instructions.md") || normalized.includes("/instructions/") || normalized.endsWith(".prompt.md") || normalized.includes("/prompts/");
-}
 function pathToFileUri(filePath) {
   const absolutePath = path22.isAbsolute(filePath) ? filePath : path22.resolve(filePath);
   const normalizedPath = absolutePath.replace(/\\/g, "/");
@@ -11675,14 +11799,6 @@ function pathToFileUri(filePath) {
   }
   return `file://${normalizedPath}`;
 }
-function composeUserQuery(request) {
-  const segments = [];
-  segments.push(request.prompt.trim());
-  if (request.guidelines && request.guidelines.trim().length > 0) {
-    segments.push("\nGuidelines:\n", request.guidelines.trim());
-  }
-  return segments.join("\n").trim();
-}
 function normalizeAttachments(attachments) {
   if (!attachments || attachments.length === 0) {
     return void 0;
@@ -11734,18 +11850,24 @@ total unlocked subagents available: ${result.created.length + result.skippedExis
 function isRecord(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value);
 }
-function checkVersion(parsed, absolutePath) {
-  const version = typeof parsed.version === "number" ? parsed.version : typeof parsed.version === "string" ? parseFloat(parsed.version) : void 0;
-  if (version === void 0) {
+function checkSchema(parsed, absolutePath) {
+  const schema = parsed.$schema;
+  if (schema === void 0) {
+    throw new Error(
+      `Missing $schema field in targets.yaml at ${absolutePath}.
+Please add '$schema: ${TARGETS_SCHEMA_V2}' at the top of the file.`
+    );
+  }
+  if (typeof schema !== "string") {
     throw new Error(
-      `Missing version field in targets.yaml at ${absolutePath}.
-Please add 'version: 2.0' at the top of the file.`
+      `Invalid $schema field in targets.yaml at ${absolutePath}.
+Expected a string value '${TARGETS_SCHEMA_V2}'.`
     );
   }
-  if (version < 2) {
+  if (schema !== TARGETS_SCHEMA_V2) {
     throw new Error(
-      `Outdated targets.yaml format (version ${version}) at ${absolutePath}.
-Please update to version 2.0 format with 'targets' array.`
+      `Invalid $schema '${schema}' in targets.yaml at ${absolutePath}.
+Expected '${TARGETS_SCHEMA_V2}'.`
     );
   }
 }
@@ -11777,7 +11899,7 @@ function assertTargetDefinition(value, index, filePath) {
     judge_target: typeof judgeTarget === "string" ? judgeTarget : void 0
   };
 }
-async function fileExists22(filePath) {
+async function fileExists3(filePath) {
   try {
     await access22(filePath, constants22.F_OK);
     return true;
@@ -11787,15 +11909,15 @@ async function fileExists22(filePath) {
 }
 async function readTargetDefinitions(filePath) {
   const absolutePath = path32.resolve(filePath);
-  if (!await fileExists22(absolutePath)) {
+  if (!await fileExists3(absolutePath)) {
     throw new Error(`targets.yaml not found at ${absolutePath}`);
   }
   const raw = await readFile3(absolutePath, "utf8");
   const parsed = parse22(raw);
   if (!isRecord(parsed)) {
-    throw new Error(`targets.yaml at ${absolutePath} must be a YAML object with 'version' and 'targets' fields`);
+    throw new Error(`targets.yaml at ${absolutePath} must be a YAML object with '$schema' and 'targets' fields`);
   }
-  checkVersion(parsed, absolutePath);
+  checkSchema(parsed, absolutePath);
   const targets = extractTargetsArray(parsed, absolutePath);
   const definitions = targets.map((entry, index) => assertTargetDefinition(entry, index, absolutePath));
   return definitions;
@@ -12002,7 +12124,7 @@ function extractKeyTerms(aspect, maxTerms = 5) {
 var HeuristicGrader = class {
   kind = "heuristic";
   grade(context2) {
-    const expectedAspects = extractAspects(context2.testCase.expected_assistant_raw);
+    const expectedAspects = extractAspects(context2.evalCase.expected_assistant_raw);
     const result = scoreCandidateResponse(context2.candidate, expectedAspects);
     const misses = [...result.misses];
     if (expectedAspects.length === 0 && isErrorLike(context2.candidate)) {
@@ -12035,14 +12157,14 @@ var QualityGrader = class {
     if (!judgeProvider) {
       throw new Error("No judge provider available for LLM grading");
     }
-    const prompt = buildQualityPrompt(context2.testCase, context2.candidate);
+    const prompt = buildQualityPrompt(context2.evalCase, context2.candidate);
     const metadata = {
       systemPrompt: QUALITY_SYSTEM_PROMPT
     };
     const response = await judgeProvider.invoke({
       prompt,
       metadata,
-      testCaseId: context2.testCase.id,
+      evalCaseId: context2.evalCase.id,
       attempt: context2.attempt,
       maxOutputTokens: this.maxOutputTokens,
       temperature: this.temperature
@@ -12088,16 +12210,16 @@ var QUALITY_SYSTEM_PROMPT = [
 function buildQualityPrompt(testCase, candidate) {
   const parts = [
     "[[ ## expected_outcome ## ]]",
-    testCase.outcome,
+    testCase.outcome.trim(),
     "",
     "[[ ## request ## ]]",
-    testCase.task,
+    testCase.task.trim(),
     "",
     "[[ ## reference_answer ## ]]",
-    testCase.expected_assistant_raw,
+    testCase.expected_assistant_raw.trim(),
     "",
     "[[ ## generated_answer ## ]]",
-    candidate,
+    candidate.trim(),
     "",
     "Respond with a single JSON object matching the schema described in the system prompt."
   ];
@@ -12330,17 +12452,17 @@ async function runEvaluation(options) {
     cache,
     useCache,
     now,
-    testId,
+    evalId,
     verbose,
     onResult,
     onProgress
   } = options;
-  const load = loadTestCases;
-  const testCases = await load(testFilePath, repoRoot, { verbose });
-  const filteredTestCases = filterTestCases(testCases, testId);
-  if (filteredTestCases.length === 0) {
-    if (testId) {
-      throw new Error(`Test case with id '${testId}' not found in ${testFilePath}`);
+  const load = loadEvalCases;
+  const evalCases = await load(testFilePath, repoRoot, { verbose });
+  const filteredEvalCases = filterEvalCases(evalCases, evalId);
+  if (filteredEvalCases.length === 0) {
+    if (evalId) {
+      throw new Error(`Test case with id '${evalId}' not found in ${testFilePath}`);
     }
     return [];
   }
@@ -12384,11 +12506,11 @@ async function runEvaluation(options) {
   };
   const graderRegistry = buildGraderRegistry(graders, resolveJudgeProvider);
   const primaryProvider = getOrCreateProvider(target);
-  if (onProgress && filteredTestCases.length > 0) {
-    for (let i6 = 0; i6 < filteredTestCases.length; i6++) {
+  if (onProgress && filteredEvalCases.length > 0) {
+    for (let i6 = 0; i6 < filteredEvalCases.length; i6++) {
       await onProgress({
         workerId: i6 + 1,
-        testId: filteredTestCases[i6].id,
+        evalId: filteredEvalCases[i6].id,
         status: "pending"
       });
     }
@@ -12396,23 +12518,23 @@ async function runEvaluation(options) {
   const workers = options.maxConcurrency ?? target.workers ?? 1;
   const limit = pLimit(workers);
   let nextWorkerId = 1;
-  const workerIdByTestId = /* @__PURE__ */ new Map();
-  const promises = filteredTestCases.map(
-    (testCase) => limit(async () => {
+  const workerIdByEvalId = /* @__PURE__ */ new Map();
+  const promises = filteredEvalCases.map(
+    (evalCase) => limit(async () => {
       const workerId = nextWorkerId++;
-      workerIdByTestId.set(testCase.id, workerId);
+      workerIdByEvalId.set(evalCase.id, workerId);
       if (onProgress) {
         await onProgress({
           workerId,
-          testId: testCase.id,
+          evalId: evalCase.id,
           status: "running",
           startedAt: Date.now()
         });
       }
       try {
         const judgeProvider = await resolveJudgeProvider(target);
-        const result = await runTestCase({
-          testCase,
+        const result = await runEvalCase({
+          evalCase,
           provider: primaryProvider,
           target,
           graders: graderRegistry,
@@ -12427,7 +12549,7 @@ async function runEvaluation(options) {
         if (onProgress) {
           await onProgress({
             workerId,
-            testId: testCase.id,
+            evalId: evalCase.id,
             status: "completed",
             startedAt: 0,
             // Not used for completed status
@@ -12442,7 +12564,7 @@ async function runEvaluation(options) {
         if (onProgress) {
           await onProgress({
             workerId,
-            testId: testCase.id,
+            evalId: evalCase.id,
             status: "failed",
             completedAt: Date.now(),
             error: error instanceof Error ? error.message : String(error)
@@ -12459,10 +12581,10 @@ async function runEvaluation(options) {
     if (outcome.status === "fulfilled") {
       results.push(outcome.value);
     } else {
-      const testCase = filteredTestCases[i6];
-      const promptInputs = await buildPromptInputs(testCase);
+      const evalCase = filteredEvalCases[i6];
+      const promptInputs = await buildPromptInputs(evalCase);
       const errorResult = buildErrorResult(
-        testCase,
+        evalCase,
         target.name,
         (now ?? (() => /* @__PURE__ */ new Date()))(),
         outcome.reason,
@@ -12476,9 +12598,9 @@ async function runEvaluation(options) {
   }
   return results;
 }
-async function runTestCase(options) {
+async function runEvalCase(options) {
   const {
-    testCase,
+    evalCase,
     provider,
     target,
     graders,
@@ -12491,11 +12613,11 @@ async function runTestCase(options) {
     signal,
     judgeProvider
   } = options;
-  const promptInputs = await buildPromptInputs(testCase);
+  const promptInputs = await buildPromptInputs(evalCase);
   if (promptDumpDir) {
-    await dumpPrompt(promptDumpDir, testCase, promptInputs);
+    await dumpPrompt(promptDumpDir, evalCase, promptInputs);
   }
-  const cacheKey = useCache ? createCacheKey(provider, target, testCase, promptInputs) : void 0;
+  const cacheKey = useCache ? createCacheKey(provider, target, evalCase, promptInputs) : void 0;
   let cachedResponse;
   if (cacheKey && cache) {
     cachedResponse = await cache.get(cacheKey);
@@ -12508,7 +12630,7 @@ async function runTestCase(options) {
   while (!providerResponse && attempt < attemptBudget) {
     try {
       providerResponse = await invokeProvider(provider, {
-        testCase,
+        evalCase,
         target,
         promptInputs,
         attempt,
@@ -12521,12 +12643,12 @@ async function runTestCase(options) {
         attempt += 1;
         continue;
       }
-      return buildErrorResult(testCase, target.name, nowFn(), error, promptInputs);
+      return buildErrorResult(evalCase, target.name, nowFn(), error, promptInputs);
     }
   }
   if (!providerResponse) {
     return buildErrorResult(
-      testCase,
+      evalCase,
       target.name,
       nowFn(),
       lastError ?? new Error("Provider did not return a response"),
@@ -12536,7 +12658,7 @@ async function runTestCase(options) {
   if (cacheKey && cache && !cachedResponse) {
     await cache.set(cacheKey, providerResponse);
   }
-  const graderKind = testCase.grader ?? "heuristic";
+  const graderKind = evalCase.grader ?? "heuristic";
   const activeGrader = graders[graderKind] ?? graders.heuristic;
   if (!activeGrader) {
     throw new Error(`No grader registered for kind '${graderKind}'`);
@@ -12545,7 +12667,7 @@ async function runTestCase(options) {
   try {
     const gradeTimestamp = nowFn();
     grade = await activeGrader.grade({
-      testCase,
+      evalCase,
       candidate: providerResponse.text ?? "",
       target,
       provider,
@@ -12555,17 +12677,18 @@ async function runTestCase(options) {
       judgeProvider
     });
   } catch (error) {
-    return buildErrorResult(testCase, target.name, nowFn(), error, promptInputs);
+    return buildErrorResult(evalCase, target.name, nowFn(), error, promptInputs);
   }
   const completedAt = nowFn();
   const rawRequest = {
     request: promptInputs.request,
     guidelines: promptInputs.guidelines,
-    guideline_paths: testCase.guideline_paths
+    guideline_paths: evalCase.guideline_paths,
+    system_message: promptInputs.systemMessage ?? ""
   };
   return {
-    test_id: testCase.id,
-    conversation_id: testCase.conversation_id,
+    eval_id: evalCase.id,
+    conversation_id: evalCase.conversation_id,
     score: grade.score,
     hits: grade.hits,
     misses: grade.misses,
@@ -12579,11 +12702,11 @@ async function runTestCase(options) {
     grader_raw_request: grade.graderRawRequest
   };
 }
-function filterTestCases(testCases, testId) {
-  if (!testId) {
-    return testCases;
+function filterEvalCases(evalCases, evalId) {
+  if (!evalId) {
+    return evalCases;
   }
-  return testCases.filter((testCase) => testCase.id === testId);
+  return evalCases.filter((evalCase) => evalCase.id === evalId);
 }
 function buildGraderRegistry(overrides, resolveJudgeProvider) {
   const heuristic = overrides?.heuristic ?? new HeuristicGrader();
@@ -12601,16 +12724,16 @@ function buildGraderRegistry(overrides, resolveJudgeProvider) {
     llm_judge: llmJudge
   };
 }
-async function dumpPrompt(directory, testCase, promptInputs) {
+async function dumpPrompt(directory, evalCase, promptInputs) {
   const timestamp = (/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-");
-  const filename = `${timestamp}_${sanitizeFilename(testCase.id)}.json`;
+  const filename = `${timestamp}_${sanitizeFilename(evalCase.id)}.json`;
   const filePath = path42.resolve(directory, filename);
   await mkdir3(path42.dirname(filePath), { recursive: true });
   const payload = {
-    test_id: testCase.id,
+    eval_id: evalCase.id,
     request: promptInputs.request,
     guidelines: promptInputs.guidelines,
-    guideline_paths: testCase.guideline_paths
+    guideline_paths: evalCase.guideline_paths
   };
   await writeFile22(filePath, JSON.stringify(payload, null, 2), "utf8");
 }
@@ -12622,7 +12745,7 @@ function sanitizeFilename(value) {
   return sanitized.length > 0 ? sanitized : randomUUID2();
 }
 async function invokeProvider(provider, options) {
-  const { testCase, target, promptInputs, attempt, agentTimeoutMs, signal } = options;
+  const { evalCase, target, promptInputs, attempt, agentTimeoutMs, signal } = options;
   const controller = new AbortController();
   const timeout = agentTimeoutMs ? setTimeout(() => controller.abort(), agentTimeoutMs) : void 0;
   if (signal) {
@@ -12632,12 +12755,12 @@ async function invokeProvider(provider, options) {
     return await provider.invoke({
       prompt: promptInputs.request,
       guidelines: promptInputs.guidelines,
-      attachments: testCase.guideline_paths,
-      testCaseId: testCase.id,
+      guideline_patterns: evalCase.guideline_patterns,
+      attachments: evalCase.file_paths,
+      evalCaseId: evalCase.id,
       attempt,
       metadata: {
-        target: target.name,
-        grader: testCase.grader
+        systemPrompt: promptInputs.systemMessage ?? ""
       },
       signal: controller.signal
     });
@@ -12647,17 +12770,18 @@ async function invokeProvider(provider, options) {
     }
   }
 }
-function buildErrorResult(testCase, targetName, timestamp, error, promptInputs) {
+function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs) {
   const message = error instanceof Error ? error.message : String(error);
   const rawRequest = {
     request: promptInputs.request,
     guidelines: promptInputs.guidelines,
-    guideline_paths: testCase.guideline_paths,
+    guideline_paths: evalCase.guideline_paths,
+    system_message: promptInputs.systemMessage ?? "",
     error: message
   };
   return {
-    test_id: testCase.id,
-    conversation_id: testCase.conversation_id,
+    eval_id: evalCase.id,
+    conversation_id: evalCase.conversation_id,
     score: 0,
     hits: [],
     misses: [`Error: ${message}`],
@@ -12669,13 +12793,14 @@ function buildErrorResult(testCase, targetName, timestamp, error, promptInputs)
     raw_request: rawRequest
   };
 }
-function createCacheKey(provider, target, testCase, promptInputs) {
+function createCacheKey(provider, target, evalCase, promptInputs) {
   const hash = createHash("sha256");
   hash.update(provider.id);
   hash.update(target.name);
-  hash.update(testCase.id);
+  hash.update(evalCase.id);
   hash.update(promptInputs.request);
   hash.update(promptInputs.guidelines);
+  hash.update(promptInputs.systemMessage ?? "");
   return hash.digest("hex");
 }
 function isTimeoutLike(error) {
@@ -12721,7 +12846,7 @@ function uniqueDirs(directories) {
   }
   return result;
 }
-async function fileExists3(filePath) {
+async function fileExists4(filePath) {
   try {
     await access4(filePath, constants4.F_OK);
     return true;
@@ -12757,7 +12882,7 @@ async function loadEnvFromHierarchy(options) {
   ]);
   for (const dir of searchDirs) {
     const candidate = path8.join(dir, ".env");
-    if (await fileExists3(candidate)) {
+    if (await fileExists4(candidate)) {
       loadDotenv({ path: candidate, override: false });
       if (verbose) {
         console.log(`Loaded environment from: ${candidate}`);
@@ -13127,9 +13252,9 @@ var ProgressDisplay = class {
       this.scheduleRender();
     } else {
       if (progress.status === "completed") {
-        console.log(`\u2713 Test ${progress.testId} completed`);
+        console.log(`\u2713 Test ${progress.evalId} completed`);
       } else if (progress.status === "failed") {
-        console.log(`\u2717 Test ${progress.testId} failed${progress.error ? `: ${progress.error}` : ""}`);
+        console.log(`\u2717 Test ${progress.evalId} failed${progress.error ? `: ${progress.error}` : ""}`);
       }
     }
   }
@@ -13162,7 +13287,7 @@ var ProgressDisplay = class {
     const statusIcon = this.getStatusIcon(worker.status);
     const elapsed = worker.startedAt ? this.formatElapsed(Date.now() - worker.startedAt) : "";
     const timeLabel = elapsed ? ` (${elapsed})` : "";
-    let testLabel = worker.testId;
+    let testLabel = worker.evalId;
     if (testLabel.length > 50) {
       testLabel = testLabel.substring(0, 47) + "...";
     }
@@ -13316,7 +13441,7 @@ function formatEvaluationSummary(summary) {
   lines.push("\n==================================================");
   lines.push("EVALUATION SUMMARY");
   lines.push("==================================================");
-  lines.push(`Total test cases: ${summary.total}`);
+  lines.push(`Total eval cases: ${summary.total}`);
   lines.push(`Mean score: ${formatScore(summary.mean)}`);
   lines.push(`Median score: ${formatScore(summary.median)}`);
   lines.push(`Min score: ${formatScore(summary.min)}`);
@@ -13329,13 +13454,13 @@ function formatEvaluationSummary(summary) {
     const [start, end] = bin.range;
     lines.push(`  ${start.toFixed(1)}-${end.toFixed(1)}: ${bin.count}`);
   }
-  lines.push("\nTop performing test cases:");
+  lines.push("\nTop performing eval cases:");
   summary.topResults.forEach((result, index) => {
-    lines.push(`  ${index + 1}. ${result.test_id}: ${formatScore(result.score)}`);
+    lines.push(`  ${index + 1}. ${result.eval_id}: ${formatScore(result.score)}`);
   });
-  lines.push("\nLowest performing test cases:");
+  lines.push("\nLowest performing eval cases:");
   summary.bottomResults.forEach((result, index) => {
-    lines.push(`  ${index + 1}. ${result.test_id}: ${formatScore(result.score)}`);
+    lines.push(`  ${index + 1}. ${result.eval_id}: ${formatScore(result.score)}`);
   });
   return lines.join("\n");
 }
@@ -13349,11 +13474,9 @@ var TARGET_FILE_CANDIDATES = [
   "targets.yaml",
   "targets.yml",
   path11.join(".agentv", "targets.yaml"),
-  path11.join(".agentv", "targets.yml"),
-  path11.join(".bbeval", "targets.yaml"),
-  path11.join(".bbeval", "targets.yml")
+  path11.join(".agentv", "targets.yml")
 ];
-async function fileExists4(filePath) {
+async function fileExists5(filePath) {
   try {
     await access5(filePath, constants5.F_OK);
     return true;
@@ -13375,56 +13498,30 @@ async function readTestSuiteTarget(testFilePath) {
   }
   return void 0;
 }
-function buildDirectoryChain(testFilePath, repoRoot, cwd) {
-  const directories = [];
-  const seen = /* @__PURE__ */ new Set();
-  const boundary = path11.resolve(repoRoot);
-  let current = path11.resolve(path11.dirname(testFilePath));
-  while (current !== void 0) {
-    if (!seen.has(current)) {
-      directories.push(current);
-      seen.add(current);
-    }
-    if (current === boundary) {
-      break;
-    }
-    const parent = path11.dirname(current);
-    if (parent === current) {
-      break;
-    }
-    current = parent;
-  }
-  if (!seen.has(boundary)) {
-    directories.push(boundary);
-    seen.add(boundary);
-  }
-  const resolvedCwd = path11.resolve(cwd);
-  if (!seen.has(resolvedCwd)) {
-    directories.push(resolvedCwd);
-    seen.add(resolvedCwd);
-  }
-  return directories;
-}
 async function discoverTargetsFile(options) {
   const { explicitPath, testFilePath, repoRoot, cwd } = options;
   if (explicitPath) {
     const resolvedExplicit = path11.resolve(explicitPath);
-    if (await fileExists4(resolvedExplicit)) {
+    if (await fileExists5(resolvedExplicit)) {
       return resolvedExplicit;
     }
     for (const candidate of TARGET_FILE_CANDIDATES) {
       const nested = path11.join(resolvedExplicit, candidate);
-      if (await fileExists4(nested)) {
+      if (await fileExists5(nested)) {
         return nested;
       }
     }
     throw new Error(`targets.yaml not found at provided path: ${resolvedExplicit}`);
   }
-  const directories = buildDirectoryChain(testFilePath, repoRoot, cwd);
+  const directories = [...buildDirectoryChain(testFilePath, repoRoot)];
+  const resolvedCwd = path11.resolve(cwd);
+  if (!directories.includes(resolvedCwd)) {
+    directories.push(resolvedCwd);
+  }
   for (const directory of directories) {
     for (const candidate of TARGET_FILE_CANDIDATES) {
       const fullPath = path11.join(directory, candidate);
-      if (await fileExists4(fullPath)) {
+      if (await fileExists5(fullPath)) {
         return fullPath;
       }
     }
@@ -13525,7 +13622,7 @@ function normalizeOptions(rawOptions) {
   return {
     target: normalizeString(rawOptions.target),
     targetsPath: normalizeString(rawOptions.targets),
-    testId: normalizeString(rawOptions.testId),
+    evalId: normalizeString(rawOptions.evalId),
     workers: workers > 0 ? workers : void 0,
     outPath: normalizeString(rawOptions.out),
     format,
@@ -13672,7 +13769,7 @@ async function runEvalCommand(input) {
       promptDumpDir,
       cache,
       useCache: options.cache,
-      testId: options.testId,
+      evalId: options.evalId,
       verbose: options.verbose,
       maxConcurrency: resolvedWorkers,
       onResult: async (result) => {
@@ -13685,7 +13782,7 @@ async function runEvalCommand(input) {
         }
         progressDisplay.updateWorker({
           workerId: event.workerId,
-          testId: event.testId,
+          evalId: event.evalId,
           status: event.status,
           startedAt: event.startedAt,
           completedAt: event.completedAt,
@@ -13735,7 +13832,7 @@ function parseInteger(value, fallback) {
   return parsed;
 }
 function registerEvalCommand(program) {
-  program.command("eval").description("Run BbEval test suites and report results").argument("<test-file>", "Path to the evaluation .test.yaml file").option("--target <name>", "Override target name from targets.yaml", "default").option("--targets <path>", "Path to targets.yaml (overrides discovery)").option("--test-id <id>", "Run only the test case with this identifier").option(
+  program.command("eval").description("Run eval suites and report results").argument("<eval-file>", "Path to the evaluation .yaml file").option("--target <name>", "Override target name from targets.yaml", "default").option("--targets <path>", "Path to targets.yaml (overrides discovery)").option("--eval-id <id>", "Run only the eval case with this identifier").option(
     "--workers <count>",
     "Number of parallel workers (default: 1, max: 50). Can also be set per-target in targets.yaml",
     (value) => parseInteger(value, 1)
@@ -13773,25 +13870,164 @@ function registerEvalCommand(program) {
   return program;
 }
 
-// src/commands/lint/lint-files.ts
-import { constants as constants7 } from "node:fs";
-import { access as access7, readdir as readdir3, stat as stat3 } from "node:fs/promises";
+// src/commands/init/index.ts
+import { existsSync, mkdirSync, writeFileSync } from "node:fs";
 import path14 from "node:path";
 
+// src/templates/index.ts
+import { readFileSync } from "node:fs";
+import path13 from "node:path";
+import { fileURLToPath as fileURLToPath2 } from "node:url";
+var TemplateManager = class {
+  static getTemplates() {
+    const currentDir = path13.dirname(fileURLToPath2(import.meta.url));
+    let templatesDir;
+    if (currentDir.includes(path13.sep + "dist")) {
+      templatesDir = path13.join(currentDir, "templates");
+    } else {
+      templatesDir = currentDir;
+    }
+    const evalBuildPrompt = readFileSync(
+      path13.join(templatesDir, "eval-build.prompt.md"),
+      "utf-8"
+    );
+    const evalSchema = readFileSync(
+      path13.join(templatesDir, "eval-schema.json"),
+      "utf-8"
+    );
+    const configSchema = readFileSync(
+      path13.join(templatesDir, "config-schema.json"),
+      "utf-8"
+    );
+    return [
+      {
+        path: "prompts/eval-build.prompt.md",
+        content: evalBuildPrompt
+      },
+      {
+        path: "contexts/eval-schema.json",
+        content: evalSchema
+      },
+      {
+        path: "contexts/config-schema.json",
+        content: configSchema
+      }
+    ];
+  }
+};
+
+// src/commands/init/index.ts
+async function initCommand(options = {}) {
+  const targetPath = path14.resolve(options.targetPath ?? ".");
+  const githubDir = path14.join(targetPath, ".github");
+  if (!existsSync(githubDir)) {
+    mkdirSync(githubDir, { recursive: true });
+  }
+  const templates = TemplateManager.getTemplates();
+  for (const template of templates) {
+    const targetFilePath = path14.join(githubDir, template.path);
+    const targetDirPath = path14.dirname(targetFilePath);
+    if (!existsSync(targetDirPath)) {
+      mkdirSync(targetDirPath, { recursive: true });
+    }
+    writeFileSync(targetFilePath, template.content, "utf-8");
+    console.log(`Created ${path14.relative(targetPath, targetFilePath)}`);
+  }
+  console.log("\nAgentV initialized successfully!");
+  console.log(`
+Files installed to ${path14.relative(targetPath, githubDir)}:`);
+  templates.forEach((t) => console.log(`  - ${t.path}`));
+  console.log("\nYou can now create eval files using the schema and prompt templates.");
+}
+
+// src/commands/validate/format-output.ts
+var ANSI_RED = "\x1B[31m";
+var ANSI_YELLOW2 = "\x1B[33m";
+var ANSI_GREEN = "\x1B[32m";
+var ANSI_CYAN = "\x1B[36m";
+var ANSI_BOLD = "\x1B[1m";
+var ANSI_RESET2 = "\x1B[0m";
+function formatSummary(summary, useColors) {
+  const lines = [];
+  lines.push("");
+  lines.push(formatHeader("Validation Summary", useColors));
+  lines.push("");
+  for (const result of summary.results) {
+    lines.push(formatFileResult(result, useColors));
+  }
+  lines.push("");
+  lines.push(formatStats(summary, useColors));
+  lines.push("");
+  return lines.join("\n");
+}
+function formatHeader(text, useColors) {
+  if (useColors) {
+    return `${ANSI_BOLD}${ANSI_CYAN}${text}${ANSI_RESET2}`;
+  }
+  return text;
+}
+function formatFileResult(result, useColors) {
+  const lines = [];
+  const status = result.valid ? "\u2713" : "\u2717";
+  const statusColor = result.valid ? ANSI_GREEN : ANSI_RED;
+  const statusText = useColors ? `${statusColor}${status}${ANSI_RESET2}` : status;
+  const fileName = result.filePath;
+  lines.push(`${statusText} ${fileName}`);
+  if (result.errors.length > 0) {
+    for (const error of result.errors) {
+      lines.push(formatError(error, useColors));
+    }
+  }
+  return lines.join("\n");
+}
+function formatError(error, useColors) {
+  const prefix = error.severity === "error" ? "  \u2717" : "  \u26A0";
+  const color = error.severity === "error" ? ANSI_RED : ANSI_YELLOW2;
+  const coloredPrefix = useColors ? `${color}${prefix}${ANSI_RESET2}` : prefix;
+  const location = error.location ? ` [${error.location}]` : "";
+  return `${coloredPrefix}${location} ${error.message}`;
+}
+function formatStats(summary, useColors) {
+  const lines = [];
+  const totalText = `Total files: ${summary.totalFiles}`;
+  const validText = `Valid: ${summary.validFiles}`;
+  const invalidText = `Invalid: ${summary.invalidFiles}`;
+  if (useColors) {
+    lines.push(`${ANSI_BOLD}${totalText}${ANSI_RESET2}`);
+    lines.push(`${ANSI_GREEN}${validText}${ANSI_RESET2}`);
+    if (summary.invalidFiles > 0) {
+      lines.push(`${ANSI_RED}${invalidText}${ANSI_RESET2}`);
+    } else {
+      lines.push(invalidText);
+    }
+  } else {
+    lines.push(totalText);
+    lines.push(validText);
+    lines.push(invalidText);
+  }
+  return lines.join("\n");
+}
+function isTTY() {
+  return process.stdout.isTTY ?? false;
+}
+
 // ../../packages/core/dist/evaluation/validation/index.js
 import { readFile as readFile5 } from "node:fs/promises";
 import { parse as parse5 } from "yaml";
 import { readFile as readFile23 } from "node:fs/promises";
-import path13 from "node:path";
+import path15 from "node:path";
 import { parse as parse23 } from "yaml";
 import { readFile as readFile32 } from "node:fs/promises";
 import path23 from "node:path";
 import { parse as parse32 } from "yaml";
 import { readFile as readFile42 } from "node:fs/promises";
-import path33 from "node:path";
 import { parse as parse42 } from "yaml";
+import { readFile as readFile52 } from "node:fs/promises";
+import path33 from "node:path";
+import { parse as parse52 } from "yaml";
 var SCHEMA_EVAL_V22 = "agentv-eval-v2";
 var SCHEMA_TARGETS_V2 = "agentv-targets-v2";
+var SCHEMA_CONFIG_V22 = "agentv-config-v2";
 async function detectFileType(filePath) {
   try {
     const content = await readFile5(filePath, "utf8");
@@ -13809,6 +14045,8 @@ async function detectFileType(filePath) {
         return "eval";
       case SCHEMA_TARGETS_V2:
         return "targets";
+      case SCHEMA_CONFIG_V22:
+        return "config";
       default:
         return "unknown";
     }
@@ -13822,7 +14060,7 @@ function isObject(value) {
 }
 async function validateEvalFile(filePath) {
   const errors = [];
-  const absolutePath = path13.resolve(filePath);
+  const absolutePath = path15.resolve(filePath);
   let parsed;
   try {
     const content = await readFile23(absolutePath, "utf8");
@@ -14008,7 +14246,6 @@ function validateMessages(messages, location, filePath, errors) {
     }
   }
 }
-var SCHEMA_TARGETS_V22 = "agentv-targets-v2";
 function isObject2(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value);
 }
@@ -14046,8 +14283,8 @@ async function validateTargetsFile(filePath) {
     };
   }
   const schema = parsed["$schema"];
-  if (schema !== SCHEMA_TARGETS_V22) {
-    const message = typeof schema === "string" ? `Invalid $schema value '${schema}'. Expected '${SCHEMA_TARGETS_V22}'` : `Missing required field '$schema'. Expected '${SCHEMA_TARGETS_V22}'`;
+  if (schema !== TARGETS_SCHEMA_V2) {
+    const message = typeof schema === "string" ? `Invalid $schema value '${schema}'. Expected '${TARGETS_SCHEMA_V2}'` : `Missing required field '$schema'. Expected '${TARGETS_SCHEMA_V2}'`;
     errors.push({
       severity: "error",
       filePath: absolutePath,
@@ -14070,7 +14307,7 @@ async function validateTargetsFile(filePath) {
       errors
     };
   }
-  const knownProviders = ["azure", "openai", "anthropic", "bedrock", "vertex"];
+  const knownProviders = [...KNOWN_PROVIDERS, ...PROVIDER_ALIASES];
   for (let i6 = 0; i6 < targets.length; i6++) {
     const target = targets[i6];
     const location = `targets[${i6}]`;
@@ -14134,6 +14371,80 @@ async function validateTargetsFile(filePath) {
     errors
   };
 }
+var SCHEMA_CONFIG_V222 = "agentv-config-v2";
+async function validateConfigFile(filePath) {
+  const errors = [];
+  try {
+    const content = await readFile42(filePath, "utf8");
+    const parsed = parse42(content);
+    if (typeof parsed !== "object" || parsed === null) {
+      errors.push({
+        severity: "error",
+        filePath,
+        message: "Config file must contain a valid YAML object"
+      });
+      return { valid: false, filePath, fileType: "config", errors };
+    }
+    const config = parsed;
+    const schema = config["$schema"];
+    if (schema !== SCHEMA_CONFIG_V222) {
+      const message = typeof schema === "string" ? `Invalid $schema value '${schema}'. Expected '${SCHEMA_CONFIG_V222}'` : `Missing required field '$schema'. Please add '$schema: ${SCHEMA_CONFIG_V222}' at the top of the file.`;
+      errors.push({
+        severity: "error",
+        filePath,
+        location: "$schema",
+        message
+      });
+    }
+    const guidelinePatterns = config["guideline_patterns"];
+    if (guidelinePatterns !== void 0) {
+      if (!Array.isArray(guidelinePatterns)) {
+        errors.push({
+          severity: "error",
+          filePath,
+          location: "guideline_patterns",
+          message: "Field 'guideline_patterns' must be an array"
+        });
+      } else if (!guidelinePatterns.every((p) => typeof p === "string")) {
+        errors.push({
+          severity: "error",
+          filePath,
+          location: "guideline_patterns",
+          message: "All entries in 'guideline_patterns' must be strings"
+        });
+      } else if (guidelinePatterns.length === 0) {
+        errors.push({
+          severity: "warning",
+          filePath,
+          location: "guideline_patterns",
+          message: "Field 'guideline_patterns' is empty. Consider removing it or adding patterns."
+        });
+      }
+    }
+    const allowedFields = /* @__PURE__ */ new Set(["$schema", "guideline_patterns"]);
+    const unexpectedFields = Object.keys(config).filter((key2) => !allowedFields.has(key2));
+    if (unexpectedFields.length > 0) {
+      errors.push({
+        severity: "warning",
+        filePath,
+        message: `Unexpected fields: ${unexpectedFields.join(", ")}`
+      });
+    }
+    return {
+      valid: errors.filter((e) => e.severity === "error").length === 0,
+      filePath,
+      fileType: "config",
+      errors
+    };
+  } catch (error) {
+    errors.push({
+      severity: "error",
+      filePath,
+      message: `Failed to parse config file: ${error.message}`
+    });
+    return { valid: false, filePath, fileType: "config", errors };
+  }
+}
 function isObject3(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value);
 }
@@ -14152,8 +14463,8 @@ async function validateFileReferences(evalFilePath) {
   const searchRoots = buildSearchRoots(absolutePath, gitRoot);
   let parsed;
   try {
-    const content = await readFile42(absolutePath, "utf8");
-    parsed = parse42(content);
+    const content = await readFile52(absolutePath, "utf8");
+    parsed = parse52(content);
   } catch {
     return errors;
   }
@@ -14222,7 +14533,7 @@ async function validateMessagesFileRefs(messages, location, searchRoots, filePat
         });
       } else {
         try {
-          const fileContent = await readFile42(resolvedPath, "utf8");
+          const fileContent = await readFile52(resolvedPath, "utf8");
           if (fileContent.trim().length === 0) {
             errors.push({
               severity: "warning",
@@ -14244,12 +14555,15 @@ async function validateMessagesFileRefs(messages, location, searchRoots, filePat
   }
 }
 
-// src/commands/lint/lint-files.ts
-async function lintFiles(paths) {
+// src/commands/validate/validate-files.ts
+import { constants as constants7 } from "node:fs";
+import { access as access7, readdir as readdir3, stat as stat3 } from "node:fs/promises";
+import path16 from "node:path";
+async function validateFiles(paths) {
   const filePaths = await expandPaths(paths);
   const results = [];
   for (const filePath of filePaths) {
-    const result = await lintSingleFile(filePath);
+    const result = await validateSingleFile(filePath);
     results.push(result);
   }
   const validFiles = results.filter((r) => r.valid).length;
@@ -14261,8 +14575,8 @@ async function lintFiles(paths) {
     results
   };
 }
-async function lintSingleFile(filePath) {
-  const absolutePath = path14.resolve(filePath);
+async function validateSingleFile(filePath) {
+  const absolutePath = path16.resolve(filePath);
   const fileType = await detectFileType(absolutePath);
   if (fileType === "unknown") {
     return {
@@ -14273,7 +14587,7 @@ async function lintSingleFile(filePath) {
         {
           severity: "error",
           filePath: absolutePath,
-          message: "Missing or invalid $schema field. File must declare schema: 'agentv-eval-v2' or 'agentv-targets-v2'"
+          message: "Missing or invalid $schema field. File must declare schema: 'agentv-eval-v2', 'agentv-targets-v2', or 'agentv-config-v2'"
         }
       ]
     };
@@ -14291,15 +14605,17 @@ async function lintSingleFile(filePath) {
         };
       }
     }
-  } else {
+  } else if (fileType === "targets") {
     result = await validateTargetsFile(absolutePath);
+  } else {
+    result = await validateConfigFile(absolutePath);
   }
   return result;
 }
 async function expandPaths(paths) {
   const expanded = [];
   for (const inputPath of paths) {
-    const absolutePath = path14.resolve(inputPath);
+    const absolutePath = path16.resolve(inputPath);
     try {
       await access7(absolutePath, constants7.F_OK);
     } catch {
@@ -14323,7 +14639,7 @@ async function findYamlFiles(dirPath) {
   try {
     const entries = await readdir3(dirPath, { withFileTypes: true });
     for (const entry of entries) {
-      const fullPath = path14.join(dirPath, entry.name);
+      const fullPath = path16.join(dirPath, entry.name);
       if (entry.isDirectory()) {
         if (entry.name === "node_modules" || entry.name.startsWith(".")) {
           continue;
@@ -14340,98 +14656,27 @@ async function findYamlFiles(dirPath) {
   return results;
 }
 function isYamlFile(filePath) {
-  const ext = path14.extname(filePath).toLowerCase();
+  const ext = path16.extname(filePath).toLowerCase();
   return ext === ".yaml" || ext === ".yml";
 }
 
-// src/commands/lint/format-output.ts
-var ANSI_RED = "\x1B[31m";
-var ANSI_YELLOW2 = "\x1B[33m";
-var ANSI_GREEN = "\x1B[32m";
-var ANSI_CYAN = "\x1B[36m";
-var ANSI_BOLD = "\x1B[1m";
-var ANSI_RESET2 = "\x1B[0m";
-function formatSummary(summary, useColors) {
-  const lines = [];
-  lines.push("");
-  lines.push(formatHeader("Validation Summary", useColors));
-  lines.push("");
-  for (const result of summary.results) {
-    lines.push(formatFileResult(result, useColors));
-  }
-  lines.push("");
-  lines.push(formatStats(summary, useColors));
-  lines.push("");
-  return lines.join("\n");
-}
-function formatHeader(text, useColors) {
-  if (useColors) {
-    return `${ANSI_BOLD}${ANSI_CYAN}${text}${ANSI_RESET2}`;
-  }
-  return text;
-}
-function formatFileResult(result, useColors) {
-  const lines = [];
-  const status = result.valid ? "\u2713" : "\u2717";
-  const statusColor = result.valid ? ANSI_GREEN : ANSI_RED;
-  const statusText = useColors ? `${statusColor}${status}${ANSI_RESET2}` : status;
-  const fileName = result.filePath;
-  lines.push(`${statusText} ${fileName}`);
-  if (result.errors.length > 0) {
-    for (const error of result.errors) {
-      lines.push(formatError(error, useColors));
-    }
-  }
-  return lines.join("\n");
-}
-function formatError(error, useColors) {
-  const prefix = error.severity === "error" ? "  \u2717" : "  \u26A0";
-  const color = error.severity === "error" ? ANSI_RED : ANSI_YELLOW2;
-  const coloredPrefix = useColors ? `${color}${prefix}${ANSI_RESET2}` : prefix;
-  const location = error.location ? ` [${error.location}]` : "";
-  return `${coloredPrefix}${location} ${error.message}`;
-}
-function formatStats(summary, useColors) {
-  const lines = [];
-  const totalText = `Total files: ${summary.totalFiles}`;
-  const validText = `Valid: ${summary.validFiles}`;
-  const invalidText = `Invalid: ${summary.invalidFiles}`;
-  if (useColors) {
-    lines.push(`${ANSI_BOLD}${totalText}${ANSI_RESET2}`);
-    lines.push(`${ANSI_GREEN}${validText}${ANSI_RESET2}`);
-    if (summary.invalidFiles > 0) {
-      lines.push(`${ANSI_RED}${invalidText}${ANSI_RESET2}`);
-    } else {
-      lines.push(invalidText);
-    }
-  } else {
-    lines.push(totalText);
-    lines.push(validText);
-    lines.push(invalidText);
-  }
-  return lines.join("\n");
-}
-function isTTY() {
-  return process.stdout.isTTY ?? false;
-}
-
-// src/commands/lint/index.ts
-async function runLintCommand(paths, options) {
+// src/commands/validate/index.ts
+async function runValidateCommand(paths, _options) {
   if (paths.length === 0) {
-    console.error("Error: No paths specified. Usage: agentv lint <paths...>");
+    console.error("Error: No paths specified. Usage: agentv validate <paths...>");
     process.exit(1);
   }
-  const summary = await lintFiles(paths);
+  const summary = await validateFiles(paths);
   const useColors = isTTY();
   console.log(formatSummary(summary, useColors));
   if (summary.invalidFiles > 0) {
     process.exit(1);
   }
 }
-function registerLintCommand(program) {
-  program.command("lint").description("Validate AgentV eval and targets YAML files").argument("<paths...>", "Files or directories to lint").action(async (paths, options) => {
+function registerValidateCommand(program) {
+  program.command("validate").description("Validate AgentV eval and targets YAML files").argument("<paths...>", "Files or directories to validate").action(async (paths, _options) => {
     try {
-      await runLintCommand(paths, options);
+      await runValidateCommand(paths, _options);
     } catch (error) {
       console.error(`Error: ${error.message}`);
       process.exit(1);
@@ -14449,68 +14694,6 @@ function registerStatusCommand(program) {
   return program;
 }
 
-// src/commands/init/index.ts
-import { existsSync as existsSync2, mkdirSync, writeFileSync } from "node:fs";
-import path16 from "node:path";
-
-// src/templates/index.ts
-import { readFileSync } from "node:fs";
-import { fileURLToPath as fileURLToPath2 } from "node:url";
-import path15 from "node:path";
-var TemplateManager = class {
-  static getTemplates() {
-    const currentDir = path15.dirname(fileURLToPath2(import.meta.url));
-    let templatesDir;
-    if (currentDir.includes(path15.sep + "dist")) {
-      templatesDir = path15.join(currentDir, "templates");
-    } else {
-      templatesDir = currentDir;
-    }
-    const evalBuildPrompt = readFileSync(
-      path15.join(templatesDir, "eval-build.prompt.md"),
-      "utf-8"
-    );
-    const evalSchema = readFileSync(
-      path15.join(templatesDir, "eval-schema.json"),
-      "utf-8"
-    );
-    return [
-      {
-        path: "prompts/eval-build.prompt.md",
-        content: evalBuildPrompt
-      },
-      {
-        path: "contexts/eval-schema.json",
-        content: evalSchema
-      }
-    ];
-  }
-};
-
-// src/commands/init/index.ts
-async function initCommand(options = {}) {
-  const targetPath = path16.resolve(options.targetPath ?? ".");
-  const githubDir = path16.join(targetPath, ".github");
-  if (!existsSync2(githubDir)) {
-    mkdirSync(githubDir, { recursive: true });
-  }
-  const templates = TemplateManager.getTemplates();
-  for (const template of templates) {
-    const targetFilePath = path16.join(githubDir, template.path);
-    const targetDirPath = path16.dirname(targetFilePath);
-    if (!existsSync2(targetDirPath)) {
-      mkdirSync(targetDirPath, { recursive: true });
-    }
-    writeFileSync(targetFilePath, template.content, "utf-8");
-    console.log(`Created ${path16.relative(targetPath, targetFilePath)}`);
-  }
-  console.log("\nAgentV initialized successfully!");
-  console.log(`
-Files installed to ${path16.relative(targetPath, githubDir)}:`);
-  templates.forEach((t) => console.log(`  - ${t.path}`));
-  console.log("\nYou can now create eval files using the schema and prompt templates.");
-}
-
 // src/index.ts
 var packageJson = JSON.parse(readFileSync2(new URL("../package.json", import.meta.url), "utf8"));
 function createProgram() {
@@ -14518,7 +14701,7 @@ function createProgram() {
   program.name("agentv").description("AgentV CLI scaffolding").version(packageJson.version);
   registerStatusCommand(program);
   registerEvalCommand(program);
-  registerLintCommand(program);
+  registerValidateCommand(program);
   program.command("init [path]").description("Initialize AgentV in your project (installs prompt templates and schema to .github)").action(async (targetPath) => {
     try {
       await initCommand({ targetPath });
@@ -14539,4 +14722,4 @@ export {
   createProgram,
   runCli
 };
-//# sourceMappingURL=chunk-S3RN2GSO.js.map
+//# sourceMappingURL=chunk-RLBRJX7V.js.map