blun-king-cli 9.1.596 → 9.1.597

package/blun.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-// BLUN_BUILD_INPUT_SHA256:2c4b3f451c41d328bb5eb43ffb226f0f3b0c458270217cecb6e1ffc63287e8bd
+// BLUN_BUILD_INPUT_SHA256:631215107f223f910482352d64f8081de5c7ebb282f9e66a53e7cd6dc93ba305
 import { fileURLToPath as __cjsShimFileURLToPath } from 'node:url';
 import { dirname as __cjsShimDirname } from 'node:path';
 const __filename = __cjsShimFileURLToPath(import.meta.url);
@@ -1536,6 +1536,7 @@ function parseExplicitMaxContextTokens(message) {
 	return Number.isSafeInteger(parsed) && parsed > 0 ? parsed : void 0;
 }
 function isRetryableGenerateError(error) {
+	if (error instanceof APIContextOverflowError) return false;
 	if (error instanceof CompactionStallError$1) return true;
 	if (error instanceof APIConnectionError || error instanceof APITimeoutError) return true;
 	if (error instanceof APIEmptyResponseError) return true;
@@ -1572,7 +1573,8 @@ function isProviderQuotaExhaustedMessage(message) {
 	return PROVIDER_QUOTA_EXHAUSTED_MESSAGE_PATTERN.test(lowerMessage) && PROVIDER_QUOTA_EXHAUSTED_CONTEXT_PATTERNS.some((pattern) => pattern.test(lowerMessage));
 }
 function isContextOverflowStatusError(statusCode, message) {
-	if (statusCode !== 400 && statusCode !== 413 && statusCode !== 422) return false;
+	const wrappedRejection = statusCode === 502 && /\bHTTP(?: Error)? (?:400|413|422):/i.test(message);
+	if (statusCode !== 400 && statusCode !== 413 && statusCode !== 422 && !wrappedRejection) return false;
 	const lowerMessage = message.toLowerCase();
 	return CONTEXT_OVERFLOW_MESSAGE_PATTERNS.some((pattern) => pattern.test(lowerMessage));
 }
@@ -1998,7 +2000,8 @@ var init_fetch_http_client = __esmMin((() => {
 				const startedAt = performance.now();
 				try {
 					const response = await this.fetch(`${this.baseUrl}${pathname}`, init);
-					const willRetry = attempt < maxRetries && shouldRetryResponse(response);
+					const gatewayError = response.status === 502 ? await responseError(response) : void 0;
+					const willRetry = attempt < maxRetries && shouldRetryResponse(response) && !(gatewayError instanceof APIContextOverflowError);
 					notifyTransportAttempt(requestOptions?.onTransportAttempt, {
 						transportAttempt: attempt + 1,
 						...requestBytes === void 0 ? {} : { requestBytes },
@@ -2010,12 +2013,12 @@ var init_fetch_http_client = __esmMin((() => {
 					if (response.ok) return response;
 					if (willRetry) {
 						const delayMs = retryDelayMs(response.headers, attempt);
-						await response.body?.cancel();
+						if (gatewayError === void 0) await response.body?.cancel();
 						await waitForRetry(delayMs, init.signal);
 						attempt += 1;
 						continue;
 					}
-					throw await responseError(response);
+					throw gatewayError ?? await responseError(response);
 				} catch (error) {
 					if (error instanceof ChatProviderError) throw error;
 					const willRetry = attempt < maxRetries && init.signal?.aborted !== true && isRetryableTransportFailure(error);
@@ -2578,9 +2581,23 @@ function extractUsageFromChunk(chunk) {
 	if (choiceUsage !== null && choiceUsage !== void 0 && typeof choiceUsage === "object") return choiceUsage;
 	return null;
 }
+function contextOverflowOutputCap(error, currentCap) {
+	const maxContext = error.maxContextTokens;
+	const counts = /requested ([\d,_]+) output tokens and your prompt contains (?:at least )?([\d,_]+) input tokens/i.exec(error.message);
+	if (maxContext === void 0 || counts === null || !Number.isSafeInteger(maxContext) || maxContext <= 0) return;
+	const requestedOutput = Number(counts[1].replaceAll(/[,_]/g, ""));
+	const input = Number(counts[2].replaceAll(/[,_]/g, ""));
+	if (!Number.isSafeInteger(input) || input < 0 || !Number.isSafeInteger(requestedOutput) || requestedOutput <= 0 || !Number.isSafeInteger(input + requestedOutput) || input + requestedOutput <= maxContext) return;
+	const available = maxContext - input;
+	const headroom = Math.min(1024, Math.max(1, Math.ceil(available * .01)));
+	const cap = Math.min(requestedOutput - 1, available - headroom);
+	if (cap < 1 || currentCap !== void 0 && (typeof currentCap !== "number" || !Number.isSafeInteger(currentCap) || cap >= currentCap)) return;
+	return cap;
+}
 var BLUN_TOOL_CALL_ID_POLICY, BLUN_INBOUND_REASONING_KEYS, BLUN_VISION_DATA_URL, BLUN_VISION_MAX_ATTACHMENTS, BLUN_VISION_MAX_IMAGE_BYTES, BlunStreamedMessage, BlunChatProvider;
 var init_blun = __esmMin((() => {
 	init_blun_schema();
+	init_errors$10();
 	init_blun_files();
 	init_chat_completions_wire();
 	init_chat_completions_stream();
@@ -2844,8 +2861,23 @@ var init_blun = __esmMin((() => {
 					...options.onTransportAttempt !== void 0 ? { onTransportAttempt: options.onTransportAttempt } : {}
 				} : void 0;
 				options?.onRequestSent?.();
-				return new BlunStreamedMessage(await client.chat.completions.create(createParams, requestOptions), this._stream, reasoningRequested);
+				let response;
+				try {
+					response = await client.chat.completions.create(createParams, requestOptions);
+				} catch (error) {
+					const rejection = normalizeFetchHttpError(error);
+					const cap = rejection instanceof APIContextOverflowError ? contextOverflowOutputCap(rejection, createParams["max_completion_tokens"]) : void 0;
+					if (cap === void 0) throw rejection;
+					options?.signal?.throwIfAborted();
+					options?.onRequestSent?.();
+					response = await client.chat.completions.create({
+						...createParams,
+						max_completion_tokens: cap
+					}, requestOptions);
+				}
+				return new BlunStreamedMessage(response, this._stream, reasoningRequested);
 			} catch (error) {
+				if (options?.signal?.aborted === true) throw options.signal.reason ?? error;
 				throw normalizeFetchHttpError(error);
 			}
 		}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "blun-king-cli",
-  "version": "9.1.596",
+  "version": "9.1.597",
   "description": "BLUN CLI - your own AI agent with a Telegram channel. Get it done. With BLUN.",
   "license": "MIT",
   "bin": {

package/worker-host.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-// BLUN_BUILD_INPUT_SHA256:2c4b3f451c41d328bb5eb43ffb226f0f3b0c458270217cecb6e1ffc63287e8bd
+// BLUN_BUILD_INPUT_SHA256:631215107f223f910482352d64f8081de5c7ebb282f9e66a53e7cd6dc93ba305
 import { fileURLToPath as __cjsShimFileURLToPath } from 'node:url';
 import { dirname as __cjsShimDirname } from 'node:path';
 const __filename = __cjsShimFileURLToPath(import.meta.url);
@@ -1664,6 +1664,7 @@ var CompactionStallError$1 = class extends ChatProviderError {
 	}
 };
 function isRetryableGenerateError(error) {
+	if (error instanceof APIContextOverflowError) return false;
 	if (error instanceof CompactionStallError$1) return true;
 	if (error instanceof APIConnectionError || error instanceof APITimeoutError) return true;
 	if (error instanceof APIEmptyResponseError) return true;
@@ -1728,7 +1729,8 @@ function isProviderQuotaExhaustedMessage(message) {
 	return PROVIDER_QUOTA_EXHAUSTED_MESSAGE_PATTERN.test(lowerMessage) && PROVIDER_QUOTA_EXHAUSTED_CONTEXT_PATTERNS.some((pattern) => pattern.test(lowerMessage));
 }
 function isContextOverflowStatusError(statusCode, message) {
-	if (statusCode !== 400 && statusCode !== 413 && statusCode !== 422) return false;
+	const wrappedRejection = statusCode === 502 && /\bHTTP(?: Error)? (?:400|413|422):/i.test(message);
+	if (statusCode !== 400 && statusCode !== 413 && statusCode !== 422 && !wrappedRejection) return false;
 	const lowerMessage = message.toLowerCase();
 	return CONTEXT_OVERFLOW_MESSAGE_PATTERNS.some((pattern) => pattern.test(lowerMessage));
 }
@@ -1854,7 +1856,8 @@ var FetchTransport = class {
 			const startedAt = performance.now();
 			try {
 				const response = await this.fetch(`${this.baseUrl}${pathname}`, init);
-				const willRetry = attempt < maxRetries && shouldRetryResponse(response);
+				const gatewayError = response.status === 502 ? await responseError(response) : void 0;
+				const willRetry = attempt < maxRetries && shouldRetryResponse(response) && !(gatewayError instanceof APIContextOverflowError);
 				notifyTransportAttempt(requestOptions?.onTransportAttempt, {
 					transportAttempt: attempt + 1,
 					...requestBytes === void 0 ? {} : { requestBytes },
@@ -1866,12 +1869,12 @@ var FetchTransport = class {
 				if (response.ok) return response;
 				if (willRetry) {
 					const delayMs = retryDelayMs(response.headers, attempt);
-					await response.body?.cancel();
+					if (gatewayError === void 0) await response.body?.cancel();
 					await waitForRetry(delayMs, init.signal);
 					attempt += 1;
 					continue;
 				}
-				throw await responseError(response);
+				throw gatewayError ?? await responseError(response);
 			} catch (error) {
 				if (error instanceof ChatProviderError) throw error;
 				const willRetry = attempt < maxRetries && init.signal?.aborted !== true && isRetryableTransportFailure(error);
@@ -2698,6 +2701,19 @@ var BlunStreamedMessage = class {
 		}
 	}
 };
+function contextOverflowOutputCap(error, currentCap) {
+	const maxContext = error.maxContextTokens;
+	const counts = /requested ([\d,_]+) output tokens and your prompt contains (?:at least )?([\d,_]+) input tokens/i.exec(error.message);
+	if (maxContext === void 0 || counts === null || !Number.isSafeInteger(maxContext) || maxContext <= 0) return;
+	const requestedOutput = Number(counts[1].replaceAll(/[,_]/g, ""));
+	const input = Number(counts[2].replaceAll(/[,_]/g, ""));
+	if (!Number.isSafeInteger(input) || input < 0 || !Number.isSafeInteger(requestedOutput) || requestedOutput <= 0 || !Number.isSafeInteger(input + requestedOutput) || input + requestedOutput <= maxContext) return;
+	const available = maxContext - input;
+	const headroom = Math.min(1024, Math.max(1, Math.ceil(available * .01)));
+	const cap = Math.min(requestedOutput - 1, available - headroom);
+	if (cap < 1 || currentCap !== void 0 && (typeof currentCap !== "number" || !Number.isSafeInteger(currentCap) || cap >= currentCap)) return;
+	return cap;
+}
 var BlunChatProvider = class {
 	name = "blun";
 	_model;
@@ -2852,8 +2868,23 @@ var BlunChatProvider = class {
 				...options.onTransportAttempt !== void 0 ? { onTransportAttempt: options.onTransportAttempt } : {}
 			} : void 0;
 			options?.onRequestSent?.();
-			return new BlunStreamedMessage(await client.chat.completions.create(createParams, requestOptions), this._stream, reasoningRequested);
+			let response;
+			try {
+				response = await client.chat.completions.create(createParams, requestOptions);
+			} catch (error) {
+				const rejection = normalizeFetchHttpError(error);
+				const cap = rejection instanceof APIContextOverflowError ? contextOverflowOutputCap(rejection, createParams["max_completion_tokens"]) : void 0;
+				if (cap === void 0) throw rejection;
+				options?.signal?.throwIfAborted();
+				options?.onRequestSent?.();
+				response = await client.chat.completions.create({
+					...createParams,
+					max_completion_tokens: cap
+				}, requestOptions);
+			}
+			return new BlunStreamedMessage(response, this._stream, reasoningRequested);
 		} catch (error) {
+			if (options?.signal?.aborted === true) throw options.signal.reason ?? error;
 			throw normalizeFetchHttpError(error);
 		}
 	}