@wix/evalforge-evaluator 0.153.0 → 0.155.0

package/build/index.js

@@ -56,6 +56,8 @@ function loadConfig() {
     }
   }
   aiGatewayHeaders["x-wix-ai-gateway-disable-cache"] = "true";
+  aiGatewayHeaders["x-time-budget"] = "300000";
+  aiGatewayHeaders["x-wix-time-budget"] = "300000";
   const tracePushUrl = process.env.TRACE_PUSH_URL;
   const routeHeader = process.env.EVAL_ROUTE_HEADER;
   const authToken = process.env.EVAL_AUTH_TOKEN;
@@ -3704,6 +3706,16 @@ async function executeWithAiSdk(context) {
   if (traceContext) {
     emitStartEvent(traceContext, startTime);
   }
+  const effectiveMaxTurns = cfg.maxTurns === 0 ? void 0 : cfg.maxTurns ?? DEFAULT_MAX_TOOL_STEPS;
+  const SDK_TIMEOUT_MS = cfg.maxDurationMs ?? Math.max(3e5, (effectiveMaxTurns ?? 25) * 6e4);
+  const abortController = new AbortController();
+  const timeoutHandle = setTimeout(() => {
+    abortController.abort(
+      new Error(
+        `Simple Agent generateText timed out after ${SDK_TIMEOUT_MS}ms (model=${modelId}, scenario=${scenario.name})`
+      )
+    );
+  }, SDK_TIMEOUT_MS);
   try {
     const isAnthropic = provider === PROVIDER_ANTHROPIC2;
     const isResponsesAPI = [...import_evalforge_types11.OPENAI_RESPONSES_MODEL_IDS].some(
@@ -3744,12 +3756,12 @@ async function executeWithAiSdk(context) {
       ...computedProviderOpts
     };
     const stepTimestamps = [];
-    const effectiveMaxTurns = cfg.maxTurns === 0 ? void 0 : cfg.maxTurns ?? DEFAULT_MAX_TOOL_STEPS;
     const { triggerPromptImages } = context;
     const hasImages = triggerPromptImages && triggerPromptImages.length > 0;
     const result = await (0, import_ai.generateText)({
       ...topLevelExtras,
       model,
+      abortSignal: abortController.signal,
       system: systemPrompt,
       ...hasImages ? {
         messages: [
@@ -3837,14 +3849,90 @@ async function executeWithAiSdk(context) {
     };
   } catch (err) {
     const baseMsg = err instanceof Error ? err.message : String(err);
+    const ctx = extractGatewayErrorContext(err);
     throw new Error(
-      `AI gateway request failed (provider=${provider}, model=${modelId}): ${baseMsg}`,
+      `AI gateway request failed (provider=${provider}, model=${modelId}): ${baseMsg}${ctx}`,
       { cause: err }
     );
   } finally {
+    clearTimeout(timeoutHandle);
     await closeMcpClients(clients);
   }
 }
+var UPSTREAM_REQUEST_ID_HEADERS = [
+  "openai-request-id",
+  "x-request-id",
+  "cf-ray",
+  "x-wix-request-id"
+];
+var PROXY_BODY_FIELDS = [
+  "proxy_request_id",
+  "upstream_request_id",
+  "failure_phase"
+];
+var MAX_ERROR_WALK_DEPTH = 5;
+function findApiCallError(err, depth = 0) {
+  if (err == null || depth > MAX_ERROR_WALK_DEPTH) return void 0;
+  if (import_ai.APICallError.isInstance(err)) return err;
+  if (typeof err === "object") {
+    const errors = err.errors;
+    if (Array.isArray(errors)) {
+      for (let i = errors.length - 1; i >= 0; i--) {
+        const found = findApiCallError(errors[i], depth + 1);
+        if (found) return found;
+      }
+    }
+    const cause = err.cause;
+    if (cause != null) return findApiCallError(cause, depth + 1);
+  }
+  return void 0;
+}
+function getHeader(headers, name) {
+  if (!headers) return void 0;
+  const target = name.toLowerCase();
+  for (const [key, value] of Object.entries(headers)) {
+    if (key.toLowerCase() === target) return value;
+  }
+  return void 0;
+}
+function getProxyBodyFields(responseBody) {
+  if (typeof responseBody !== "string" || responseBody.length === 0) {
+    return {};
+  }
+  let parsed;
+  try {
+    parsed = JSON.parse(responseBody);
+  } catch {
+    return {};
+  }
+  if (parsed == null || typeof parsed !== "object") return {};
+  const out = {};
+  for (const name of PROXY_BODY_FIELDS) {
+    const value = parsed[name];
+    if (typeof value === "string" && value.length > 0) {
+      out[name] = value;
+    }
+  }
+  return out;
+}
+function extractGatewayErrorContext(err) {
+  const apiError = findApiCallError(err);
+  if (!apiError) return "";
+  const fields = [];
+  if (typeof apiError.statusCode === "number") {
+    fields.push(`status=${apiError.statusCode}`);
+  }
+  for (const name of UPSTREAM_REQUEST_ID_HEADERS) {
+    const value = getHeader(apiError.responseHeaders, name);
+    if (value) fields.push(`${name}=${value}`);
+  }
+  const bodyFields = getProxyBodyFields(apiError.responseBody);
+  for (const name of PROXY_BODY_FIELDS) {
+    const value = bodyFields[name];
+    if (value) fields.push(`${name}=${value}`);
+  }
+  return fields.length > 0 ? ` [${fields.join(", ")}]` : "";
+}
 function composeSystemPrompt(context) {
   const parts = [];
   if (context.systemPrompt) {
@@ -4888,6 +4976,7 @@ async function runEvaluationLoop(scenarioItems, evalData, callbacks) {
   const runsPerScenario = evalData.evalRun.runsPerScenario ?? 1;
   let completedExecutions = 0;
   let erroredExecutions = 0;
+  let firstErrorMessage;
   const totalExecutions = scenarioItems.length * runsPerScenario;
   const targetId = evalData.evalRun.presetId ?? evalData.agent?.id ?? evalData.evalRun.id;
   const targetName = evalData.presetName ?? evalData.agent?.name ?? "";
@@ -4925,6 +5014,9 @@ async function runEvaluationLoop(scenarioItems, evalData, callbacks) {
         };
         await callbacks.addResult(errorResult);
         erroredExecutions++;
+        if (firstErrorMessage === void 0) {
+          firstErrorMessage = errorResult.outputText;
+        }
       }
       if (scenarioResult !== null) {
         await callbacks.addResult({ ...scenarioResult, iterationIndex });
@@ -4932,7 +5024,12 @@ async function runEvaluationLoop(scenarioItems, evalData, callbacks) {
       completedExecutions++;
     }
   }
-  return { completedExecutions, totalExecutions, erroredExecutions };
+  return {
+    completedExecutions,
+    totalExecutions,
+    erroredExecutions,
+    firstErrorMessage
+  };
 }
 
 // src/error-reporter.ts
@@ -5116,7 +5213,12 @@ async function runEvaluation(projectId2, evalRunId2) {
     );
   }
   const skillNames = evalData.skills.map((s) => s.name).join(", ");
-  const { completedExecutions, totalExecutions, erroredExecutions } = await runEvaluationLoop(scenarioItems, evalData, {
+  const {
+    completedExecutions,
+    totalExecutions,
+    erroredExecutions,
+    firstErrorMessage
+  } = await runEvaluationLoop(scenarioItems, evalData, {
     runScenario: (scenario, template, resolvedAssertions) => {
       state.currentPhase = ExecutionPhase.EXECUTE_SKILL;
       state.currentContext = {
@@ -5165,10 +5267,14 @@ async function runEvaluation(projectId2, evalRunId2) {
   };
   const allFailed = totalExecutions > 0 && erroredExecutions === totalExecutions;
   const finalStatus = allFailed ? import_evalforge_types15.EvalStatus.FAILED : import_evalforge_types15.EvalStatus.COMPLETED;
+  const jobErrorOnAllFailed = allFailed ? truncateForJobError(
+    firstErrorMessage ?? `All ${totalExecutions} executions errored without an error message`
+  ) : void 0;
   try {
     await api.updateEvalRun(projectId2, evalRunId2, {
       status: finalStatus,
-      completedAt: (/* @__PURE__ */ new Date()).toISOString()
+      completedAt: (/* @__PURE__ */ new Date()).toISOString(),
+      ...allFailed ? { jobError: jobErrorOnAllFailed, jobStatus: "FAILED" } : {}
     });
   } catch (updateErr) {
     throw new Error(
@@ -5176,6 +5282,11 @@ async function runEvaluation(projectId2, evalRunId2) {
     );
   }
 }
+var JOB_ERROR_MAX_LENGTH = 1e3;
+function truncateForJobError(message) {
+  if (message.length <= JOB_ERROR_MAX_LENGTH) return message;
+  return `${message.slice(0, JOB_ERROR_MAX_LENGTH)}\u2026 [truncated]`;
+}
 var projectId = process.argv[2];
 var evalRunId = process.argv[3];
 console.error(