@prestyj/agent 4.3.239 → 4.5.0

package/dist/index.js

@@ -5,7 +5,8 @@ import { EventStream as EventStream2 } from "@prestyj/ai";
 import { ZodError, prettifyError } from "zod";
 import {
   stream,
-  EventStream
+  EventStream,
+  isHardBillingMessage
 } from "@prestyj/ai";
 var DEFAULT_MAX_TURNS = 300;
 var _diagFn = null;
@@ -23,6 +24,8 @@ function isAbortError(err) {
 }
 function isContextOverflow(err) {
   if (!(err instanceof Error)) return false;
+  const overflowStatus = err.statusCode;
+  if (overflowStatus === 402) return false;
   if (isBillingError(err)) return false;
   const msg = err.message.toLowerCase();
   return msg.includes("prompt is too long") || msg.includes("prompt too long") || msg.includes("input is too long") || msg.includes("context_length_exceeded") || msg.includes("context_window_exceeded") || msg.includes("maximum context length") || msg.includes("exceeds model context window") || msg.includes("exceeds the context window") || msg.includes("content_too_large") || msg.includes("request_too_large") || msg.includes("reduce the length") || msg.includes("please shorten") || msg.includes("token") && msg.includes("exceed");
@@ -67,13 +70,21 @@ function extractContextOverflowDetails(err) {
 }
 function isBillingError(err) {
   if (!(err instanceof Error)) return false;
-  const msg = err.message.toLowerCase();
-  return msg.includes("insufficient balance") || msg.includes("no resource package") || msg.includes("quota exceeded") || msg.includes("billing") || msg.includes("recharge") || msg.includes("subscription plan") || msg.includes("does not yet include access") || msg.includes("token quota") || msg.includes("exceeded_current_quota_error") || msg.includes("check your account balance");
+  const statusCode = err.statusCode;
+  if (statusCode === 402) return true;
+  return isHardBillingMessage(err.message);
 }
 function isUsageLimitError(err) {
   if (!(err instanceof Error)) return false;
   return /usage limit reached/i.test(err.message);
 }
+function serverResetDelayMs(err) {
+  if (!(err instanceof Error)) return void 0;
+  const resetsAt = err.resetsAt;
+  if (typeof resetsAt !== "number" || !Number.isFinite(resetsAt)) return void 0;
+  const delayMs = resetsAt * 1e3 - Date.now();
+  return delayMs > 0 ? delayMs : void 0;
+}
 function isToolPairingError(err) {
   if (!(err instanceof Error)) return false;
   const msg = err.message.toLowerCase();
@@ -94,6 +105,7 @@ function classifyOverload(err) {
   const msg = err.message.toLowerCase();
   const errorWithStatus = err;
   const statusCode = typeof errorWithStatus.statusCode === "number" ? errorWithStatus.statusCode : void 0;
+  if (statusCode === 402) return null;
   if (statusCode === 429 || msg.includes("rate_limit") || msg.includes("rate limit") || msg.includes("too many requests") || msg.includes("429")) {
     return "rate_limit";
   }
@@ -153,23 +165,51 @@ function isTransportFailure(err) {
   }
   return false;
 }
+function createAbortError() {
+  return new DOMException("Aborted", "AbortError");
+}
+function abortablePromise(promise, signal) {
+  if (!signal) return promise;
+  if (signal.aborted) return Promise.reject(createAbortError());
+  return new Promise((resolve, reject) => {
+    let settled = false;
+    const cleanup = () => signal.removeEventListener("abort", onAbort);
+    const resolveOnce = (value) => {
+      if (settled) return;
+      settled = true;
+      cleanup();
+      resolve(value);
+    };
+    const rejectOnce = (err) => {
+      if (settled) return;
+      settled = true;
+      cleanup();
+      reject(err instanceof Error ? err : new Error(String(err)));
+    };
+    const onAbort = () => rejectOnce(createAbortError());
+    signal.addEventListener("abort", onAbort, { once: true });
+    promise.then(resolveOnce, rejectOnce);
+  });
+}
 function abortableSleep(ms, signal) {
-  if (signal?.aborted) {
-    return Promise.reject(new DOMException("Aborted", "AbortError"));
-  }
+  if (signal?.aborted) return Promise.reject(createAbortError());
   return new Promise((resolve, reject) => {
-    let onAbort = null;
     const timer = setTimeout(() => {
-      if (onAbort) signal?.removeEventListener("abort", onAbort);
+      signal?.removeEventListener("abort", onAbort);
       resolve();
     }, ms);
-    onAbort = () => {
+    const onAbort = () => {
       clearTimeout(timer);
-      reject(new DOMException("Aborted", "AbortError"));
+      reject(createAbortError());
     };
     signal?.addEventListener("abort", onAbort, { once: true });
   });
 }
+function closeIterator(iterator) {
+  if (!iterator?.return) return;
+  Promise.resolve(iterator.return()).catch(() => {
+  });
+}
 async function* agentLoop(messages, options) {
   const maxTurns = options.maxTurns ?? DEFAULT_MAX_TURNS;
   const maxContinuations = options.maxContinuations ?? 5;
@@ -294,6 +334,7 @@ async function* agentLoop(messages, options) {
         idleTimedOut = true;
         streamController.abort();
       }, hardTimeoutMs);
+      let streamIterator = null;
       try {
         diag("stream_call", { nonStreaming: useNonStreamingFallback });
         streamCallStart = Date.now();
@@ -316,9 +357,11 @@ async function* agentLoop(messages, options) {
           promptCacheKey: options.promptCacheKey,
           serviceTier: options.serviceTier,
           supportsImages: options.supportsImages,
+          supportsVideo: options.supportsVideo,
           compaction: options.compaction,
           clearToolUses: options.clearToolUses,
           userAgent: options.userAgent,
+          defaultHeaders: options.defaultHeaders,
           // Flip to non-streaming fallback after repeated stream stalls.
           ...useNonStreamingFallback ? { streaming: false } : {}
         });
@@ -331,7 +374,11 @@ async function* agentLoop(messages, options) {
         streamCallStart = Date.now();
         lastYieldEndTime = Date.now();
         resetIdleTimer();
-        for await (const event of result) {
+        streamIterator = result[Symbol.asyncIterator]();
+        while (true) {
+          const next = await abortablePromise(streamIterator.next(), streamController.signal);
+          if (next.done) break;
+          const event = next.value;
           const pullTime = Date.now();
           const consumerLag = pullTime - lastYieldEndTime;
           if (streamEventCount > 0 && consumerLag > maxConsumerLagMs) {
@@ -428,8 +475,9 @@ async function* agentLoop(messages, options) {
           maxConsumerLagMs,
           eventTypes: eventTypeCounts
         });
-        response = await result.response;
+        response = await abortablePromise(result.response, streamController.signal);
       } catch (err) {
+        if (streamController.signal.aborted) closeIterator(streamIterator);
         const errMsg = err instanceof Error ? err.message : String(err);
         diag("stream_error", {
           error: errMsg.slice(0, 200),
@@ -528,7 +576,8 @@ async function* agentLoop(messages, options) {
         const overloadKind = classifyOverload(err);
         if (overloadRetries < MAX_OVERLOAD_RETRIES && overloadKind) {
           overloadRetries++;
-          const delayMs = Math.min(
+          const serverDelayMs = serverResetDelayMs(err);
+          const delayMs = serverDelayMs !== void 0 ? Math.min(serverDelayMs, OVERLOAD_MAX_DELAY_MS) : Math.min(
             OVERLOAD_BASE_DELAY_MS * 2 ** (overloadRetries - 1),
             OVERLOAD_MAX_DELAY_MS
           );
@@ -828,7 +877,10 @@ async function executeSingleToolCall(toolCall, options, pushEvent) {
           });
         }
       };
-      const raw = await tool.execute(parsed, ctx);
+      const raw = await abortablePromise(
+        Promise.resolve().then(() => tool.execute(parsed, ctx)),
+        ctx.signal
+      );
       const normalized = normalizeToolResult(raw);
       resultContent = normalized.content;
       details = normalized.details;