@agentv/core 0.5.1 → 0.6.1

package/dist/index.cjs

@@ -36,6 +36,7 @@ __export(index_exports, {
   buildDirectoryChain: () => buildDirectoryChain,
   buildPromptInputs: () => buildPromptInputs,
   buildSearchRoots: () => buildSearchRoots,
+  consumeCodexLogEntries: () => consumeCodexLogEntries,
   createAgentKernel: () => createAgentKernel,
   createProvider: () => createProvider,
   ensureVSCodeSubagents: () => ensureVSCodeSubagents,
@@ -52,11 +53,13 @@ __export(index_exports, {
   listTargetNames: () => listTargetNames,
   loadEvalCases: () => loadEvalCases,
   readTargetDefinitions: () => readTargetDefinitions,
+  readTextFile: () => readTextFile,
   resolveAndCreateProvider: () => resolveAndCreateProvider,
   resolveFileReference: () => resolveFileReference,
   resolveTargetDefinition: () => resolveTargetDefinition,
   runEvalCase: () => runEvalCase,
-  runEvaluation: () => runEvaluation
+  runEvaluation: () => runEvaluation,
+  subscribeToCodexLogEntries: () => subscribeToCodexLogEntries
 });
 module.exports = __toCommonJS(index_exports);
 
@@ -130,6 +133,10 @@ async function fileExists(filePath) {
     return false;
   }
 }
+async function readTextFile(filePath) {
+  const content = await (0, import_promises.readFile)(filePath, "utf8");
+  return content.replace(/\r\n/g, "\n");
+}
 async function findGitRoot(startPath) {
   let currentDir = import_node_path.default.dirname(import_node_path.default.resolve(startPath));
   const root = import_node_path.default.parse(currentDir).root;
@@ -308,6 +315,9 @@ async function loadEvalCases(evalFilePath, repoRoot, options) {
     throw new Error(`Invalid test file format: ${evalFilePath}`);
   }
   const suite = parsed;
+  const datasetNameFromSuite = asString(suite.dataset)?.trim();
+  const fallbackDataset = import_node_path2.default.basename(absoluteTestPath).replace(/\.ya?ml$/i, "") || "eval";
+  const datasetName = datasetNameFromSuite && datasetNameFromSuite.length > 0 ? datasetNameFromSuite : fallbackDataset;
   const schema = suite.$schema;
   if (schema !== SCHEMA_EVAL_V2) {
     const message = typeof schema === "string" ? `Invalid $schema value '${schema}' in ${evalFilePath}. Expected '${SCHEMA_EVAL_V2}'` : `Missing required field '$schema' in ${evalFilePath}.
@@ -455,6 +465,7 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
     ];
     const testCase = {
       id,
+      dataset: datasetName,
       conversation_id: conversationId,
       task: userTextPrompt,
       user_segments: userSegments,
@@ -835,6 +846,9 @@ var AzureProvider = class {
     );
     return mapResponse(ensureChatResponse(response));
   }
+  getAxAI() {
+    return this.ai;
+  }
 };
 var AnthropicProvider = class {
   constructor(targetName, config) {
@@ -869,6 +883,9 @@ var AnthropicProvider = class {
     );
     return mapResponse(ensureChatResponse(response));
   }
+  getAxAI() {
+    return this.ai;
+  }
 };
 var GeminiProvider = class {
   constructor(targetName, config) {
@@ -902,6 +919,9 @@ var GeminiProvider = class {
     );
     return mapResponse(ensureChatResponse(response));
   }
+  getAxAI() {
+    return this.ai;
+  }
 };
 
 // src/evaluation/providers/cli.ts
@@ -1114,6 +1134,7 @@ function formatTimeoutSuffix(timeoutMs) {
 
 // src/evaluation/providers/codex.ts
 var import_node_child_process2 = require("child_process");
+var import_node_crypto = require("crypto");
 var import_node_fs3 = require("fs");
 var import_promises3 = require("fs/promises");
 var import_node_os = require("os");
@@ -1221,6 +1242,59 @@ function pathToFileUri(filePath) {
   return `file://${normalizedPath}`;
 }
 
+// src/evaluation/providers/codex-log-tracker.ts
+var GLOBAL_LOGS_KEY = Symbol.for("agentv.codexLogs");
+var GLOBAL_SUBSCRIBERS_KEY = Symbol.for("agentv.codexLogSubscribers");
+function getCodexLogStore() {
+  const globalObject = globalThis;
+  const existing = globalObject[GLOBAL_LOGS_KEY];
+  if (existing) {
+    return existing;
+  }
+  const created = [];
+  globalObject[GLOBAL_LOGS_KEY] = created;
+  return created;
+}
+function getSubscriberStore() {
+  const globalObject = globalThis;
+  const existing = globalObject[GLOBAL_SUBSCRIBERS_KEY];
+  if (existing) {
+    return existing;
+  }
+  const created = /* @__PURE__ */ new Set();
+  globalObject[GLOBAL_SUBSCRIBERS_KEY] = created;
+  return created;
+}
+function notifySubscribers(entry) {
+  const subscribers = Array.from(getSubscriberStore());
+  for (const listener of subscribers) {
+    try {
+      listener(entry);
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error);
+      console.warn(`Codex log subscriber failed: ${message}`);
+    }
+  }
+}
+function recordCodexLogEntry(entry) {
+  getCodexLogStore().push(entry);
+  notifySubscribers(entry);
+}
+function consumeCodexLogEntries() {
+  const store = getCodexLogStore();
+  if (store.length === 0) {
+    return [];
+  }
+  return store.splice(0, store.length);
+}
+function subscribeToCodexLogEntries(listener) {
+  const store = getSubscriberStore();
+  store.add(listener);
+  return () => {
+    store.delete(listener);
+  };
+}
+
 // src/evaluation/providers/codex.ts
 var execAsync2 = (0, import_node_util2.promisify)(import_node_child_process2.exec);
 var WORKSPACE_PREFIX = "agentv-codex-";
@@ -1252,6 +1326,7 @@ var CodexProvider = class {
       collectGuidelineFiles(inputFiles, request.guideline_patterns).map((file) => import_node_path5.default.resolve(file))
     );
     const workspaceRoot = await this.createWorkspace();
+    const logger = await this.createStreamLogger(request).catch(() => void 0);
     try {
       const { mirroredInputFiles, guidelineMirrors } = await this.mirrorInputFiles(
         inputFiles,
@@ -1266,7 +1341,7 @@ var CodexProvider = class {
       await (0, import_promises3.writeFile)(promptFile, promptContent, "utf8");
       const args = this.buildCodexArgs();
       const cwd = this.resolveCwd(workspaceRoot);
-      const result = await this.executeCodex(args, cwd, promptContent, request.signal);
+      const result = await this.executeCodex(args, cwd, promptContent, request.signal, logger);
       if (result.timedOut) {
         throw new Error(
           `Codex CLI timed out${formatTimeoutSuffix2(this.config.timeoutMs ?? void 0)}`
@@ -1290,10 +1365,12 @@ var CodexProvider = class {
           executable: this.resolvedExecutable ?? this.config.executable,
           promptFile,
           workspace: workspaceRoot,
-          inputFiles: mirroredInputFiles
+          inputFiles: mirroredInputFiles,
+          logFile: logger?.filePath
         }
       };
     } finally {
+      await logger?.close();
       await this.cleanupWorkspace(workspaceRoot);
     }
   }
@@ -1320,7 +1397,7 @@ var CodexProvider = class {
     args.push("-");
     return args;
   }
-  async executeCodex(args, cwd, promptContent, signal) {
+  async executeCodex(args, cwd, promptContent, signal, logger) {
     try {
       return await this.runCodex({
         executable: this.resolvedExecutable ?? this.config.executable,
@@ -1329,7 +1406,9 @@ var CodexProvider = class {
         prompt: promptContent,
         timeoutMs: this.config.timeoutMs,
         env: process.env,
-        signal
+        signal,
+        onStdoutChunk: logger ? (chunk) => logger.handleStdoutChunk(chunk) : void 0,
+        onStderrChunk: logger ? (chunk) => logger.handleStderrChunk(chunk) : void 0
       });
     } catch (error) {
       const err = error;
@@ -1381,7 +1460,240 @@ var CodexProvider = class {
     } catch {
     }
   }
+  resolveLogDirectory() {
+    const disabled = isCodexLogStreamingDisabled();
+    if (disabled) {
+      return void 0;
+    }
+    if (this.config.logDir) {
+      return import_node_path5.default.resolve(this.config.logDir);
+    }
+    return import_node_path5.default.join(process.cwd(), ".agentv", "logs", "codex");
+  }
+  async createStreamLogger(request) {
+    const logDir = this.resolveLogDirectory();
+    if (!logDir) {
+      return void 0;
+    }
+    try {
+      await (0, import_promises3.mkdir)(logDir, { recursive: true });
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error);
+      console.warn(`Skipping Codex stream logging (could not create ${logDir}): ${message}`);
+      return void 0;
+    }
+    const filePath = import_node_path5.default.join(logDir, buildLogFilename(request, this.targetName));
+    try {
+      const logger = await CodexStreamLogger.create({
+        filePath,
+        targetName: this.targetName,
+        evalCaseId: request.evalCaseId,
+        attempt: request.attempt,
+        format: this.config.logFormat ?? "summary"
+      });
+      recordCodexLogEntry({
+        filePath,
+        targetName: this.targetName,
+        evalCaseId: request.evalCaseId,
+        attempt: request.attempt
+      });
+      return logger;
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error);
+      console.warn(`Skipping Codex stream logging for ${filePath}: ${message}`);
+      return void 0;
+    }
+  }
+};
+var CodexStreamLogger = class _CodexStreamLogger {
+  filePath;
+  stream;
+  startedAt = Date.now();
+  stdoutBuffer = "";
+  stderrBuffer = "";
+  format;
+  constructor(filePath, format) {
+    this.filePath = filePath;
+    this.format = format;
+    this.stream = (0, import_node_fs3.createWriteStream)(filePath, { flags: "a" });
+  }
+  static async create(options) {
+    const logger = new _CodexStreamLogger(options.filePath, options.format);
+    const header = [
+      "# Codex CLI stream log",
+      `# target: ${options.targetName}`,
+      options.evalCaseId ? `# eval: ${options.evalCaseId}` : void 0,
+      options.attempt !== void 0 ? `# attempt: ${options.attempt + 1}` : void 0,
+      `# started: ${(/* @__PURE__ */ new Date()).toISOString()}`,
+      ""
+    ].filter((line) => Boolean(line));
+    logger.writeLines(header);
+    return logger;
+  }
+  handleStdoutChunk(chunk) {
+    this.stdoutBuffer += chunk;
+    this.flushBuffer("stdout");
+  }
+  handleStderrChunk(chunk) {
+    this.stderrBuffer += chunk;
+    this.flushBuffer("stderr");
+  }
+  async close() {
+    this.flushBuffer("stdout");
+    this.flushBuffer("stderr");
+    this.flushRemainder();
+    await new Promise((resolve, reject) => {
+      this.stream.once("error", reject);
+      this.stream.end(() => resolve());
+    });
+  }
+  writeLines(lines) {
+    for (const line of lines) {
+      this.stream.write(`${line}
+`);
+    }
+  }
+  flushBuffer(source) {
+    const buffer = source === "stdout" ? this.stdoutBuffer : this.stderrBuffer;
+    const lines = buffer.split(/\r?\n/);
+    const remainder = lines.pop() ?? "";
+    if (source === "stdout") {
+      this.stdoutBuffer = remainder;
+    } else {
+      this.stderrBuffer = remainder;
+    }
+    for (const line of lines) {
+      const formatted = this.formatLine(line, source);
+      if (formatted) {
+        this.stream.write(formatted);
+        this.stream.write("\n");
+      }
+    }
+  }
+  formatLine(rawLine, source) {
+    const trimmed = rawLine.trim();
+    if (trimmed.length === 0) {
+      return void 0;
+    }
+    const message = this.format === "json" ? formatCodexJsonLog(trimmed) : formatCodexLogMessage(trimmed, source);
+    return `[+${formatElapsed(this.startedAt)}] [${source}] ${message}`;
+  }
+  flushRemainder() {
+    const stdoutRemainder = this.stdoutBuffer.trim();
+    if (stdoutRemainder.length > 0) {
+      const formatted = this.formatLine(stdoutRemainder, "stdout");
+      if (formatted) {
+        this.stream.write(formatted);
+        this.stream.write("\n");
+      }
+    }
+    const stderrRemainder = this.stderrBuffer.trim();
+    if (stderrRemainder.length > 0) {
+      const formatted = this.formatLine(stderrRemainder, "stderr");
+      if (formatted) {
+        this.stream.write(formatted);
+        this.stream.write("\n");
+      }
+    }
+    this.stdoutBuffer = "";
+    this.stderrBuffer = "";
+  }
 };
+function isCodexLogStreamingDisabled() {
+  const envValue = process.env.AGENTV_CODEX_STREAM_LOGS;
+  if (!envValue) {
+    return false;
+  }
+  const normalized = envValue.trim().toLowerCase();
+  return normalized === "false" || normalized === "0" || normalized === "off";
+}
+function buildLogFilename(request, targetName) {
+  const timestamp = (/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-");
+  const evalId = sanitizeForFilename(request.evalCaseId ?? "codex");
+  const attemptSuffix = request.attempt !== void 0 ? `_attempt-${request.attempt + 1}` : "";
+  const target = sanitizeForFilename(targetName);
+  return `${timestamp}_${target}_${evalId}${attemptSuffix}_${(0, import_node_crypto.randomUUID)().slice(0, 8)}.log`;
+}
+function sanitizeForFilename(value) {
+  const sanitized = value.replace(/[^A-Za-z0-9._-]+/g, "_");
+  return sanitized.length > 0 ? sanitized : "codex";
+}
+function formatElapsed(startedAt) {
+  const elapsedSeconds = Math.floor((Date.now() - startedAt) / 1e3);
+  const hours = Math.floor(elapsedSeconds / 3600);
+  const minutes = Math.floor(elapsedSeconds % 3600 / 60);
+  const seconds = elapsedSeconds % 60;
+  if (hours > 0) {
+    return `${hours.toString().padStart(2, "0")}:${minutes.toString().padStart(2, "0")}:${seconds.toString().padStart(2, "0")}`;
+  }
+  return `${minutes.toString().padStart(2, "0")}:${seconds.toString().padStart(2, "0")}`;
+}
+function formatCodexLogMessage(rawLine, source) {
+  const parsed = tryParseJsonValue(rawLine);
+  if (parsed) {
+    const summary = summarizeCodexEvent(parsed);
+    if (summary) {
+      return summary;
+    }
+  }
+  if (source === "stderr") {
+    return `stderr: ${rawLine}`;
+  }
+  return rawLine;
+}
+function formatCodexJsonLog(rawLine) {
+  const parsed = tryParseJsonValue(rawLine);
+  if (!parsed) {
+    return rawLine;
+  }
+  try {
+    return JSON.stringify(parsed, null, 2);
+  } catch {
+    return rawLine;
+  }
+}
+function summarizeCodexEvent(event) {
+  if (!event || typeof event !== "object") {
+    return void 0;
+  }
+  const record = event;
+  const type = typeof record.type === "string" ? record.type : void 0;
+  let message = extractFromEvent(event) ?? extractFromItem(record.item) ?? flattenContent(record.output ?? record.content);
+  if (!message && type === JSONL_TYPE_ITEM_COMPLETED) {
+    const item = record.item;
+    if (item && typeof item === "object") {
+      const candidate = flattenContent(
+        item.text ?? item.content ?? item.output
+      );
+      if (candidate) {
+        message = candidate;
+      }
+    }
+  }
+  if (!message) {
+    const itemType = typeof record.item?.type === "string" ? record.item.type : void 0;
+    if (type && itemType) {
+      return `${type}:${itemType}`;
+    }
+    if (type) {
+      return type;
+    }
+  }
+  if (type && message) {
+    return `${type}: ${message}`;
+  }
+  if (message) {
+    return message;
+  }
+  return type;
+}
+function tryParseJsonValue(rawLine) {
+  try {
+    return JSON.parse(rawLine);
+  } catch {
+    return void 0;
+  }
+}
 async function locateExecutable(candidate) {
   const includesPathSeparator = candidate.includes("/") || candidate.includes("\\");
   if (includesPathSeparator) {
@@ -1651,10 +1963,12 @@ async function defaultCodexRunner(options) {
     child.stdout.setEncoding("utf8");
     child.stdout.on("data", (chunk) => {
       stdout += chunk;
+      options.onStdoutChunk?.(chunk);
     });
     child.stderr.setEncoding("utf8");
     child.stderr.on("data", (chunk) => {
       stderr += chunk;
+      options.onStderrChunk?.(chunk);
     });
     child.stdin.end(options.prompt);
     const cleanup = () => {
@@ -1899,6 +2213,8 @@ function resolveCodexConfig(target, env) {
   const argsSource = settings.args ?? settings.arguments;
   const cwdSource = settings.cwd;
   const timeoutSource = settings.timeout_seconds ?? settings.timeoutSeconds;
+  const logDirSource = settings.log_dir ?? settings.logDir ?? settings.log_directory ?? settings.logDirectory;
+  const logFormatSource = settings.log_format ?? settings.logFormat ?? settings.log_output_format ?? settings.logOutputFormat ?? env.AGENTV_CODEX_LOG_FORMAT;
   const executable = resolveOptionalString(executableSource, env, `${target.name} codex executable`, {
     allowLiteral: true,
     optionalEnv: true
@@ -1909,13 +2225,33 @@ function resolveCodexConfig(target, env) {
     optionalEnv: true
   });
   const timeoutMs = resolveTimeoutMs(timeoutSource, `${target.name} codex timeout`);
+  const logDir = resolveOptionalString(logDirSource, env, `${target.name} codex log directory`, {
+    allowLiteral: true,
+    optionalEnv: true
+  });
+  const logFormat = normalizeCodexLogFormat(logFormatSource);
   return {
     executable,
     args,
     cwd,
-    timeoutMs
+    timeoutMs,
+    logDir,
+    logFormat
   };
 }
+function normalizeCodexLogFormat(value) {
+  if (value === void 0 || value === null) {
+    return void 0;
+  }
+  if (typeof value !== "string") {
+    throw new Error("codex log format must be 'summary' or 'json'");
+  }
+  const normalized = value.trim().toLowerCase();
+  if (normalized === "json" || normalized === "summary") {
+    return normalized;
+  }
+  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;
@@ -2550,7 +2886,30 @@ function resolveAndCreateProvider(definition, env = process.env) {
 }
 
 // src/evaluation/evaluators.ts
-var import_node_crypto = require("crypto");
+var import_ax3 = require("@ax-llm/ax");
+var import_node_crypto2 = require("crypto");
+var LLM_JUDGE_SIGNATURE = (0, import_ax3.f)().input(
+  "evaluationContext",
+  import_ax3.f.object(
+    {
+      expectedOutcome: import_ax3.f.string("The expected outcome for the original task"),
+      request: import_ax3.f.string("The original task request"),
+      referenceAnswer: import_ax3.f.string("The gold standard reference answer"),
+      generatedAnswer: import_ax3.f.string("The answer to evaluate"),
+      guidelines: import_ax3.f.string("Additional evaluation guidelines or instructions").optional()
+    },
+    "Complete evaluation context for the judge"
+  )
+).output(
+  "evaluation",
+  import_ax3.f.object({
+    score: import_ax3.f.number("Score between 0.0 and 1.0").min(0).max(1),
+    hits: import_ax3.f.string("Brief specific achievement").array(),
+    misses: import_ax3.f.string("Brief specific failure or omission").array(),
+    reasoning: import_ax3.f.string("Concise explanation for the score").max(500)
+  })
+).build();
+var LLM_JUDGE = (0, import_ax3.ax)(LLM_JUDGE_SIGNATURE);
 var LlmJudgeEvaluator = class {
   kind = "llm_judge";
   resolveJudgeProvider;
@@ -2568,6 +2927,44 @@ var LlmJudgeEvaluator = class {
     if (!judgeProvider) {
       throw new Error("No judge provider available for LLM grading");
     }
+    if (providerSupportsAx(judgeProvider)) {
+      return this.evaluateWithAx(context, judgeProvider);
+    }
+    return this.evaluateWithPrompt(context, judgeProvider);
+  }
+  async evaluateWithAx(context, judgeProvider) {
+    const ai = judgeProvider.getAxAI();
+    const guidelines = context.promptInputs.guidelines?.trim();
+    const evaluationContext = {
+      expectedOutcome: context.evalCase.outcome.trim(),
+      request: context.evalCase.task.trim(),
+      referenceAnswer: context.evalCase.expected_assistant_raw.trim(),
+      generatedAnswer: context.candidate.trim(),
+      ...guidelines ? { guidelines } : {}
+    };
+    const options = this.buildJudgeForwardOptions(context);
+    const result = await LLM_JUDGE.forward(ai, { evaluationContext }, options);
+    const evaluation = result.evaluation;
+    const expectedAspectCount = Math.max(
+      evaluation.hits.length + evaluation.misses.length,
+      1
+    );
+    return {
+      score: evaluation.score,
+      hits: evaluation.hits,
+      misses: evaluation.misses,
+      expectedAspectCount,
+      reasoning: evaluation.reasoning,
+      evaluatorRawRequest: {
+        id: (0, import_node_crypto2.randomUUID)(),
+        provider: judgeProvider.id,
+        target: context.target.name,
+        method: "ax-structured-output",
+        signature: LLM_JUDGE_SIGNATURE.toString()
+      }
+    };
+  }
+  async evaluateWithPrompt(context, judgeProvider) {
     const prompt = buildQualityPrompt(context.evalCase, context.candidate);
     const systemPrompt = context.systemPrompt ?? this.customPrompt ?? QUALITY_SYSTEM_PROMPT;
     const metadata = {
@@ -2587,8 +2984,9 @@ var LlmJudgeEvaluator = class {
     const hits = Array.isArray(parsed.hits) ? parsed.hits.filter(isNonEmptyString).slice(0, 4) : [];
     const misses = Array.isArray(parsed.misses) ? parsed.misses.filter(isNonEmptyString).slice(0, 4) : [];
     const reasoning = parsed.reasoning ?? response.reasoning;
+    const expectedAspectCount = Math.max(hits.length + misses.length, 1);
     const evaluatorRawRequest = {
-      id: (0, import_node_crypto.randomUUID)(),
+      id: (0, import_node_crypto2.randomUUID)(),
       provider: judgeProvider.id,
       prompt,
       target: context.target.name,
@@ -2599,12 +2997,34 @@ var LlmJudgeEvaluator = class {
       score,
       hits,
       misses,
-      expectedAspectCount: hits.length + misses.length || 1,
+      expectedAspectCount,
       reasoning,
       evaluatorRawRequest
     };
   }
+  buildJudgeForwardOptions(context) {
+    const modelConfig = this.buildJudgeModelConfig();
+    if (modelConfig === void 0 && context.judgeModel === void 0) {
+      return void 0;
+    }
+    return {
+      ...context.judgeModel ? { model: context.judgeModel } : {},
+      ...modelConfig ? { modelConfig } : {}
+    };
+  }
+  buildJudgeModelConfig() {
+    if (this.maxOutputTokens === void 0 && this.temperature === void 0) {
+      return void 0;
+    }
+    return {
+      ...this.maxOutputTokens !== void 0 ? { maxTokens: this.maxOutputTokens } : {},
+      ...this.temperature !== void 0 ? { temperature: this.temperature } : {}
+    };
+  }
 };
+function providerSupportsAx(provider) {
+  return typeof provider.getAxAI === "function";
+}
 var QUALITY_SYSTEM_PROMPT = [
   "You are an expert evaluator. Your goal is to grade the generated_answer based on how well it achieves the expected_outcome for the original task.",
   "",
@@ -2827,7 +3247,7 @@ function parseJsonSafe(payload) {
 }
 
 // src/evaluation/orchestrator.ts
-var import_node_crypto2 = require("crypto");
+var import_node_crypto3 = require("crypto");
 var import_promises6 = require("fs/promises");
 var import_node_path8 = __toESM(require("path"), 1);
 
@@ -3375,6 +3795,7 @@ async function evaluateCandidate(options) {
   };
   return {
     eval_id: evalCase.id,
+    dataset: evalCase.dataset,
     conversation_id: evalCase.conversation_id,
     score: score.score,
     hits: score.hits,
@@ -3551,7 +3972,7 @@ async function runLlmJudgeEvaluator(options) {
 async function resolveCustomPrompt(config) {
   if (config.promptPath) {
     try {
-      return await (0, import_promises6.readFile)(config.promptPath, "utf8");
+      return await readTextFile(config.promptPath);
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
       console.warn(`Could not read custom prompt at ${config.promptPath}: ${message}`);
@@ -3600,7 +4021,7 @@ function sanitizeFilename(value) {
     return "prompt";
   }
   const sanitized = value.replace(/[^A-Za-z0-9._-]+/g, "_");
-  return sanitized.length > 0 ? sanitized : (0, import_node_crypto2.randomUUID)();
+  return sanitized.length > 0 ? sanitized : (0, import_node_crypto3.randomUUID)();
 }
 async function invokeProvider(provider, options) {
   const { evalCase, promptInputs, attempt, agentTimeoutMs, signal } = options;
@@ -3639,6 +4060,7 @@ function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs)
   };
   return {
     eval_id: evalCase.id,
+    dataset: evalCase.dataset,
     conversation_id: evalCase.conversation_id,
     score: 0,
     hits: [],
@@ -3652,7 +4074,7 @@ function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs)
   };
 }
 function createCacheKey(provider, target, evalCase, promptInputs) {
-  const hash = (0, import_node_crypto2.createHash)("sha256");
+  const hash = (0, import_node_crypto3.createHash)("sha256");
   hash.update(provider.id);
   hash.update(target.name);
   hash.update(evalCase.id);
@@ -3689,6 +4111,7 @@ function createAgentKernel() {
   buildDirectoryChain,
   buildPromptInputs,
   buildSearchRoots,
+  consumeCodexLogEntries,
   createAgentKernel,
   createProvider,
   ensureVSCodeSubagents,
@@ -3705,10 +4128,12 @@ function createAgentKernel() {
   listTargetNames,
   loadEvalCases,
   readTargetDefinitions,
+  readTextFile,
   resolveAndCreateProvider,
   resolveFileReference,
   resolveTargetDefinition,
   runEvalCase,
-  runEvaluation
+  runEvaluation,
+  subscribeToCodexLogEntries
 });
 //# sourceMappingURL=index.cjs.map