@openclaw/ai 2026.7.2-beta.7 → 2026.8.1-beta.1

package/dist/{azure-openai-responses-CImcwB83.mjs → azure-openai-responses-Cc_kasye.mjs}

@@ -1,10 +1,10 @@
 import { n as getEnvApiKey } from "./env-api-keys-DrgeBuva.mjs";
-import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
-import { n as getAiTransportHost } from "./host-Dog2WQiR.mjs";
+import { t as AssistantMessageEventStream } from "./event-stream-CEV9t6da.mjs";
+import { n as getAiTransportHost } from "./host-Bl7Kgddo.mjs";
 import { n as clampOpenAIPromptCacheKey } from "./openai-prompt-cache-mZTCdRPo.mjs";
+import { c as buildBaseOptions } from "./tool-result-text-Dvkp2Dus.mjs";
 import { i as resolveAzureDeploymentNameFromMap, t as isOpenAICompatibleAzureResponsesBaseUrl } from "./azure-openai-responses-client-compat-C7K7QfUE.mjs";
-import { n as buildBaseOptions } from "./simple-options-9lhRrN73.mjs";
-import { a as runResponsesStreamLifecycle, i as resolveResponsesReasoningEffort, n as convertResponsesMessages, r as createResponsesAssistantOutput, t as applyCommonResponsesParams } from "./openai-responses-shared-pXl6Wd8S.mjs";
+import { a as resolveResponsesReasoningEffort, i as createResponsesAssistantOutput, o as runResponsesStreamLifecycle, r as convertResponsesMessages, t as applyCommonResponsesParams } from "./openai-responses-shared-CCyMvg7M.mjs";
 import OpenAI, { AzureOpenAI } from "openai";
 //#region packages/ai/src/providers/azure-openai-responses.ts
 const DEFAULT_AZURE_API_VERSION = "v1";

package/dist/{event-stream-D8n2uFee.mjs → event-stream-CEV9t6da.mjs}

@@ -3,6 +3,7 @@
 var EventStream = class {
 	constructor(isComplete, extractResult) {
 		this.queue = [];
+		this.queueHead = 0;
 		this.waiting = [];
 		this.done = false;
 		this.isComplete = isComplete;
@@ -41,8 +42,15 @@ var EventStream = class {
 		}
 	}
 	async *[Symbol.asyncIterator]() {
-		while (true) if (this.queue.length > 0) for (const event of this.queue.splice(0, 1)) yield event;
-		else if (this.done) return;
+		while (true) if (this.queueHead < this.queue.length) {
+			const event = this.queue[this.queueHead];
+			this.queueHead += 1;
+			if (this.queueHead >= 1024 && this.queueHead * 2 >= this.queue.length) {
+				this.queue = this.queue.slice(this.queueHead);
+				this.queueHead = 0;
+			}
+			yield event;
+		} else if (this.done) return;
 		else {
 			const result = await new Promise((resolve) => {
 				this.waiting.push(resolve);

package/dist/{event-stream-YjaPW20U.d.mts → event-stream-DRmDMRHH.d.mts}

@@ -1,9 +1,10 @@
-import { a as AssistantMessageEvent, i as AssistantMessage, o as AssistantMessageEventStreamContract } from "./types-bzp5k29J.mjs";
+import { a as AssistantMessageEvent, i as AssistantMessage, o as AssistantMessageEventStreamContract } from "./types-CH7ReIcU.mjs";
 
 //#region packages/llm-core/src/utils/event-stream.d.ts
 /** Generic async-iterable event stream with a separately awaited final result. */
 declare class EventStream<T, R = T> implements AsyncIterable<T> {
   private queue;
+  private queueHead;
   private waiting;
   private done;
   private finalResultPromise;

package/dist/event-stream.d.mts

@@ -1,2 +1,2 @@
-import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-YjaPW20U.mjs";
+import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-DRmDMRHH.mjs";
 export { AssistantMessageEventStream, EventStream, createAssistantMessageEventStream };

package/dist/event-stream.mjs

@@ -1,2 +1,2 @@
-import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
+import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-CEV9t6da.mjs";
 export { AssistantMessageEventStream, EventStream, createAssistantMessageEventStream };

package/dist/{google-CtSg0iTS.mjs → google-CGQu21aw.mjs}

@@ -1,8 +1,8 @@
 import { n as getEnvApiKey } from "./env-api-keys-DrgeBuva.mjs";
-import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
-import { n as getAiTransportHost, r as resolveAiTransportHeaderSentinels } from "./host-Dog2WQiR.mjs";
-import { n as buildBaseOptions } from "./simple-options-9lhRrN73.mjs";
-import { a as runGoogleGenerateContentLifecycle, i as getDisabledGoogleThinkingConfig, n as buildGoogleSimpleThinking, r as createGoogleAssistantOutput, t as buildGoogleGenerateContentParams } from "./google-shared-DNBz5rcD.mjs";
+import { t as AssistantMessageEventStream } from "./event-stream-CEV9t6da.mjs";
+import { n as getAiTransportHost, r as resolveAiTransportHeaderSentinels } from "./host-Bl7Kgddo.mjs";
+import { c as buildBaseOptions } from "./tool-result-text-Dvkp2Dus.mjs";
+import { i as runGoogleGenerateContentLifecycle, n as buildGoogleSimpleThinking, r as createGoogleAssistantOutput, t as buildGoogleGenerateContentParams } from "./google-shared-Cq2UKx1l.mjs";
 import { GoogleGenAI } from "@google/genai";
 //#region packages/ai/src/providers/google.ts
 let toolCallCounter = 0;
@@ -49,7 +49,7 @@ function createClient(model, apiKey, optionsHeaders) {
 	});
 }
 function buildParams(model, context, options = {}) {
-	return buildGoogleGenerateContentParams(model, context, options, { getDisabledThinkingConfig: (modelLocal) => getDisabledGoogleThinkingConfig(modelLocal, { includeGemma4: true }) });
+	return buildGoogleGenerateContentParams(model, context, options);
 }
 //#endregion
 export { streamGoogle, streamSimpleGoogle };

package/dist/{google-shared-DNBz5rcD.mjs → google-shared-Cq2UKx1l.mjs}

@@ -1,10 +1,10 @@
-import { i as transformMessages } from "./host-Dog2WQiR.mjs";
+import { i as transformMessages } from "./host-Bl7Kgddo.mjs";
+import { n as calculateCost, r as clampThinkingLevel } from "./model-utils-Dau5dlgm.mjs";
 import { t as sanitizeSurrogates } from "./sanitize-unicode-DT5o51ur.mjs";
-import { a as isImageWithMediaPayload, d as stripSystemPromptCacheBoundary, r as extractToolResultText, t as describeToolResultMediaPlaceholder } from "./tool-result-text-CTpIRbYd.mjs";
-import { c as calculateCost, l as clampThinkingLevel } from "./number-coercion-DvG7SNMg.mjs";
-import { d as transportAbortError } from "./transport-stream-shared-D81p90xq.mjs";
-import { t as formatProviderError } from "./provider-error-CAEvRjry.mjs";
-import "./transform-messages-C8mBqZxF.mjs";
+import { a as isImageWithMediaPayload, h as stripSystemPromptCacheBoundary, r as extractToolResultText, t as describeToolResultMediaPlaceholder, v as sortPromptCacheToolsByName } from "./tool-result-text-Dvkp2Dus.mjs";
+import { d as transportAbortError, n as coerceTransportToolCallArguments, t as assignTransportErrorDetails } from "./transport-stream-shared-CPNv7A3r.mjs";
+import { t as formatProviderError } from "./provider-error-DI0Ts28U.mjs";
+import "./transform-messages-DfjpNXNQ.mjs";
 import { FinishReason, FunctionCallingConfigMode, ThinkingLevel } from "@google/genai";
 //#region packages/ai/src/providers/google-shared.ts
 /**
@@ -83,6 +83,7 @@ function convertMessages(model, context) {
 		return id.replace(/[^a-zA-Z0-9_-]/g, "_").slice(0, 64);
 	};
 	const transformedMessages = transformMessages(context.messages, model, normalizeToolCallId);
+	const requiresToolCallThoughtSignature = model.provider !== "google-gemini-cli" && (isGemini3ProModel(model) || isGemini3FlashModel(model));
 	const pendingToolResultImageTurns = [];
 	let activeToolResultParts;
 	const flushToolResultRun = () => {
@@ -94,17 +95,17 @@ function convertMessages(model, context) {
 		if (msg.role !== "toolResult") flushToolResultRun();
 		if (msg.role === "user") if (typeof msg.content === "string") contents.push({
 			role: "user",
-			parts: [{ text: sanitizeSurrogates(msg.content) }]
+			parts: [{ text: sanitizeSurrogates(msg.content) || " " }]
 		});
 		else {
 			const parts = msg.content.map((item) => {
-				if (item.type === "text") return { text: sanitizeSurrogates(item.text) };
+				if (item.type === "text") return { text: sanitizeSurrogates(item.text) || " " };
 				return { inlineData: {
 					mimeType: item.mimeType,
 					data: item.data
 				} };
 			});
-			if (parts.length === 0) continue;
+			if (parts.length === 0) parts.push({ text: " " });
 			contents.push({
 				role: "user",
 				parts
@@ -112,30 +113,32 @@ function convertMessages(model, context) {
 		}
 		else if (msg.role === "assistant") {
 			const parts = [];
-			const isSameProviderAndModel = msg.provider === model.provider && msg.model === model.id;
+			let sawFunctionCall = false;
+			const isSameProviderAndModel = msg.provider === model.provider && msg.api === model.api && msg.model === model.id;
 			for (const block of msg.content) if (block.type === "text") {
-				if (!block.text || block.text.trim() === "") continue;
 				const thoughtSignature = resolveThoughtSignature(isSameProviderAndModel, block.textSignature);
+				if ((!block.text || block.text.trim() === "") && !thoughtSignature) continue;
 				parts.push({
 					text: sanitizeSurrogates(block.text),
 					...thoughtSignature && { thoughtSignature }
 				});
 			} else if (block.type === "thinking") {
-				if (!block.thinking || block.thinking.trim() === "") continue;
-				if (isSameProviderAndModel) {
-					const thoughtSignature = resolveThoughtSignature(isSameProviderAndModel, block.thinkingSignature);
-					parts.push({
-						thought: true,
-						text: sanitizeSurrogates(block.thinking),
-						...thoughtSignature && { thoughtSignature }
-					});
-				} else parts.push({ text: sanitizeSurrogates(block.thinking) });
+				const thoughtSignature = resolveThoughtSignature(isSameProviderAndModel, block.thinkingSignature);
+				if ((!block.thinking || block.thinking.trim() === "") && !thoughtSignature) continue;
+				if (isSameProviderAndModel) parts.push({
+					thought: true,
+					text: sanitizeSurrogates(block.thinking),
+					...thoughtSignature && { thoughtSignature }
+				});
+				else parts.push({ text: sanitizeSurrogates(block.thinking) });
 			} else if (block.type === "toolCall") {
-				const thoughtSignature = resolveThoughtSignature(isSameProviderAndModel, block.thoughtSignature);
+				const args = coerceTransportToolCallArguments(block.arguments);
+				const thoughtSignature = resolveThoughtSignature(isSameProviderAndModel, block.thoughtSignature) ?? (!sawFunctionCall && requiresToolCallThoughtSignature ? "skip_thought_signature_validator" : void 0);
+				sawFunctionCall = true;
 				const part = {
 					functionCall: {
 						name: block.name,
-						args: block.arguments ?? {},
+						args,
 						...requiresToolCallId(model.id) ? { id: block.id } : {}
 					},
 					...thoughtSignature && { thoughtSignature }
@@ -181,45 +184,22 @@ function convertMessages(model, context) {
 		}
 	}
 	flushToolResultRun();
+	if (contents.length === 0) contents.push({
+		role: "user",
+		parts: [{ text: " " }]
+	});
 	return contents;
 }
-const JSON_SCHEMA_META_DECLARATIONS = /* @__PURE__ */ new Set([
-	"$schema",
-	"$id",
-	"$anchor",
-	"$dynamicAnchor",
-	"$vocabulary",
-	"$comment",
-	"$defs",
-	"definitions"
-]);
-/**
-* Strip meta-declarations from a schema obj
-*/
-function sanitizeForOpenApi(schema) {
-	if (typeof schema !== "object" || schema === null || Array.isArray(schema)) return schema;
-	const result = {};
-	for (const [key, value] of Object.entries(schema)) {
-		if (JSON_SCHEMA_META_DECLARATIONS.has(key)) continue;
-		result[key] = sanitizeForOpenApi(value);
-	}
-	return result;
-}
 /**
 * Convert tools to Gemini function declarations format.
-*
-* By default uses `parametersJsonSchema` which supports full JSON Schema (including
-* anyOf, oneOf, const, etc.). Set `useParameters` to true to use the legacy `parameters`
-* field instead (OpenAPI 3.03 Schema). This is needed for Cloud Code Assist with Claude
-* models, where the API translates `parameters` into Anthropic's `input_schema`.
 * @internal Directly tested provider implementation detail.
 */
-function convertTools(tools, useParameters = false) {
+function convertTools(tools) {
 	if (tools.length === 0) return;
-	return [{ functionDeclarations: tools.map((tool) => ({
+	return [{ functionDeclarations: sortPromptCacheToolsByName(tools).map((tool) => ({
 		name: tool.name,
 		description: tool.description,
-		...useParameters ? { parameters: sanitizeForOpenApi(tool.parameters) } : { parametersJsonSchema: tool.parameters }
+		parametersJsonSchema: tool.parameters
 	})) }];
 }
 /**
@@ -276,17 +256,23 @@ async function runGoogleGenerateContentLifecycle(params) {
 		});
 	} catch (error) {
 		for (const block of output.content) if ("index" in block) delete block.index;
-		output.stopReason = options?.signal?.aborted ? "aborted" : "error";
-		output.errorMessage = formatProviderError(error);
+		const failure = options?.signal?.aborted ? transportAbortError(options.signal) : error;
+		assignTransportErrorDetails(output, failure, options?.signal);
+		const formattedError = formatProviderError(failure);
+		const status = failure instanceof Error && "status" in failure ? failure.status : void 0;
+		if (typeof status === "number" && Number.isFinite(status)) {
+			output.errorCode ||= String(status);
+			output.errorMessage = formattedError.startsWith(`${status}:`) ? formattedError : `${status}: ${formattedError}`;
+		} else output.errorMessage = formattedError;
 		stream.push({
 			type: "error",
-			reason: output.stopReason,
+			reason: output.stopReason === "aborted" ? "aborted" : "error",
 			error: output
 		});
 		stream.end();
 	}
 }
-function buildGoogleGenerateContentParams(model, context, options = {}, configHooks) {
+function buildGoogleGenerateContentParams(model, context, options = {}) {
 	const contents = convertMessages(model, context);
 	const generationConfig = {};
 	if (options.temperature !== void 0) generationConfig.temperature = options.temperature;
@@ -304,7 +290,10 @@ function buildGoogleGenerateContentParams(model, context, options = {}, configHo
 		if (options.thinking.level !== void 0) thinkingConfig.thinkingLevel = ThinkingLevel[options.thinking.level];
 		else if (options.thinking.budgetTokens !== void 0) thinkingConfig.thinkingBudget = options.thinking.budgetTokens;
 		config.thinkingConfig = thinkingConfig;
-	} else if (model.reasoning && options.thinking && !options.thinking.enabled) config.thinkingConfig = configHooks?.getDisabledThinkingConfig ? configHooks.getDisabledThinkingConfig(model) : getDisabledGoogleThinkingConfig(model);
+	} else if (model.reasoning && options.thinking && !options.thinking.enabled) {
+		const disabledThinkingConfig = getDisabledGoogleThinkingConfig(model);
+		if (Object.keys(disabledThinkingConfig).length > 0) config.thinkingConfig = disabledThinkingConfig;
+	}
 	if (options.signal) {
 		if (options.signal.aborted) throw new Error("Request aborted");
 		config.abortSignal = options.signal;
@@ -315,8 +304,22 @@ function buildGoogleGenerateContentParams(model, context, options = {}, configHo
 		config
 	};
 }
+function isAdaptiveGoogleReasoningLevel(value) {
+	return value === "adaptive";
+}
 function buildGoogleSimpleThinking(model, options, config) {
 	if (!options?.reasoning || options.reasoning === "off") return { enabled: false };
+	if (isAdaptiveGoogleReasoningLevel(options.reasoning)) {
+		if (!model.reasoning) return { enabled: false };
+		if (isGemma4Model(model)) return {
+			enabled: true,
+			level: ThinkingLevel.HIGH
+		};
+		return isGemini3ProModel(model) || isGemini3FlashModel(model) ? { enabled: true } : {
+			enabled: true,
+			budgetTokens: -1
+		};
+	}
 	const clampedReasoning = clampThinkingLevel(model, options.reasoning);
 	if (clampedReasoning === "off") return { enabled: false };
 	const effort = clampedReasoning === "max" ? "high" : clampedReasoning;
@@ -329,10 +332,10 @@ function buildGoogleSimpleThinking(model, options, config) {
 		budgetTokens: getGoogleBudget(model, effort, options.thinkingBudgets, { useFlashLiteBudgets: config?.useFlashLiteBudgets })
 	};
 }
-function getDisabledGoogleThinkingConfig(model, config) {
+function getDisabledGoogleThinkingConfig(model) {
 	if (isGemini3ProModel(model)) return { thinkingLevel: ThinkingLevel.LOW };
 	if (isGemini3FlashModel(model)) return { thinkingLevel: ThinkingLevel.MINIMAL };
-	if (config?.includeGemma4 && isGemma4Model(model)) return { thinkingLevel: ThinkingLevel.MINIMAL };
+	if (isGemma4Model(model) || model.id.toLowerCase().includes("gemini-2.5-pro")) return {};
 	return { thinkingBudget: 0 };
 }
 /** @internal Directly tested provider implementation detail. */
@@ -340,10 +343,10 @@ function isGemma4Model(model) {
 	return /gemma-?4/.test(model.id.toLowerCase());
 }
 function isGemini3ProModel(model) {
-	return /gemini-3(?:\.\d+)?-pro/.test(model.id.toLowerCase());
+	return /gemini-(?:3(?:\.\d+)?-pro|pro-latest)/.test(model.id.toLowerCase());
 }
 function isGemini3FlashModel(model) {
-	return /gemini-3(?:\.\d+)?-flash/.test(model.id.toLowerCase());
+	return /gemini-(?:3(?:\.\d+)?-flash|flash(?:-lite)?-latest)/.test(model.id.toLowerCase());
 }
 function getGoogleThinkingLevel(effort, model, config) {
 	if (isGemini3ProModel(model)) switch (effort) {
@@ -422,6 +425,15 @@ async function consumeGoogleGenerateContentStream(params) {
 	});
 	let currentBlock = null;
 	const blocks = params.output.content;
+	let sawTerminalReason = false;
+	let terminalGenerationError;
+	const knownUsage = {
+		promptTokenCount: 0,
+		cachedContentTokenCount: 0,
+		toolUsePromptTokenCount: 0,
+		candidatesTokenCount: 0,
+		thoughtsTokenCount: 0
+	};
 	const toolCallIds = /* @__PURE__ */ new Set();
 	for (const block of blocks) if (block.type === "toolCall") toolCallIds.add(block.id);
 	const blockIndex = () => blocks.length - 1;
@@ -443,9 +455,60 @@ async function consumeGoogleGenerateContentStream(params) {
 	};
 	for await (const chunk of params.chunks) {
 		params.output.responseId ||= chunk.responseId;
+		if (chunk.usageMetadata) {
+			for (const field of Object.keys(knownUsage)) {
+				const value = chunk.usageMetadata[field];
+				if (typeof value === "number") knownUsage[field] = value;
+			}
+			const promptTokens = knownUsage.promptTokenCount;
+			const cacheRead = knownUsage.cachedContentTokenCount;
+			const toolUsePromptTokens = knownUsage.toolUsePromptTokenCount;
+			const outputTokens = knownUsage.candidatesTokenCount + knownUsage.thoughtsTokenCount;
+			params.output.usage = {
+				input: Math.max(0, promptTokens - cacheRead) + toolUsePromptTokens,
+				output: outputTokens,
+				cacheRead,
+				cacheWrite: 0,
+				totalTokens: chunk.usageMetadata.totalTokenCount ?? promptTokens + outputTokens + toolUsePromptTokens,
+				cost: {
+					input: 0,
+					output: 0,
+					cacheRead: 0,
+					cacheWrite: 0,
+					total: 0
+				}
+			};
+			calculateCost(params.model, params.output.usage);
+		}
 		const candidate = chunk.candidates?.[0];
-		if (candidate?.content?.parts) for (const part of candidate.content.parts) {
-			if (part.text !== void 0) {
+		const promptFeedback = chunk.promptFeedback;
+		if (!candidate && promptFeedback) {
+			const blockReason = promptFeedback.blockReason ?? "PROMPT_BLOCKED";
+			const blockMessage = promptFeedback.blockReasonMessage?.trim();
+			params.output.errorCode = blockReason;
+			params.output.errorType = "google_prompt_blocked";
+			throw new Error(`Google prompt blocked (${blockReason})${blockMessage ? `: ${blockMessage}` : ""}`);
+		}
+		if (candidate?.content?.parts) for (const [partIndex, part] of candidate.content.parts.entries()) {
+			const text = part.text;
+			const hasText = typeof text === "string";
+			const hasThoughtSignature = typeof part.thoughtSignature === "string" && part.thoughtSignature.length > 0;
+			const signatureOnly = hasThoughtSignature && (!hasText || text.length === 0) && Object.keys(part).every((key) => key === "thought" || key === "thoughtSignature" || key === "text");
+			if (signatureOnly) {
+				if (!hasText && part.thought !== true) {
+					const latestBlock = blocks.at(-1);
+					if (partIndex === 0 && latestBlock?.type === "toolCall" && !latestBlock.thoughtSignature) {
+						latestBlock.thoughtSignature = retainThoughtSignature(latestBlock.thoughtSignature, part.thoughtSignature);
+						continue;
+					}
+				}
+				endCurrentBlock();
+			}
+			if (hasText || signatureOnly) {
+				if (currentBlock && (hasThoughtSignature || partIndex > 0)) {
+					const currentSignature = currentBlock.type === "thinking" ? currentBlock.thinkingSignature : currentBlock.textSignature;
+					if ((currentBlock.type === "thinking" ? currentBlock.thinking : currentBlock.text).length > 0 && (currentSignature !== part.thoughtSignature || partIndex > 0 && (currentSignature || hasThoughtSignature))) endCurrentBlock();
+				}
 				const isThinking = isThinkingPart(part);
 				if (!currentBlock || isThinking && currentBlock.type !== "thinking" || !isThinking && currentBlock.type !== "text") {
 					endCurrentBlock();
@@ -474,25 +537,27 @@ async function consumeGoogleGenerateContentStream(params) {
 						});
 					}
 				}
+				const delta = hasText ? text : "";
 				if (currentBlock.type === "thinking") {
-					currentBlock.thinking += part.text;
+					currentBlock.thinking += delta;
 					currentBlock.thinkingSignature = retainThoughtSignature(currentBlock.thinkingSignature, part.thoughtSignature);
 					params.stream.push({
 						type: "thinking_delta",
 						contentIndex: blockIndex(),
-						delta: part.text,
+						delta,
 						partial: params.output
 					});
 				} else {
-					currentBlock.text += part.text;
+					currentBlock.text += delta;
 					currentBlock.textSignature = retainThoughtSignature(currentBlock.textSignature, part.thoughtSignature);
 					params.stream.push({
 						type: "text_delta",
 						contentIndex: blockIndex(),
-						delta: part.text,
+						delta,
 						partial: params.output
 					});
 				}
+				if (signatureOnly) endCurrentBlock();
 			}
 			if (part.functionCall) {
 				endCurrentBlock();
@@ -526,29 +591,30 @@ async function consumeGoogleGenerateContentStream(params) {
 			}
 		}
 		if (candidate?.finishReason) {
+			sawTerminalReason = true;
 			params.output.stopReason = mapStopReason(candidate.finishReason);
+			if (params.output.stopReason === "error") {
+				const finishMessage = candidate.finishMessage?.trim();
+				terminalGenerationError = Object.assign(/* @__PURE__ */ new Error(`Google generation stopped (${candidate.finishReason})${finishMessage ? `: ${finishMessage}` : ""}`), {
+					code: candidate.finishReason,
+					type: "google_generation_failed"
+				});
+			}
 			if (params.output.stopReason === "stop" && params.output.content.some((block) => block.type === "toolCall")) params.output.stopReason = "toolUse";
 		}
-		if (chunk.usageMetadata) {
-			params.output.usage = {
-				input: (chunk.usageMetadata.promptTokenCount || 0) - (chunk.usageMetadata.cachedContentTokenCount || 0),
-				output: (chunk.usageMetadata.candidatesTokenCount || 0) + (chunk.usageMetadata.thoughtsTokenCount || 0),
-				cacheRead: chunk.usageMetadata.cachedContentTokenCount || 0,
-				cacheWrite: 0,
-				totalTokens: chunk.usageMetadata.totalTokenCount || 0,
-				cost: {
-					input: 0,
-					output: 0,
-					cacheRead: 0,
-					cacheWrite: 0,
-					total: 0
-				}
-			};
-			calculateCost(params.model, params.output.usage);
-		}
 	}
 	endCurrentBlock();
 	if (params.signal?.aborted) throw transportAbortError(params.signal);
+	if (terminalGenerationError) {
+		params.output.errorCode = terminalGenerationError.code;
+		params.output.errorType = terminalGenerationError.type;
+		throw terminalGenerationError;
+	}
+	if (!sawTerminalReason) {
+		params.output.errorCode = "STREAM_INCOMPLETE";
+		params.output.errorType = "google_incomplete_stream";
+		throw new Error("Google stream ended before a terminal finish reason");
+	}
 	if (params.output.stopReason === "aborted" || params.output.stopReason === "error") throw new Error("An unknown error occurred");
 	params.stream.push({
 		type: "done",
@@ -558,4 +624,4 @@ async function consumeGoogleGenerateContentStream(params) {
 	params.stream.end();
 }
 //#endregion
-export { runGoogleGenerateContentLifecycle as a, getDisabledGoogleThinkingConfig as i, buildGoogleSimpleThinking as n, createGoogleAssistantOutput as r, buildGoogleGenerateContentParams as t };
+export { runGoogleGenerateContentLifecycle as i, buildGoogleSimpleThinking as n, createGoogleAssistantOutput as r, buildGoogleGenerateContentParams as t };

package/dist/{google-vertex-31f1uS9L.mjs → google-vertex-CusxWzvE.mjs}

@@ -1,7 +1,7 @@
-import { t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
-import { n as getAiTransportHost, r as resolveAiTransportHeaderSentinels } from "./host-Dog2WQiR.mjs";
-import { n as buildBaseOptions } from "./simple-options-9lhRrN73.mjs";
-import { a as runGoogleGenerateContentLifecycle, n as buildGoogleSimpleThinking, r as createGoogleAssistantOutput, t as buildGoogleGenerateContentParams } from "./google-shared-DNBz5rcD.mjs";
+import { t as AssistantMessageEventStream } from "./event-stream-CEV9t6da.mjs";
+import { n as getAiTransportHost, r as resolveAiTransportHeaderSentinels, v as normalizeOptionalString } from "./host-Bl7Kgddo.mjs";
+import { c as buildBaseOptions } from "./tool-result-text-Dvkp2Dus.mjs";
+import { i as runGoogleGenerateContentLifecycle, n as buildGoogleSimpleThinking, r as createGoogleAssistantOutput, t as buildGoogleGenerateContentParams } from "./google-shared-Cq2UKx1l.mjs";
 import { GoogleGenAI, ResourceScope } from "@google/genai";
 //#region packages/ai/src/providers/google-vertex.ts
 const API_VERSION = "v1";
@@ -27,7 +27,10 @@ const streamSimpleGoogleVertex = (model, context, options) => {
 	const base = buildBaseOptions(model, options, void 0);
 	return streamGoogleVertex(model, context, {
 		...base,
-		thinking: buildGoogleSimpleThinking(model, options)
+		thinking: buildGoogleSimpleThinking(model, options, {
+			includeGemma4ThinkingLevel: true,
+			useFlashLiteBudgets: true
+		})
 	});
 };
 function createClient(model, project, location, optionsHeaders) {
@@ -82,12 +85,12 @@ function isPlaceholderApiKey(apiKey) {
 	return /^<[^>]+>$/.test(apiKey);
 }
 function resolveProject(options) {
-	const project = options?.project || process.env.GOOGLE_CLOUD_PROJECT || process.env.GCLOUD_PROJECT;
+	const project = normalizeOptionalString(options?.project) || normalizeOptionalString(process.env.GOOGLE_CLOUD_PROJECT) || normalizeOptionalString(process.env.GCLOUD_PROJECT);
 	if (!project) throw new Error("Vertex AI requires a project ID. Set GOOGLE_CLOUD_PROJECT/GCLOUD_PROJECT or pass project in options.");
 	return project;
 }
 function resolveLocation(options) {
-	const location = options?.location || process.env.GOOGLE_CLOUD_LOCATION;
+	const location = normalizeOptionalString(options?.location) || normalizeOptionalString(process.env.GOOGLE_CLOUD_LOCATION);
 	if (!location) throw new Error("Vertex AI requires a location. Set GOOGLE_CLOUD_LOCATION or pass location in options.");
 	return location;
 }

package/dist/{host-Dog2WQiR.mjs → host-Bl7Kgddo.mjs}

@@ -1,4 +1,5 @@
-import { a as requiresClaudeMandatoryAdaptiveThinking, c as resolveClaudeMythos5ModelIdentity, d as resolveClaudeSonnet5ModelIdentity, i as requiresClaudeDefaultSampling, o as resolveClaudeFable5ModelIdentity, u as resolveClaudeOpus5ModelIdentity } from "./src-QkygScBs.mjs";
+import { a as requiresClaudeMandatoryAdaptiveThinking, c as resolveClaudeMythos5ModelIdentity, d as resolveClaudeSonnet5ModelIdentity, g as supportsClaudeNativeXhighEffort, h as supportsClaudeNativeMaxEffort, i as requiresClaudeDefaultSampling, l as resolveClaudeNativeThinkingLevelMap, o as resolveClaudeFable5ModelIdentity, u as resolveClaudeOpus5ModelIdentity } from "./src-QkygScBs.mjs";
+import { r as clampThinkingLevel } from "./model-utils-Dau5dlgm.mjs";
 //#region packages/normalization-core/src/record-coerce.ts
 /** Type guard for non-array object records at browser-safe boundaries. */
 function isRecord(value) {
@@ -34,6 +35,8 @@ function hasNonEmptyString(value) {
 }
 //#endregion
 //#region packages/ai/src/providers/anthropic-model-contract.ts
+const ANTHROPIC_CLAUDE_CODE_VERSION = "2.1.75";
+const ANTHROPIC_CLAUDE_CODE_BILLING_SYSTEM_BLOCK = `x-anthropic-billing-header: cc_version=${ANTHROPIC_CLAUDE_CODE_VERSION}; cc_entrypoint=sdk-cli;`;
 function normalizeModelId(modelId) {
 	const normalized = normalizeLowercaseStringOrEmpty(modelId);
 	return (normalized.startsWith("anthropic/") ? normalized.slice(10) : normalized).replace(/[._\s]+/g, "-");
@@ -62,6 +65,41 @@ function requiresClaudeAdaptiveThinking(model) {
 function defaultsClaudeAdaptiveThinking(model) {
 	return requiresClaudeAdaptiveThinking(model) || normalizeApi(model.api) === "anthropic-messages" && (resolveClaudeOpus5ModelIdentity(model) !== void 0 || resolveClaudeSonnet5ModelIdentity(model) !== void 0);
 }
+/** Resolve provider-native effort once for direct and managed Claude requests. */
+function resolveAnthropicThinkingEffort(model, level) {
+	const requestedLevel = level;
+	const thinkingLevelMap = resolveClaudeNativeThinkingLevelMap(model);
+	const clampModel = {
+		...model,
+		...typeof model.params?.canonicalModelId === "string" ? { reasoning: true } : {},
+		...thinkingLevelMap ? { thinkingLevelMap } : {}
+	};
+	const resolvedLevel = requestedLevel ? clampThinkingLevel(clampModel, requestedLevel) : void 0;
+	const mapped = resolvedLevel ? thinkingLevelMap?.[resolvedLevel] : void 0;
+	if (typeof mapped === "string") return mapped;
+	switch (resolvedLevel) {
+		case "off":
+		case "minimal":
+		case "low": return "low";
+		case "medium": return "medium";
+		case "xhigh": return supportsClaudeNativeXhighEffort(model) ? "xhigh" : "high";
+		case "max": return supportsClaudeNativeMaxEffort(model) ? "max" : "high";
+		default: return "high";
+	}
+}
+/** Normalize Anthropic and Anthropic-compatible terminal reasons identically. */
+function mapAnthropicStopReason(reason) {
+	switch (reason) {
+		case "end_turn":
+		case "pause_turn":
+		case "stop_sequence": return "stop";
+		case "max_tokens": return "length";
+		case "tool_use": return "toolUse";
+		case "refusal":
+		case "sensitive": return "error";
+		default: throw new Error(`Unhandled stop reason: ${String(reason)}`);
+	}
+}
 /** Remove unsupported assistant prefills while preserving completed tool-use turns. */
 function prepareClaudeNoPrefillRequestContext(model, context) {
 	if (!resolveClaudeOpus5ModelIdentity(model) && !resolveClaudeSonnet5ModelIdentity(model)) return context;
@@ -366,4 +404,4 @@ function resolveAiTransportHeaderSentinels(headers) {
 	return resolvedHeaders ?? headers;
 }
 //#endregion
-export { applyClaudeRequestContract as a, requiresClaudeAdaptiveThinking as c, usesClaudeStreamingRefusalContract as d, hasNonEmptyString as f, isRecord as g, readStringValue as h, transformMessages as i, resolveModelBoundThinkingReplayMode as l, normalizeOptionalString as m, getAiTransportHost as n, defaultsClaudeAdaptiveThinking as o, normalizeLowercaseStringOrEmpty as p, resolveAiTransportHeaderSentinels as r, prepareClaudeNoPrefillRequestContext as s, configureAiTransportHost as t, usesClaudeFable5MessagesContract as u };
+export { normalizeLowercaseStringOrEmpty as _, ANTHROPIC_CLAUDE_CODE_BILLING_SYSTEM_BLOCK as a, isRecord as b, defaultsClaudeAdaptiveThinking as c, requiresClaudeAdaptiveThinking as d, resolveAnthropicThinkingEffort as f, hasNonEmptyString as g, usesClaudeStreamingRefusalContract as h, transformMessages as i, mapAnthropicStopReason as l, usesClaudeFable5MessagesContract as m, getAiTransportHost as n, ANTHROPIC_CLAUDE_CODE_VERSION as o, resolveModelBoundThinkingReplayMode as p, resolveAiTransportHeaderSentinels as r, applyClaudeRequestContract as s, configureAiTransportHost as t, prepareClaudeNoPrefillRequestContext as u, normalizeOptionalString as v, readStringValue as y };

package/dist/{host-B9GUmcra.d.mts → host-Dn4_SZI2.d.mts}

@@ -1,5 +1,5 @@
-import { E as Model, L as StreamFn, n as Api, u as Context } from "./types-bzp5k29J.mjs";
-import { n as ApiRegistry } from "./api-registry-DlMgPR39.mjs";
+import { E as Model, L as StreamFn, n as Api, u as Context } from "./types-CH7ReIcU.mjs";
+import { n as ApiRegistry } from "./api-registry-EpJoVwM1.mjs";
 
 //#region packages/ai/src/host.d.ts
 /** Provider capability facts needed by the package-owned transports. */

package/dist/index.d.mts

@@ -1,10 +1,10 @@
 import { a as requiresClaudeMandatoryAdaptiveThinking, c as resolveClaudeMythos5ModelIdentity, d as resolveClaudeSonnet5ModelIdentity, f as supportsClaude1MContext, g as supportsClaudeNativeXhighEffort, h as supportsClaudeNativeMaxEffort, i as requiresClaudeDefaultSampling, l as resolveClaudeNativeThinkingLevelMap, m as supportsClaudeFastMode, n as CLAUDE_OPUS_5_THINKING_PROFILE, o as resolveClaudeFable5ModelIdentity, p as supportsClaudeAdaptiveThinking, r as CLAUDE_SONNET_5_THINKING_PROFILE, s as resolveClaudeModelIdentity, t as CLAUDE_FABLE_5_THINKING_PROFILE, u as resolveClaudeOpus5ModelIdentity } from "./anthropic-SrGtwsJu.mjs";
 import { a as extractDiagnosticError, i as createAssistantMessageDiagnostic, n as DiagnosticErrorInfo, o as formatThrownValue, r as appendAssistantMessageDiagnostic, t as AssistantMessageDiagnostic } from "./diagnostics-BaTA9eVl.mjs";
-import { $ as VercelGatewayRouting, A as OpenRouterRouting, B as TextContent, C as KnownImagesProvider, D as ModelThinkingLevel, E as Model, F as SimpleStreamOptions, G as ThinkingLevelMap, H as ThinkingBudgets, I as StopReason, J as ToolResultMessage, K as Tool, L as StreamFn, M as ProviderImagesOptions, N as ProviderResponse, O as OpenAICompletionsCompat, P as ProviderStreamOptions, Q as ValidateToolArgumentsFn, R as StreamFunction, S as KnownImagesApi, T as Message, U as ThinkingContent, V as TextSignatureV1, W as ThinkingLevel, X as Usage, Y as Transport, Z as UserMessage, _ as ImagesOptions, a as AssistantMessageEvent, b as ImagesStopReason, c as CacheRetention, d as ImageContent, f as ImagesApi, g as ImagesModel, h as ImagesInputContent, i as AssistantMessage, j as Provider, k as OpenAIResponsesCompat, l as CompleteSimpleFn, m as ImagesFunction, n as Api, o as AssistantMessageEventStreamContract, p as ImagesContext, q as ToolCall, r as AssistantImages, s as AssistantMessageEventStreamLike, t as AnthropicMessagesCompat, u as Context, v as ImagesOutputContent, w as MaybePromise, x as KnownApi, y as ImagesProvider, z as StreamOptions } from "./types-bzp5k29J.mjs";
-import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-YjaPW20U.mjs";
-import { n as validateToolCall, t as validateToolArguments } from "./validation-B-j7cOYp.mjs";
-import { a as RegisteredApiProvider, i as ApiStreamSimpleFunction, n as ApiRegistry, o as createApiRegistry, r as ApiStreamFunction, t as ApiProvider } from "./api-registry-DlMgPR39.mjs";
-import { a as AiProviderRequestPolicyInput, c as AiTransportHost, d as configureAiTransportHost, f as getAiTransportHost, i as AiProviderRequestCapabilities, l as AiTransportPluginHost, n as AiInlineImageBlock, o as AiProviderStreamHookContext, p as resolveAiTransportHeaderSentinels, r as AiInlineTextBlock, s as AiTransformTransportMessages, t as AiInlineContentBlock, u as OpenAIStrictToolSettingOptions } from "./host-B9GUmcra.mjs";
+import { $ as VercelGatewayRouting, A as OpenRouterRouting, B as TextContent, C as KnownImagesProvider, D as ModelThinkingLevel, E as Model, F as SimpleStreamOptions, G as ThinkingLevelMap, H as ThinkingBudgets, I as StopReason, J as ToolResultMessage, K as Tool, L as StreamFn, M as ProviderImagesOptions, N as ProviderResponse, O as OpenAICompletionsCompat, P as ProviderStreamOptions, Q as ValidateToolArgumentsFn, R as StreamFunction, S as KnownImagesApi, T as Message, U as ThinkingContent, V as TextSignatureV1, W as ThinkingLevel, X as Usage, Y as Transport, Z as UserMessage, _ as ImagesOptions, a as AssistantMessageEvent, b as ImagesStopReason, c as CacheRetention, d as ImageContent, f as ImagesApi, g as ImagesModel, h as ImagesInputContent, i as AssistantMessage, j as Provider, k as OpenAIResponsesCompat, l as CompleteSimpleFn, m as ImagesFunction, n as Api, o as AssistantMessageEventStreamContract, p as ImagesContext, q as ToolCall, r as AssistantImages, s as AssistantMessageEventStreamLike, t as AnthropicMessagesCompat, u as Context, v as ImagesOutputContent, w as MaybePromise, x as KnownApi, y as ImagesProvider, z as StreamOptions } from "./types-CH7ReIcU.mjs";
+import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-DRmDMRHH.mjs";
+import { n as validateToolCall, t as validateToolArguments } from "./validation-CcPcuEbR.mjs";
+import { a as RegisteredApiProvider, i as ApiStreamSimpleFunction, n as ApiRegistry, o as createApiRegistry, r as ApiStreamFunction, t as ApiProvider } from "./api-registry-EpJoVwM1.mjs";
+import { a as AiProviderRequestPolicyInput, c as AiTransportHost, d as configureAiTransportHost, f as getAiTransportHost, i as AiProviderRequestCapabilities, l as AiTransportPluginHost, n as AiInlineImageBlock, o as AiProviderStreamHookContext, p as resolveAiTransportHeaderSentinels, r as AiInlineTextBlock, s as AiTransformTransportMessages, t as AiInlineContentBlock, u as OpenAIStrictToolSettingOptions } from "./host-Dn4_SZI2.mjs";
 
 //#region packages/ai/src/stream.d.ts
 /** Creates an isolated LLM runtime backed by the supplied provider registry. */

package/dist/index.mjs

@@ -1,7 +1,7 @@
 import { a as requiresClaudeMandatoryAdaptiveThinking, c as resolveClaudeMythos5ModelIdentity, d as resolveClaudeSonnet5ModelIdentity, f as supportsClaude1MContext, g as supportsClaudeNativeXhighEffort, h as supportsClaudeNativeMaxEffort, i as requiresClaudeDefaultSampling, l as resolveClaudeNativeThinkingLevelMap, m as supportsClaudeFastMode, n as CLAUDE_OPUS_5_THINKING_PROFILE, o as resolveClaudeFable5ModelIdentity, p as supportsClaudeAdaptiveThinking, r as CLAUDE_SONNET_5_THINKING_PROFILE, s as resolveClaudeModelIdentity, t as CLAUDE_FABLE_5_THINKING_PROFILE, u as resolveClaudeOpus5ModelIdentity } from "./src-QkygScBs.mjs";
 import { i as formatThrownValue, n as createAssistantMessageDiagnostic, r as extractDiagnosticError, t as appendAssistantMessageDiagnostic } from "./diagnostics-COpOtRwq.mjs";
-import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-D8n2uFee.mjs";
+import { n as EventStream, r as createAssistantMessageEventStream, t as AssistantMessageEventStream } from "./event-stream-CEV9t6da.mjs";
 import { n as validateToolCall, t as validateToolArguments } from "./validation-DAa_yFOM.mjs";
 import { n as createApiRegistry, t as createLlmRuntime } from "./stream-CREqxHgU.mjs";
-import { n as getAiTransportHost, r as resolveAiTransportHeaderSentinels, t as configureAiTransportHost } from "./host-Dog2WQiR.mjs";
+import { n as getAiTransportHost, r as resolveAiTransportHeaderSentinels, t as configureAiTransportHost } from "./host-Bl7Kgddo.mjs";
 export { AssistantMessageEventStream, CLAUDE_FABLE_5_THINKING_PROFILE, CLAUDE_OPUS_5_THINKING_PROFILE, CLAUDE_SONNET_5_THINKING_PROFILE, EventStream, appendAssistantMessageDiagnostic, configureAiTransportHost, createApiRegistry, createAssistantMessageDiagnostic, createAssistantMessageEventStream, createLlmRuntime, extractDiagnosticError, formatThrownValue, getAiTransportHost, requiresClaudeDefaultSampling, requiresClaudeMandatoryAdaptiveThinking, resolveAiTransportHeaderSentinels, resolveClaudeFable5ModelIdentity, resolveClaudeModelIdentity, resolveClaudeMythos5ModelIdentity, resolveClaudeNativeThinkingLevelMap, resolveClaudeOpus5ModelIdentity, resolveClaudeSonnet5ModelIdentity, supportsClaude1MContext, supportsClaudeAdaptiveThinking, supportsClaudeFastMode, supportsClaudeNativeMaxEffort, supportsClaudeNativeXhighEffort, validateToolArguments, validateToolCall };