agentv 0.7.4 → 0.9.0

package/dist/{chunk-R2OCC2OH.js → chunk-X2VVUCIB.js}

@@ -590,7 +590,7 @@ import fg from "fast-glob";
 import { stat as stat3 } from "node:fs/promises";
 import path15 from "node:path";
 
-// ../../packages/core/dist/chunk-L6RCDZ4Z.js
+// ../../packages/core/dist/chunk-SNTZFB24.js
 import { constants } from "node:fs";
 import { access, readFile } from "node:fs/promises";
 import path from "node:path";
@@ -4636,7 +4636,7 @@ var coerce = {
 };
 var NEVER = INVALID;
 
-// ../../packages/core/dist/chunk-L6RCDZ4Z.js
+// ../../packages/core/dist/chunk-SNTZFB24.js
 async function fileExists(filePath) {
   try {
     await access(filePath, constants.F_OK);
@@ -4645,9 +4645,12 @@ async function fileExists(filePath) {
     return false;
   }
 }
+function normalizeLineEndings(content) {
+  return content.replace(/\r\n/g, "\n");
+}
 async function readTextFile(filePath) {
   const content = await readFile(filePath, "utf8");
-  return content.replace(/\r\n/g, "\n");
+  return normalizeLineEndings(content);
 }
 async function findGitRoot(startPath) {
   let currentDir = path.dirname(path.resolve(startPath));
@@ -4744,10 +4747,9 @@ var CLI_PLACEHOLDERS = /* @__PURE__ */ new Set(["PROMPT", "GUIDELINES", "EVAL_ID
 var BASE_TARGET_SCHEMA = external_exports.object({
   name: external_exports.string().min(1, "target name is required"),
   provider: external_exports.string().min(1, "provider is required"),
-  settings: external_exports.record(external_exports.unknown()).optional(),
   judge_target: external_exports.string().optional(),
   workers: external_exports.number().int().min(1).optional()
-});
+}).passthrough();
 var DEFAULT_AZURE_API_VERSION = "2024-10-01-preview";
 function normalizeAzureApiVersion(value) {
   if (!value) {
@@ -4760,11 +4762,43 @@ function normalizeAzureApiVersion(value) {
   const withoutPrefix = trimmed.replace(/^api[-_]?version\s*=\s*/i, "").trim();
   return withoutPrefix.length > 0 ? withoutPrefix : DEFAULT_AZURE_API_VERSION;
 }
+function resolveRetryConfig(target) {
+  const maxRetries = resolveOptionalNumber(
+    target.max_retries ?? target.maxRetries,
+    `${target.name} max retries`
+  );
+  const initialDelayMs = resolveOptionalNumber(
+    target.retry_initial_delay_ms ?? target.retryInitialDelayMs,
+    `${target.name} retry initial delay`
+  );
+  const maxDelayMs = resolveOptionalNumber(
+    target.retry_max_delay_ms ?? target.retryMaxDelayMs,
+    `${target.name} retry max delay`
+  );
+  const backoffFactor = resolveOptionalNumber(
+    target.retry_backoff_factor ?? target.retryBackoffFactor,
+    `${target.name} retry backoff factor`
+  );
+  const retryableStatusCodes = resolveOptionalNumberArray(
+    target.retry_status_codes ?? target.retryStatusCodes,
+    `${target.name} retry status codes`
+  );
+  if (maxRetries === void 0 && initialDelayMs === void 0 && maxDelayMs === void 0 && backoffFactor === void 0 && retryableStatusCodes === void 0) {
+    return void 0;
+  }
+  return {
+    maxRetries,
+    initialDelayMs,
+    maxDelayMs,
+    backoffFactor,
+    retryableStatusCodes
+  };
+}
 function resolveTargetDefinition(definition, env = process.env) {
   const parsed = BASE_TARGET_SCHEMA.parse(definition);
   const provider = parsed.provider.toLowerCase();
   const providerBatching = resolveOptionalBoolean(
-    parsed.settings?.provider_batching ?? parsed.settings?.providerBatching
+    parsed.provider_batching ?? parsed.providerBatching
   );
   switch (provider) {
     case "azure":
@@ -4840,13 +4874,12 @@ function resolveTargetDefinition(definition, env = process.env) {
   }
 }
 function resolveAzureConfig(target, env) {
-  const settings = target.settings ?? {};
-  const endpointSource = settings.endpoint ?? settings.resource ?? settings.resourceName;
-  const apiKeySource = settings.api_key ?? settings.apiKey;
-  const deploymentSource = settings.deployment ?? settings.deploymentName ?? settings.model;
-  const versionSource = settings.version ?? settings.api_version;
-  const temperatureSource = settings.temperature;
-  const maxTokensSource = settings.max_output_tokens ?? settings.maxTokens;
+  const endpointSource = target.endpoint ?? target.resource ?? target.resourceName;
+  const apiKeySource = target.api_key ?? target.apiKey;
+  const deploymentSource = target.deployment ?? target.deploymentName ?? target.model;
+  const versionSource = target.version ?? target.api_version;
+  const temperatureSource = target.temperature;
+  const maxTokensSource = target.max_output_tokens ?? target.maxTokens;
   const resourceName = resolveString(endpointSource, env, `${target.name} endpoint`);
   const apiKey = resolveString(apiKeySource, env, `${target.name} api key`);
   const deploymentName = resolveString(deploymentSource, env, `${target.name} deployment`);
@@ -4858,58 +4891,61 @@ function resolveAzureConfig(target, env) {
     maxTokensSource,
     `${target.name} max output tokens`
   );
+  const retry = resolveRetryConfig(target);
   return {
     resourceName,
     deploymentName,
     apiKey,
     version,
     temperature,
-    maxOutputTokens
+    maxOutputTokens,
+    retry
   };
 }
 function resolveAnthropicConfig(target, env) {
-  const settings = target.settings ?? {};
-  const apiKeySource = settings.api_key ?? settings.apiKey;
-  const modelSource = settings.model ?? settings.deployment ?? settings.variant;
-  const temperatureSource = settings.temperature;
-  const maxTokensSource = settings.max_output_tokens ?? settings.maxTokens;
-  const thinkingBudgetSource = settings.thinking_budget ?? settings.thinkingBudget;
+  const apiKeySource = target.api_key ?? target.apiKey;
+  const modelSource = target.model ?? target.deployment ?? target.variant;
+  const temperatureSource = target.temperature;
+  const maxTokensSource = target.max_output_tokens ?? target.maxTokens;
+  const thinkingBudgetSource = target.thinking_budget ?? target.thinkingBudget;
   const apiKey = resolveString(apiKeySource, env, `${target.name} Anthropic api key`);
   const model = resolveString(modelSource, env, `${target.name} Anthropic model`);
+  const retry = resolveRetryConfig(target);
   return {
     apiKey,
     model,
     temperature: resolveOptionalNumber(temperatureSource, `${target.name} temperature`),
     maxOutputTokens: resolveOptionalNumber(maxTokensSource, `${target.name} max output tokens`),
-    thinkingBudget: resolveOptionalNumber(thinkingBudgetSource, `${target.name} thinking budget`)
+    thinkingBudget: resolveOptionalNumber(thinkingBudgetSource, `${target.name} thinking budget`),
+    retry
   };
 }
 function resolveGeminiConfig(target, env) {
-  const settings = target.settings ?? {};
-  const apiKeySource = settings.api_key ?? settings.apiKey;
-  const modelSource = settings.model ?? settings.deployment ?? settings.variant;
-  const temperatureSource = settings.temperature;
-  const maxTokensSource = settings.max_output_tokens ?? settings.maxTokens;
+  const apiKeySource = target.api_key ?? target.apiKey;
+  const modelSource = target.model ?? target.deployment ?? target.variant;
+  const temperatureSource = target.temperature;
+  const maxTokensSource = target.max_output_tokens ?? target.maxTokens;
   const apiKey = resolveString(apiKeySource, env, `${target.name} Google API key`);
   const model = resolveOptionalString(modelSource, env, `${target.name} Gemini model`, {
     allowLiteral: true,
     optionalEnv: true
   }) ?? "gemini-2.5-flash";
+  const retry = resolveRetryConfig(target);
   return {
     apiKey,
     model,
     temperature: resolveOptionalNumber(temperatureSource, `${target.name} temperature`),
-    maxOutputTokens: resolveOptionalNumber(maxTokensSource, `${target.name} max output tokens`)
+    maxOutputTokens: resolveOptionalNumber(maxTokensSource, `${target.name} max output tokens`),
+    retry
   };
 }
 function resolveCodexConfig(target, env) {
-  const settings = target.settings ?? {};
-  const executableSource = settings.executable ?? settings.command ?? settings.binary;
-  const argsSource = settings.args ?? settings.arguments;
-  const cwdSource = settings.cwd;
-  const timeoutSource = settings.timeout_seconds ?? settings.timeoutSeconds;
-  const logDirSource = settings.log_dir ?? settings.logDir ?? settings.log_directory ?? settings.logDirectory;
-  const logFormatSource = settings.log_format ?? settings.logFormat ?? settings.log_output_format ?? settings.logOutputFormat ?? env.AGENTV_CODEX_LOG_FORMAT;
+  const executableSource = target.executable ?? target.command ?? target.binary;
+  const argsSource = target.args ?? target.arguments;
+  const cwdSource = target.cwd;
+  const timeoutSource = target.timeout_seconds ?? target.timeoutSeconds;
+  const logDirSource = target.log_dir ?? target.logDir ?? target.log_directory ?? target.logDirectory;
+  const logFormatSource = target.log_format ?? target.logFormat ?? target.log_output_format ?? target.logOutputFormat ?? env.AGENTV_CODEX_LOG_FORMAT;
   const executable = resolveOptionalString(executableSource, env, `${target.name} codex executable`, {
     allowLiteral: true,
     optionalEnv: true
@@ -4948,21 +4984,19 @@ function normalizeCodexLogFormat(value) {
   throw new Error("codex log format must be 'summary' or 'json'");
 }
 function resolveMockConfig(target) {
-  const settings = target.settings ?? {};
-  const response = typeof settings.response === "string" ? settings.response : void 0;
+  const response = typeof target.response === "string" ? target.response : void 0;
   return { response };
 }
 function resolveVSCodeConfig(target, env, insiders) {
-  const settings = target.settings ?? {};
-  const workspaceTemplateEnvVar = resolveOptionalLiteralString(settings.workspace_template ?? settings.workspaceTemplate);
+  const workspaceTemplateEnvVar = resolveOptionalLiteralString(target.workspace_template ?? target.workspaceTemplate);
   const workspaceTemplate = workspaceTemplateEnvVar ? resolveOptionalString(workspaceTemplateEnvVar, env, `${target.name} workspace template path`, {
     allowLiteral: false,
     optionalEnv: true
   }) : void 0;
-  const commandSource = settings.vscode_cmd ?? settings.command;
-  const waitSource = settings.wait;
-  const dryRunSource = settings.dry_run ?? settings.dryRun;
-  const subagentRootSource = settings.subagent_root ?? settings.subagentRoot;
+  const commandSource = target.vscode_cmd ?? target.command;
+  const waitSource = target.wait;
+  const dryRunSource = target.dry_run ?? target.dryRun;
+  const subagentRootSource = target.subagent_root ?? target.subagentRoot;
   const defaultCommand = insiders ? "code-insiders" : "code";
   const command = resolveOptionalLiteralString(commandSource) ?? defaultCommand;
   return {
@@ -4977,18 +5011,16 @@ function resolveVSCodeConfig(target, env, insiders) {
   };
 }
 function resolveCliConfig(target, env) {
-  const settings = target.settings ?? {};
-  const commandTemplateSource = settings.command_template ?? settings.commandTemplate;
+  const commandTemplateSource = target.command_template ?? target.commandTemplate;
   const filesFormat = resolveOptionalLiteralString(
-    settings.files_format ?? settings.filesFormat ?? settings.attachments_format ?? settings.attachmentsFormat
+    target.files_format ?? target.filesFormat ?? target.attachments_format ?? target.attachmentsFormat
   );
-  const cwd = resolveOptionalString(settings.cwd, env, `${target.name} working directory`, {
+  const cwd = resolveOptionalString(target.cwd, env, `${target.name} working directory`, {
     allowLiteral: true,
     optionalEnv: true
   });
-  const envOverrides = resolveEnvOverrides(settings.env, env, target.name);
-  const timeoutMs = resolveTimeoutMs(settings.timeout_seconds ?? settings.timeoutSeconds, `${target.name} timeout`);
-  const healthcheck = resolveCliHealthcheck(settings.healthcheck, env, target.name);
+  const timeoutMs = resolveTimeoutMs(target.timeout_seconds ?? target.timeoutSeconds, `${target.name} timeout`);
+  const healthcheck = resolveCliHealthcheck(target.healthcheck, env, target.name);
   const commandTemplate = resolveString(
     commandTemplateSource,
     env,
@@ -5000,29 +5032,10 @@ function resolveCliConfig(target, env) {
     commandTemplate,
     filesFormat,
     cwd,
-    env: envOverrides,
     timeoutMs,
     healthcheck
   };
 }
-function resolveEnvOverrides(source2, env, targetName) {
-  if (source2 === void 0 || source2 === null) {
-    return void 0;
-  }
-  if (typeof source2 !== "object" || Array.isArray(source2)) {
-    throw new Error(`${targetName} env overrides must be an object map of strings`);
-  }
-  const entries = Object.entries(source2);
-  const resolved = {};
-  for (const [key2, value] of entries) {
-    if (typeof value !== "string") {
-      throw new Error(`${targetName} env override '${key2}' must be a string`);
-    }
-    const resolvedValue = resolveString(value, env, `${targetName} env override '${key2}'`);
-    resolved[key2] = resolvedValue;
-  }
-  return Object.keys(resolved).length > 0 ? resolved : void 0;
-}
 function resolveTimeoutMs(source2, description) {
   const seconds = resolveOptionalNumber(source2, `${description} (seconds)`);
   if (seconds === void 0) {
@@ -5218,6 +5231,26 @@ function resolveOptionalStringArray(source2, env, description) {
   }
   return resolved.length > 0 ? resolved : void 0;
 }
+function resolveOptionalNumberArray(source2, description) {
+  if (source2 === void 0 || source2 === null) {
+    return void 0;
+  }
+  if (!Array.isArray(source2)) {
+    throw new Error(`${description} must be an array of numbers`);
+  }
+  if (source2.length === 0) {
+    return void 0;
+  }
+  const resolved = [];
+  for (let i6 = 0; i6 < source2.length; i6++) {
+    const item = source2[i6];
+    if (typeof item !== "number" || !Number.isFinite(item)) {
+      throw new Error(`${description}[${i6}] must be a number`);
+    }
+    resolved.push(item);
+  }
+  return resolved.length > 0 ? resolved : void 0;
+}
 var AGENT_PROVIDER_KINDS = [
   "codex",
   "vscode",
@@ -5249,7 +5282,7 @@ var PROVIDER_ALIASES = [
   "vertex"
   // legacy/future support
 ];
-var TARGETS_SCHEMA_V2 = "agentv-targets-v2.1";
+var TARGETS_SCHEMA_V2 = "agentv-targets-v2.2";
 function isAgentProvider(provider) {
   return provider ? AGENT_PROVIDER_KINDS.includes(provider.kind) : false;
 }
@@ -12305,6 +12338,67 @@ function ensureChatResponse(result) {
   }
   return result;
 }
+function isRetryableError(error, retryableStatusCodes) {
+  if (!error || typeof error !== "object") {
+    return false;
+  }
+  if ("status" in error && typeof error.status === "number") {
+    return retryableStatusCodes.includes(error.status);
+  }
+  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);
+    }
+  }
+  if ("name" in error && error.name === "AxAIServiceNetworkError") {
+    return true;
+  }
+  return false;
+}
+function calculateRetryDelay(attempt, config) {
+  const delay = Math.min(
+    config.maxDelayMs,
+    config.initialDelayMs * config.backoffFactor ** attempt
+  );
+  return delay * (0.75 + Math.random() * 0.5);
+}
+async function sleep2(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+async function withRetry(fn, retryConfig, signal) {
+  const config = {
+    maxRetries: retryConfig?.maxRetries ?? 3,
+    initialDelayMs: retryConfig?.initialDelayMs ?? 1e3,
+    maxDelayMs: retryConfig?.maxDelayMs ?? 6e4,
+    backoffFactor: retryConfig?.backoffFactor ?? 2,
+    retryableStatusCodes: retryConfig?.retryableStatusCodes ?? [500, 408, 429, 502, 503, 504]
+  };
+  let lastError;
+  for (let attempt = 0; attempt <= config.maxRetries; attempt++) {
+    if (signal?.aborted) {
+      throw new Error(`Request aborted: ${signal.reason ?? "Unknown reason"}`);
+    }
+    try {
+      return await fn();
+    } catch (error) {
+      lastError = error;
+      if (attempt >= config.maxRetries) {
+        break;
+      }
+      if (!isRetryableError(error, config.retryableStatusCodes)) {
+        throw error;
+      }
+      const delay = calculateRetryDelay(attempt, config);
+      await sleep2(delay);
+      if (signal?.aborted) {
+        throw new Error(`Request aborted: ${signal.reason ?? "Unknown reason"}`);
+      }
+    }
+  }
+  throw lastError;
+}
 var AzureProvider = class {
   constructor(targetName, config) {
     this.config = config;
@@ -12314,6 +12408,7 @@ var AzureProvider = class {
       temperature: config.temperature,
       maxOutputTokens: config.maxOutputTokens
     };
+    this.retryConfig = config.retry;
     this.ai = Wn.create({
       name: "azure-openai",
       apiKey: config.apiKey,
@@ -12330,16 +12425,21 @@ var AzureProvider = class {
   targetName;
   ai;
   defaults;
+  retryConfig;
   async invoke(request) {
     const chatPrompt = buildChatPrompt(request);
     const modelConfig = extractModelConfig(request, this.defaults);
-    const response = await this.ai.chat(
-      {
-        chatPrompt,
-        model: this.config.deploymentName,
-        ...modelConfig ? { modelConfig } : {}
-      },
-      request.signal ? { abortSignal: request.signal } : void 0
+    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));
   }
@@ -12357,6 +12457,7 @@ var AnthropicProvider = class {
       maxOutputTokens: config.maxOutputTokens,
       thinkingBudget: config.thinkingBudget
     };
+    this.retryConfig = config.retry;
     this.ai = Wn.create({
       name: "anthropic",
       apiKey: config.apiKey
@@ -12367,16 +12468,21 @@ var AnthropicProvider = class {
   targetName;
   ai;
   defaults;
+  retryConfig;
   async invoke(request) {
     const chatPrompt = buildChatPrompt(request);
     const modelConfig = extractModelConfig(request, this.defaults);
-    const response = await this.ai.chat(
-      {
-        chatPrompt,
-        model: this.config.model,
-        ...modelConfig ? { modelConfig } : {}
-      },
-      request.signal ? { abortSignal: request.signal } : void 0
+    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));
   }
@@ -12393,6 +12499,7 @@ var GeminiProvider = class {
       temperature: config.temperature,
       maxOutputTokens: config.maxOutputTokens
     };
+    this.retryConfig = config.retry;
     this.ai = Wn.create({
       name: "google-gemini",
       apiKey: config.apiKey
@@ -12403,16 +12510,21 @@ var GeminiProvider = class {
   targetName;
   ai;
   defaults;
+  retryConfig;
   async invoke(request) {
     const chatPrompt = buildChatPrompt(request);
     const modelConfig = extractModelConfig(request, this.defaults);
-    const response = await this.ai.chat(
-      {
-        chatPrompt,
-        model: this.config.model,
-        ...modelConfig ? { modelConfig } : {}
-      },
-      request.signal ? { abortSignal: request.signal } : void 0
+    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));
   }
@@ -12433,7 +12545,6 @@ async function defaultCommandRunner(command, options) {
   };
   try {
     const { stdout, stderr } = await execAsync2(command, execOptions);
-    console.error(`[CLI DEBUG] SUCCESS - stdout: ${stdout.length} bytes, stderr: ${stderr.length} bytes`);
     return {
       stdout,
       stderr,
@@ -12444,8 +12555,6 @@ async function defaultCommandRunner(command, options) {
     };
   } catch (error) {
     const execError = error;
-    console.error(`[CLI DEBUG] ERROR - code: ${execError.code}, message: ${execError.message}`);
-    console.error(`[CLI DEBUG] stdout: ${execError.stdout?.length ?? 0} bytes, stderr: ${execError.stderr?.length ?? 0} bytes`);
     return {
       stdout: execError.stdout ?? "",
       stderr: execError.stderr ?? "",
@@ -12478,10 +12587,9 @@ var CliProvider = class {
     const outputFilePath = generateOutputFilePath(request.evalCaseId);
     const templateValues = buildTemplateValues(request, this.config, outputFilePath);
     const renderedCommand = renderTemplate(this.config.commandTemplate, templateValues);
-    const env = this.config.env ? { ...process.env, ...this.config.env } : process.env;
     const result = await this.runCommand(renderedCommand, {
       cwd: this.config.cwd,
-      env,
+      env: process.env,
       timeoutMs: this.config.timeoutMs,
       signal: request.signal
     });
@@ -12513,7 +12621,7 @@ var CliProvider = class {
   }
   async readAndCleanupOutputFile(filePath) {
     try {
-      const content = await fs.readFile(filePath, "utf-8");
+      const content = await readTextFile(filePath);
       return content;
     } catch (error) {
       const errorMsg = error instanceof Error ? error.message : String(error);
@@ -12570,10 +12678,9 @@ var CliProvider = class {
         generateOutputFilePath("healthcheck")
       )
     );
-    const env = this.config.env ? { ...process.env, ...this.config.env } : process.env;
     const result = await this.runCommand(renderedCommand, {
       cwd: healthcheck.cwd ?? this.config.cwd,
-      env,
+      env: process.env,
       timeoutMs,
       signal
     });
@@ -13771,20 +13878,13 @@ function assertTargetDefinition(value, index, filePath) {
   }
   const name = value.name;
   const provider = value.provider;
-  const settings = value.settings;
-  const judgeTarget = value.judge_target;
   if (typeof name !== "string" || name.trim().length === 0) {
     throw new Error(`targets.yaml entry at index ${index} in ${filePath} is missing a valid 'name'`);
   }
   if (typeof provider !== "string" || provider.trim().length === 0) {
     throw new Error(`targets.yaml entry '${name}' in ${filePath} is missing a valid 'provider'`);
   }
-  return {
-    name,
-    provider,
-    settings: isRecord(settings) ? settings : void 0,
-    judge_target: typeof judgeTarget === "string" ? judgeTarget : void 0
-  };
+  return value;
 }
 async function fileExists3(filePath) {
   try {
@@ -14405,10 +14505,11 @@ async function runEvaluation(options) {
           await onProgress({
             workerId,
             evalId: evalCase.id,
-            status: "completed",
+            status: result.error ? "failed" : "completed",
             startedAt: 0,
             // Not used for completed status
-            completedAt: Date.now()
+            completedAt: Date.now(),
+            error: result.error
           });
         }
         if (onResult) {
@@ -14946,7 +15047,8 @@ function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs,
     target: targetName,
     timestamp: timestamp.toISOString(),
     raw_aspects: [],
-    raw_request: rawRequest
+    raw_request: rawRequest,
+    error: message
   };
 }
 function createCacheKey(provider, target, evalCase, promptInputs) {
@@ -15331,9 +15433,10 @@ var YamlWriter = class _YamlWriter {
         // Let YAML library choose appropriate string style based on content
         // (will use block literal for multiline strings with actual newlines)
       });
+      const normalizedYaml = normalizeLineEndings(yamlDoc);
       const separator = this.isFirst ? "---\n" : "\n---\n";
       this.isFirst = false;
-      const content = `${separator}${yamlDoc}`;
+      const content = `${separator}${normalizedYaml}`;
       if (!this.stream.write(content)) {
         await new Promise((resolve, reject) => {
           this.stream.once("drain", resolve);
@@ -15379,27 +15482,42 @@ function getDefaultExtension(format) {
 }
 
 // src/commands/eval/progress-display.ts
-import logUpdate from "log-update";
+import { stripVTControlCharacters } from "node:util";
+var ESC = "\x1B[";
+var CLEAR_LINE = `${ESC}K`;
+var MOVE_CURSOR_UP = `${ESC}1A`;
+var SYNC_START = `${ESC}?2026h`;
+var SYNC_END = `${ESC}?2026l`;
 var ProgressDisplay = class {
   workers = /* @__PURE__ */ new Map();
   maxWorkers;
   totalTests = 0;
   completedTests = 0;
   renderTimer;
+  renderScheduled = false;
   isInteractive;
   logPaths = [];
   logPathSet = /* @__PURE__ */ new Set();
   hasPrintedLogHeader = false;
+  windowHeight = 0;
+  started = false;
+  finished = false;
   constructor(maxWorkers) {
     this.maxWorkers = maxWorkers;
-    this.isInteractive = process.stderr.isTTY && !process.env.CI;
+    this.isInteractive = process.stdout.isTTY && !process.env.CI;
   }
   isInteractiveMode() {
     return this.isInteractive;
   }
   start() {
+    this.started = true;
+    this.finished = false;
     if (this.isInteractive) {
-      console.log("");
+      this.write("\n");
+      this.renderTimer = setInterval(() => {
+        this.scheduleRender();
+      }, 1e3);
+      this.renderTimer.unref?.();
     }
   }
   setTotalTests(count) {
@@ -15449,22 +15567,42 @@ var ProgressDisplay = class {
     });
   }
   scheduleRender() {
-    if (this.renderTimer) {
+    if (this.renderScheduled || this.finished) {
       return;
     }
-    this.renderTimer = setTimeout(() => {
-      this.renderTimer = void 0;
+    this.renderScheduled = true;
+    setTimeout(() => {
+      this.renderScheduled = false;
       this.render();
     }, 100);
   }
+  write(content) {
+    process.stdout.write(content);
+  }
+  clearWindow() {
+    if (this.windowHeight === 0) {
+      return;
+    }
+    this.write(`\r${CLEAR_LINE}`);
+    for (let i6 = 1; i6 < this.windowHeight; i6++) {
+      this.write(`${MOVE_CURSOR_UP}\r${CLEAR_LINE}`);
+    }
+    this.windowHeight = 0;
+  }
+  getRenderedRowCount(rows) {
+    const columns = process.stdout.columns || 80;
+    let count = 0;
+    for (const row of rows) {
+      const text = stripVTControlCharacters(row);
+      count += Math.max(1, Math.ceil(text.length / columns));
+    }
+    return count;
+  }
   render() {
-    if (!this.isInteractive) {
+    if (!this.isInteractive || !this.started || this.finished) {
       return;
     }
     const lines = [];
-    const progressBar = this.buildProgressBar(this.completedTests, this.totalTests);
-    lines.push(`${progressBar} ${this.completedTests}/${this.totalTests} evals`);
-    lines.push("");
     const sortedWorkers = Array.from(this.workers.values()).sort((a, b) => a.workerId - b.workerId);
     for (const worker of sortedWorkers) {
       const line2 = this.formatWorkerLine(worker);
@@ -15477,22 +15615,26 @@ var ProgressDisplay = class {
         lines.push(`${index + 1}. ${path19}`);
       });
     }
-    logUpdate(lines.join("\n"));
+    const rowCount = this.getRenderedRowCount(lines);
+    this.clearWindow();
+    if (lines.length > 0) {
+      this.write(lines.join("\n"));
+    }
+    this.windowHeight = rowCount;
   }
   formatWorkerLine(worker) {
     const workerLabel = `${worker.workerId}.`.padEnd(4);
     const statusIcon = this.getStatusIcon(worker.status);
-    const elapsed = worker.startedAt ? this.formatElapsed(Date.now() - worker.startedAt) : "";
-    const timeLabel = elapsed ? ` (${elapsed})` : "";
     const targetLabel = worker.targetLabel ? `  | ${worker.targetLabel}` : "";
-    const maxLineLength = 90;
-    const reservedLength = workerLabel.length + statusIcon.length + timeLabel.length + targetLabel.length + 4;
+    const columns = process.stdout.columns || 80;
+    const maxLineLength = Math.max(40, columns - 4);
+    const reservedLength = workerLabel.length + statusIcon.length + targetLabel.length + 4;
     const availableLabelLength = Math.max(15, maxLineLength - reservedLength);
     let testLabel = worker.evalId;
     if (testLabel.length > availableLabelLength) {
       testLabel = `${testLabel.substring(0, Math.max(0, availableLabelLength - 3))}...`;
     }
-    return `${workerLabel} ${statusIcon} ${testLabel}${timeLabel}${targetLabel}`;
+    return `${workerLabel} ${statusIcon} ${testLabel}${targetLabel}`;
   }
   getStatusIcon(status) {
     switch (status) {
@@ -15508,39 +15650,26 @@ var ProgressDisplay = class {
         return "  ";
     }
   }
-  formatElapsed(ms) {
-    const seconds = Math.floor(ms / 1e3);
-    if (seconds < 60) {
-      return `${seconds}s`;
-    }
-    const minutes = Math.floor(seconds / 60);
-    const remainingSeconds = seconds % 60;
-    return `${minutes}m ${remainingSeconds}s`;
-  }
-  buildProgressBar(current, total) {
-    if (total === 0) {
-      return "[          ]";
-    }
-    const width = 20;
-    const filled = Math.floor(current / total * width);
-    const empty = width - filled;
-    const bar = "\u2588".repeat(filled) + "\u2591".repeat(empty);
-    const percentage = Math.floor(current / total * 100);
-    return `[${bar}] ${percentage}%`;
-  }
   finish() {
     if (this.renderTimer) {
-      clearTimeout(this.renderTimer);
+      clearInterval(this.renderTimer);
       this.renderTimer = void 0;
     }
-    if (this.isInteractive) {
-      this.render();
-      logUpdate.done();
+    this.finished = true;
+    if (this.isInteractive && this.started) {
+      this.clearWindow();
+      const sortedWorkers = Array.from(this.workers.values()).sort(
+        (a, b) => a.workerId - b.workerId
+      );
+      for (const worker of sortedWorkers) {
+        this.write(this.formatWorkerLine(worker) + "\n");
+      }
+      this.write("\n");
     }
   }
   clear() {
     if (this.isInteractive) {
-      logUpdate.clear();
+      this.clearWindow();
     }
   }
 };
@@ -15597,6 +15726,8 @@ function buildHistogram(values) {
 function calculateEvaluationSummary(results) {
   const scores = results.map((result) => result.score);
   const total = results.length;
+  const errors = results.filter((result) => result.error !== void 0).map((result) => ({ evalId: result.eval_id, error: result.error }));
+  const errorCount = errors.length;
   if (total === 0) {
     return {
       total: 0,
@@ -15607,7 +15738,9 @@ function calculateEvaluationSummary(results) {
       standardDeviation: void 0,
       histogram: buildHistogram([]),
       topResults: [],
-      bottomResults: []
+      bottomResults: [],
+      errorCount: 0,
+      errors: []
     };
   }
   const mean = computeMean(scores);
@@ -15628,7 +15761,9 @@ function calculateEvaluationSummary(results) {
     standardDeviation,
     histogram,
     topResults,
-    bottomResults
+    bottomResults,
+    errorCount,
+    errors
   };
 }
 function formatScore(value) {
@@ -15639,10 +15774,25 @@ function formatEvaluationSummary(summary) {
     return "\nNo results to summarize";
   }
   const lines = [];
+  if (summary.errorCount > 0) {
+    lines.push("\n==================================================");
+    lines.push("ERRORS");
+    lines.push("==================================================");
+    summary.errors.forEach((error) => {
+      lines.push(`
+\u274C ${error.evalId}`);
+      lines.push(`   ${error.error}`);
+    });
+    lines.push("");
+  }
   lines.push("\n==================================================");
   lines.push("EVALUATION SUMMARY");
   lines.push("==================================================");
   lines.push(`Total eval cases: ${summary.total}`);
+  if (summary.errorCount > 0) {
+    lines.push(`Failed: ${summary.errorCount}`);
+    lines.push(`Passed: ${summary.total - summary.errorCount}`);
+  }
   lines.push(`Mean score: ${formatScore(summary.mean)}`);
   lines.push(`Median score: ${formatScore(summary.median)}`);
   lines.push(`Min score: ${formatScore(summary.min)}`);
@@ -15681,7 +15831,7 @@ import { readFile as readFile5 } from "node:fs/promises";
 import path33 from "node:path";
 import { parse as parse5 } from "yaml";
 var SCHEMA_EVAL_V22 = "agentv-eval-v2";
-var SCHEMA_TARGETS_V2 = "agentv-targets-v2.1";
+var SCHEMA_TARGETS_V2 = "agentv-targets-v2.2";
 var SCHEMA_CONFIG_V22 = "agentv-config-v2";
 async function detectFileType(filePath) {
   try {
@@ -15908,8 +16058,21 @@ var COMMON_SETTINGS = /* @__PURE__ */ new Set([
   "provider_batching",
   "providerBatching"
 ]);
+var RETRY_SETTINGS = /* @__PURE__ */ new Set([
+  "max_retries",
+  "maxRetries",
+  "retry_initial_delay_ms",
+  "retryInitialDelayMs",
+  "retry_max_delay_ms",
+  "retryMaxDelayMs",
+  "retry_backoff_factor",
+  "retryBackoffFactor",
+  "retry_status_codes",
+  "retryStatusCodes"
+]);
 var AZURE_SETTINGS = /* @__PURE__ */ new Set([
   ...COMMON_SETTINGS,
+  ...RETRY_SETTINGS,
   "endpoint",
   "resource",
   "resourceName",
@@ -15926,6 +16089,7 @@ var AZURE_SETTINGS = /* @__PURE__ */ new Set([
 ]);
 var ANTHROPIC_SETTINGS = /* @__PURE__ */ new Set([
   ...COMMON_SETTINGS,
+  ...RETRY_SETTINGS,
   "api_key",
   "apiKey",
   "model",
@@ -15939,6 +16103,7 @@ var ANTHROPIC_SETTINGS = /* @__PURE__ */ new Set([
 ]);
 var GEMINI_SETTINGS = /* @__PURE__ */ new Set([
   ...COMMON_SETTINGS,
+  ...RETRY_SETTINGS,
   "api_key",
   "apiKey",
   "model",
@@ -16026,13 +16191,14 @@ function getKnownSettings(provider) {
       return null;
   }
 }
-function validateUnknownSettings(settings, provider, absolutePath, location, errors) {
+function validateUnknownSettings(target, provider, absolutePath, location, errors) {
   const knownSettings = getKnownSettings(provider);
   if (!knownSettings) {
     return;
   }
-  for (const key2 of Object.keys(settings)) {
-    if (!knownSettings.has(key2)) {
+  const baseFields = /* @__PURE__ */ new Set(["name", "provider", "judge_target", "workers", "$schema", "targets"]);
+  for (const key2 of Object.keys(target)) {
+    if (!baseFields.has(key2) && !knownSettings.has(key2)) {
       errors.push({
         severity: "warning",
         filePath: absolutePath,
@@ -16062,17 +16228,8 @@ async function validateTargetsFile(filePath) {
       errors
     };
   }
-  function validateCliSettings(settings, absolutePath2, location, errors2) {
-    if (!isObject2(settings)) {
-      errors2.push({
-        severity: "error",
-        filePath: absolutePath2,
-        location,
-        message: "CLI provider requires a 'settings' object"
-      });
-      return;
-    }
-    const commandTemplate = settings["command_template"] ?? settings["commandTemplate"];
+  function validateCliSettings(target, absolutePath2, location, errors2) {
+    const commandTemplate = target["command_template"] ?? target["commandTemplate"];
     if (typeof commandTemplate !== "string" || commandTemplate.trim().length === 0) {
       errors2.push({
         severity: "error",
@@ -16083,7 +16240,7 @@ async function validateTargetsFile(filePath) {
     } else {
       recordUnknownPlaceholders(commandTemplate, absolutePath2, `${location}.commandTemplate`, errors2);
     }
-    const attachmentsFormat = settings["attachments_format"] ?? settings["attachmentsFormat"];
+    const attachmentsFormat = target["attachments_format"] ?? target["attachmentsFormat"];
     if (attachmentsFormat !== void 0 && typeof attachmentsFormat !== "string") {
       errors2.push({
         severity: "error",
@@ -16092,7 +16249,7 @@ async function validateTargetsFile(filePath) {
         message: "'attachmentsFormat' must be a string when provided"
       });
     }
-    const filesFormat = settings["files_format"] ?? settings["filesFormat"];
+    const filesFormat = target["files_format"] ?? target["filesFormat"];
     if (filesFormat !== void 0 && typeof filesFormat !== "string") {
       errors2.push({
         severity: "error",
@@ -16101,7 +16258,7 @@ async function validateTargetsFile(filePath) {
         message: "'filesFormat' must be a string when provided"
       });
     }
-    const cwd = settings["cwd"];
+    const cwd = target["cwd"];
     if (cwd !== void 0 && typeof cwd !== "string") {
       errors2.push({
         severity: "error",
@@ -16110,7 +16267,7 @@ async function validateTargetsFile(filePath) {
         message: "'cwd' must be a string when provided"
       });
     }
-    const timeoutSeconds = settings["timeout_seconds"] ?? settings["timeoutSeconds"];
+    const timeoutSeconds = target["timeout_seconds"] ?? target["timeoutSeconds"];
     if (timeoutSeconds !== void 0) {
       const numericTimeout = Number(timeoutSeconds);
       if (!Number.isFinite(numericTimeout) || numericTimeout <= 0) {
@@ -16122,29 +16279,7 @@ async function validateTargetsFile(filePath) {
         });
       }
     }
-    const envOverrides = settings["env"];
-    if (envOverrides !== void 0) {
-      if (!isObject2(envOverrides)) {
-        errors2.push({
-          severity: "error",
-          filePath: absolutePath2,
-          location: `${location}.env`,
-          message: "'env' must be an object with string values"
-        });
-      } else {
-        for (const [key2, value] of Object.entries(envOverrides)) {
-          if (typeof value !== "string" || value.trim().length === 0) {
-            errors2.push({
-              severity: "error",
-              filePath: absolutePath2,
-              location: `${location}.env.${key2}`,
-              message: `Environment override '${key2}' must be a non-empty string`
-            });
-          }
-        }
-      }
-    }
-    const healthcheck = settings["healthcheck"];
+    const healthcheck = target["healthcheck"];
     if (healthcheck !== void 0) {
       validateCliHealthcheck(healthcheck, absolutePath2, `${location}.healthcheck`, errors2);
     }
@@ -16315,20 +16450,11 @@ async function validateTargetsFile(filePath) {
         message: `Unknown provider '${provider}'. Known providers: ${knownProviders.join(", ")}`
       });
     }
-    const settings = target["settings"];
-    if (providerValue !== "cli" && settings !== void 0 && !isObject2(settings)) {
-      errors.push({
-        severity: "error",
-        filePath: absolutePath,
-        location: `${location}.settings`,
-        message: "Invalid 'settings' field (must be an object)"
-      });
-    }
     if (providerValue === "cli") {
-      validateCliSettings(settings, absolutePath, `${location}.settings`, errors);
+      validateCliSettings(target, absolutePath, location, errors);
     }
-    if (settings !== void 0 && isObject2(settings) && typeof provider === "string") {
-      validateUnknownSettings(settings, provider, absolutePath, `${location}.settings`, errors);
+    if (typeof provider === "string") {
+      validateUnknownSettings(target, provider, absolutePath, location, errors);
     }
     const judgeTarget = target["judge_target"];
     if (judgeTarget !== void 0 && typeof judgeTarget !== "string") {
@@ -17539,4 +17665,4 @@ export {
   createProgram,
   runCli
 };
-//# sourceMappingURL=chunk-R2OCC2OH.js.map
+//# sourceMappingURL=chunk-X2VVUCIB.js.map