@openclaw/ai 2026.9.4 → 2026.9.6

package/dist/{openai-completions-stream-Da2vvl-S.mjs → openai-completions-stream-BpuJ0ZOA.mjs}

@@ -1,26 +1,27 @@
 import { r as appendAssistantThinking } from "./event-stream-D8PARQfL.mjs";
-import { _ as isImageWithMediaPayload, d as describeToolResultMediaPlaceholder, m as extractToolResultText, v as sanitizeSurrogates } from "./host-B8YfDGd4.mjs";
 import { o as isRecord } from "./record-coerce-DwRYMj3t.mjs";
+import { n as isOpenAIGpt55Model, r as isOpenAIGpt56Model, t as isOpenAIGpt54MiniModel } from "./openai-reasoning-effort-ash8dPrd.mjs";
+import { r as uniqueStrings } from "./string-normalization-J9ZiLfGO.mjs";
+import { _ as isImageWithMediaPayload, d as describeToolResultMediaPlaceholder, m as extractToolResultText, v as sanitizeSurrogates } from "./host-yTvBrYM2.mjs";
 import { t as truncateUtf16Safe } from "./utf16-slice-CvGodqok.mjs";
-import { d as shouldOmitOllamaCompatResponseFormat, i as detectOpenAICompletionsCompat, r as resolveOpenAIPromptCacheParams, u as resolveOpenAICompletionsResponseFormat } from "./openai-prompt-cache-Bds-n_9Q.mjs";
-import { r as emitModelTransportDebug } from "./diagnostics-QuErwCIl.mjs";
-import { i as asNonNegativeFiniteNumber } from "./base64-D-su8YVo.mjs";
-import { c as supportsModelTools, r as estimateStringChars } from "./transport-utils-zrYjICLZ.mjs";
-import { n as uniqueStrings } from "./string-normalization-CmLIasuf.mjs";
-import { i as resolveProviderEndpoint, r as resolveOpenAIStrictToolSetting } from "./host-policy-Zcg_cNz8.mjs";
+import { r as resolveOpenAIPromptCacheParams } from "./openai-prompt-cache-CjV_2IRI.mjs";
+import { r as emitModelTransportDebug } from "./diagnostics-CH8j2wIG.mjs";
+import { i as asNonNegativeFiniteNumber } from "./base64-CPmadnUf.mjs";
+import { c as supportsModelTools, r as estimateStringChars } from "./transport-utils-03Qi0CN2.mjs";
+import { c as shouldOmitOllamaCompatResponseFormat, n as isNativeOpenAIEndpoint, s as resolveOpenAICompletionsResponseFormat, t as detectOpenAICompletionsCompat } from "./openai-completions-compat-CUollwl6.mjs";
+import { i as resolveProviderEndpoint, r as resolveOpenAIStrictToolSetting } from "./host-policy-DZO3oPPk.mjs";
 import { t as resolveCacheRetention } from "./cache-retention-0x979a5V.mjs";
-import { f as splitSystemPromptCacheBoundary, h as stripSystemPromptRelocatableBoundary, m as stripSystemPromptCacheBoundary, p as splitSystemPromptRelocatableBoundary, v as sortPromptCacheToolsByName } from "./simple-options-BQbb4yQL.mjs";
-import { C as readOpenAICompletionsContentDeltas, D as throwIfModelStreamAborted, E as resolvePromptCacheKey, S as parseOpenAICompletionsUsage, b as log, d as resolveOpenAIReasoningEffortMap, f as projectOpenAITools, g as createModelStreamCooperativeScheduler, h as GEMINI_THOUGHT_SIGNATURE_VALIDATOR_SKIP, p as reconcileOpenAICompletionsToolChoice, s as getCompat, u as resolveOpenAIStrictToolFlagWithDiagnostics, w as readOpenAICompletionsReasoningBatch, x as measureUtf8AppendBytes, y as isOpenAICompletionsThinkingEnabled } from "./openai-transport-params-9aPuV5YY.mjs";
-import { E as parseStreamingJson, c as finalizeTerminalToolCallArguments, w as createToolArgumentPreviewSchedule } from "./transport-stream-shared-zHll9BxO.mjs";
+import { f as splitSystemPromptCacheBoundary, h as stripSystemPromptRelocatableBoundary, m as stripSystemPromptCacheBoundary, p as splitSystemPromptRelocatableBoundary, v as sortPromptCacheToolsByName } from "./simple-options-np1XPuhm.mjs";
+import { A as resolvePromptCacheKey, C as isOpenAICompletionsThinkingEnabled, D as readOpenAICompletionsContentDeltas, E as parseOpenAICompletionsUsage, O as readOpenAICompletionsReasoningBatch, T as measureUtf8AppendBytes, d as prepareOpenAITools, f as projectOpenAITools, g as resolveOpenAISimpleReasoningEffort, h as resolveOpenAIRequestReasoning, j as throwIfModelStreamAborted, p as reconcileOpenAICompletionsToolChoice, s as getCompat, u as resolveOpenAIStrictToolFlagWithDiagnostics, v as GEMINI_THOUGHT_SIGNATURE_VALIDATOR_SKIP, w as log, y as createModelStreamCooperativeScheduler } from "./openai-transport-params-DZKOTSfH.mjs";
+import { E as parseStreamingJson, c as finalizeTerminalToolCallArguments, w as createToolArgumentPreviewSchedule } from "./transport-stream-shared-B5RioB_Q.mjs";
 import { a as tagUnresolvedTextAsCommentary, i as tagPendingCommentaryText, n as rememberPendingCommentaryTags, r as tagInterruptedTextPhases, t as clearPendingCommentaryText } from "./assistant-text-phase-C20rxWwP.mjs";
 import { t as notifyLlmRequestActivity } from "./llm-request-activity-BjtkplhG.mjs";
-import { a as withFirstStreamEventTimeout } from "./stream-first-event-timeout-2hfrquuz.mjs";
-import { t as transformProviderMessages } from "./provider-transcript-transform-V5YzU9zh.mjs";
-import { _ as isOpenAIGpt56Model, b as resolveOpenAIReasoningEffortForModel, g as isOpenAIGpt55Model, h as isOpenAIGpt54MiniModel } from "./openai-responses-contracts-DDOHA62Y.mjs";
-import { r as normalizeOpenAIStrictToolParameters } from "./openai-tool-schema-CzjyYXun.mjs";
+import { a as withFirstStreamEventTimeout } from "./stream-first-event-timeout-BeptfRL6.mjs";
+import { t as transformProviderMessages } from "./provider-transcript-transform-BTocjCRs.mjs";
+import { f as withPreparedToolSchemaNormalization, r as normalizeOpenAIStrictToolParameters } from "./openai-tool-schema-D4BMcj2D.mjs";
 import { isGoogleGemini3FlashModel, isGoogleGemini3ProModel } from "./internal/google-model-family.mjs";
 import { t as mapOpenAIStopReason } from "./openai-stop-reason-Drnn_6Qj.mjs";
-import { t as createReasoningTagTextPartitioner } from "./reasoning-tag-text-partitioner-BcR5pztD.mjs";
+import { t as createReasoningTagTextPartitioner } from "./reasoning-tag-text-partitioner-Di38crhI.mjs";
 import { randomUUID } from "node:crypto";
 //#region packages/ai/src/transports/deepseek-dsml-grammar.ts
 const DEEPSEEK_DSML_MARKERS = [
@@ -254,10 +255,18 @@ function createOpenAIEncryptedToolCallReasoningTracker() {
 	const firstBlocks = /* @__PURE__ */ new Map();
 	const pendingDetails = /* @__PURE__ */ new Map();
 	return {
-		rememberToolCall(id, block) {
+		rememberToolCall(id, block, previousId = id) {
+			const previous = firstBlocks.get(previousId);
+			if (previousId !== id && previous?.block === block) {
+				firstBlocks.delete(previousId);
+				if (previous.signature && block.thoughtSignature === previous.signature) delete block.thoughtSignature;
+			}
 			if (!id || firstBlocks.has(id)) return;
-			firstBlocks.set(id, block);
 			const pendingDetail = pendingDetails.get(id);
+			firstBlocks.set(id, {
+				block,
+				signature: pendingDetail
+			});
 			if (pendingDetail) {
 				block.thoughtSignature = pendingDetail;
 				pendingDetails.delete(id);
@@ -269,8 +278,10 @@ function createOpenAIEncryptedToolCallReasoningTracker() {
 				if (!isRecord(detail) || detail.type !== "reasoning.encrypted" || typeof detail.id !== "string" || detail.id.length === 0 || typeof detail.data !== "string" || detail.data.length === 0) continue;
 				const serializedDetail = JSON.stringify(detail);
 				const matchingBlock = firstBlocks.get(detail.id);
-				if (matchingBlock) matchingBlock.thoughtSignature = serializedDetail;
-				else pendingDetails.set(detail.id, serializedDetail);
+				if (matchingBlock) {
+					matchingBlock.signature = serializedDetail;
+					matchingBlock.block.thoughtSignature = serializedDetail;
+				} else pendingDetails.set(detail.id, serializedDetail);
 			}
 		}
 	};
@@ -336,7 +347,7 @@ function createOpenAICompletionsToolCallDeltaNormalizer() {
 			}];
 		}
 		const functionCall = delta.function_call;
-		if (sawModernToolCall) return [{
+		if (sawModernToolCall || !functionCall && !pendingLegacyToolCall) return [{
 			delta: ordinaryDelta,
 			toolCalls: []
 		}];
@@ -640,12 +651,9 @@ function relocateNonBehavioralRegion(args) {
 }
 //#endregion
 //#region packages/ai/src/transports/openai-completions-direct-policy.ts
-function applyDirectCompletionsReasoningAndRouting(params, model, options, compat) {
-	const reasoningEffortMap = resolveOpenAIReasoningEffortMap(model);
-	const thinkingLevelMap = model.thinkingLevelMap;
-	const offReasoningEffort = reasoningEffortMap.off ?? model.thinkingLevelMap?.off;
-	const reasoningEffort = options?.reasoningEffort === void 0 ? offReasoningEffort ?? void 0 : reasoningEffortMap[options.reasoningEffort] ?? thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort;
-	const reasoningEnabled = reasoningEffort !== void 0 && reasoningEffort !== "none";
+function applyDirectCompletionsReasoningAndRouting(params, model, reasoning, compat) {
+	const nativeEffort = compat.thinkingFormat === "openrouter" ? void 0 : reasoning.effort;
+	const reasoningEnabled = reasoning.thinkingEnabled ?? false;
 	if (compat.thinkingFormat === "zai" && model.reasoning) params.thinking = reasoningEnabled ? {
 		type: "enabled",
 		clear_thinking: false
@@ -657,17 +665,11 @@ function applyDirectCompletionsReasoningAndRouting(params, model, options, compa
 	};
 	else if (compat.thinkingFormat === "deepseek" && model.reasoning) {
 		params.thinking = { type: reasoningEnabled ? "enabled" : "disabled" };
-		if (reasoningEnabled && compat.supportsReasoningEffort) params.reasoning_effort = reasoningEffort;
-	} else if (compat.thinkingFormat === "openrouter" && model.reasoning) {
-		if (reasoningEnabled) params.reasoning = { effort: reasoningEffort };
-		else if (offReasoningEffort !== null) params.reasoning = { effort: offReasoningEffort ?? "none" };
+		if (reasoningEnabled && compat.supportsReasoningEffort && nativeEffort !== void 0) params.reasoning_effort = nativeEffort;
 	} else if (compat.thinkingFormat === "together" && model.reasoning) {
 		params.reasoning = { enabled: reasoningEnabled };
-		if (reasoningEnabled && compat.supportsReasoningEffort) params.reasoning_effort = reasoningEffort;
-	} else if (reasoningEnabled && model.reasoning && compat.supportsReasoningEffort) params.reasoning_effort = reasoningEffort;
-	else if (model.reasoning && compat.supportsReasoningEffort) {
-		if (typeof offReasoningEffort === "string") params.reasoning_effort = offReasoningEffort;
-	}
+		if (reasoningEnabled && compat.supportsReasoningEffort && nativeEffort !== void 0) params.reasoning_effort = nativeEffort;
+	} else if (model.reasoning && compat.supportsReasoningEffort && nativeEffort !== void 0) params.reasoning_effort = nativeEffort;
 	if (compat.openRouterRouting) params.provider = compat.openRouterRouting;
 	if (model.baseUrl.includes("ai-gateway.vercel.sh") && model.compat?.vercelGatewayRouting) {
 		const routing = model.compat.vercelGatewayRouting;
@@ -774,7 +776,9 @@ const REASONING_CONTENT_REPLAY_MODEL_IDS = /* @__PURE__ */ new Set([
 	"mimo-v2-omni",
 	"mimo-v2.5",
 	"mimo-v2.5-pro",
-	"mimo-v2.6-pro"
+	"mimo-v2.6-flash",
+	"mimo-v2.6-pro",
+	"mimo-v2.6-pro-ultraspeed"
 ]);
 const REASONING_CONTENT_REPLAY_TIER_SUFFIXES = [
 	"-free",
@@ -838,7 +842,7 @@ function applyCompletionsReplay(outgoingMessages, context, model, compat) {
 //#endregion
 //#region packages/ai/src/transports/openai-completions-params.ts
 function isKnownOpenAICompletionsEndpoint(model) {
-	if (!model.baseUrl.trim()) return true;
+	if (!model.baseUrl.trim() || isNativeOpenAIEndpoint(model)) return true;
 	const endpointClass = resolveProviderEndpoint(model).endpointClass;
 	if (endpointClass === "openai-public" || endpointClass === "azure-openai") return true;
 	try {
@@ -847,9 +851,6 @@ function isKnownOpenAICompletionsEndpoint(model) {
 		return false;
 	}
 }
-function resolveOpenAICompletionsReasoningEffort(options) {
-	return options?.reasoningEffort ?? options?.reasoning ?? "high";
-}
 function resolveOpenAICompletionsMaxTokens(model, options) {
 	if (options?.maxTokens) return {
 		maxTokens: options.maxTokens,
@@ -870,6 +871,7 @@ function resolveOpenAICompletionsModelMaxTokens(model) {
 }
 const OPENAI_COMPLETIONS_INPUT_TOKEN_SAFETY_MARGIN = 1.25;
 const OPENAI_COMPLETIONS_IMAGE_CHAR_ESTIMATE = 8e3;
+const MIN_USEFUL_OUTPUT_TOKENS = 16;
 function estimateOpenAICompletionsInputTokens(payload) {
 	let adjustedChars = 0;
 	adjustedChars += estimateOpenAICompletionsMessagesChars(payload.messages);
@@ -930,9 +932,6 @@ function resolveOpenAICompletionsEffectiveContextTokens(model) {
 	if (typeof contextTokens === "number" && Number.isFinite(contextTokens) && contextTokens > 0) return contextTokens;
 	return typeof model.contextWindow === "number" && Number.isFinite(model.contextWindow) && model.contextWindow > 0 ? model.contextWindow : void 0;
 }
-function isQwenOpenAICompletionsThinkingFormat(format) {
-	return format === "qwen" || format === "qwen-chat-template";
-}
 function setQwenChatTemplateThinking(params, enabled) {
 	const existing = params.chat_template_kwargs;
 	params.chat_template_kwargs = existing && typeof existing === "object" && !Array.isArray(existing) ? {
@@ -940,41 +939,51 @@ function setQwenChatTemplateThinking(params, enabled) {
 		enable_thinking: enabled
 	} : { enable_thinking: enabled };
 }
-function applyQwenOpenAICompletionsThinkingParams(params) {
-	if (!params.modelReasoning || !isQwenOpenAICompletionsThinkingFormat(params.compatThinkingFormat)) return false;
-	const enabled = isOpenAICompletionsThinkingEnabled(params.requestedEffort);
-	if (params.compatThinkingFormat === "qwen-chat-template") setQwenChatTemplateThinking(params.payload, enabled);
-	else params.payload.enable_thinking = enabled;
-	return true;
-}
-function applyTogetherOpenAICompletionsThinkingParams(params) {
-	if (!params.modelReasoning || params.compatThinkingFormat !== "together") return;
-	params.payload.reasoning = { enabled: isOpenAICompletionsThinkingEnabled(params.requestedEffort) };
+/** Return whether the binary control replaces scalar reasoning effort. */
+function applyBinaryCompletionsThinkingParams(params) {
+	if (!params.modelReasoning) return false;
+	const enabled = params.thinkingEnabled;
+	switch (params.compatThinkingFormat) {
+		case "qwen-chat-template":
+			setQwenChatTemplateThinking(params.payload, enabled);
+			return true;
+		case "qwen":
+			params.payload.enable_thinking = enabled;
+			return true;
+		case "together":
+			params.payload.reasoning = { enabled };
+			return !enabled;
+		default: return false;
+	}
 }
 function convertTools(tools, compat, model, mode) {
-	const projection = projectOpenAITools(tools);
-	const strict = mode === "direct" ? compat.supportsStrictMode ? false : void 0 : resolveOpenAIStrictToolFlagWithDiagnostics(projection, resolveOpenAIStrictToolSetting(model, {
-		transport: "stream",
-		supportsStrictMode: compat?.supportsStrictMode
-	}), {
-		transport: "completions",
-		model
-	});
-	return {
-		projection,
-		tools: sortPromptCacheToolsByName(projection.tools).map((tool) => {
-			const functionTool = {
-				name: tool.name,
-				description: tool.description,
-				parameters: mode === "direct" ? tool.parameters : normalizeOpenAIStrictToolParameters(tool.parameters, strict === true, model.compat)
-			};
-			if (strict !== void 0) functionTool.strict = strict;
-			return {
-				type: "function",
-				function: functionTool
-			};
-		})
+	const prepared = mode === "managed" ? prepareOpenAITools(tools) : void 0;
+	const projection = prepared?.projection ?? projectOpenAITools(tools);
+	const convert = () => {
+		const strict = mode === "direct" ? compat.supportsStrictMode ? false : void 0 : resolveOpenAIStrictToolFlagWithDiagnostics(projection, resolveOpenAIStrictToolSetting(model, {
+			transport: "stream",
+			supportsStrictMode: compat?.supportsStrictMode
+		}), {
+			transport: "completions",
+			model
+		});
+		return {
+			projection,
+			tools: sortPromptCacheToolsByName(projection.tools).map((tool) => {
+				const functionTool = {
+					name: tool.name,
+					description: tool.description,
+					parameters: mode === "direct" ? tool.parameters : normalizeOpenAIStrictToolParameters(tool.parameters, strict === true, model.compat)
+				};
+				if (strict !== void 0) functionTool.strict = strict;
+				return {
+					type: "function",
+					function: functionTool
+				};
+			})
+		};
 	};
+	return prepared ? withPreparedToolSchemaNormalization(prepared.schemas, convert) : convert();
 }
 function buildOpenAICompletionsParams(model, context, options) {
 	return buildOpenAICompletionsRequest(model, context, options, { mode: "managed" });
@@ -1071,6 +1080,7 @@ function buildOpenAICompletionsRequest(model, context, options, policy) {
 			if (clampedMaxTokens > remainingBudget) {
 				clampedMaxTokens = remainingBudget;
 				emitModelTransportDebug(log, `[completions] clamp_max_tokens provider=${model.provider} api=${model.api} model=${model.id} requested=${effectiveMaxTokens} output=${clampedMaxTokens} effectiveContext=${effectiveContextTokens} estimatedInput=${estimatedInputTokens}`);
+				if (remainingBudget < MIN_USEFUL_OUTPUT_TOKENS) log.warn(`[completions] insufficient_output_budget provider=${model.provider} api=${model.api} model=${model.id} output=${clampedMaxTokens} effectiveContext=${effectiveContextTokens} estimatedInput=${estimatedInputTokens}`);
 			}
 		}
 		if (policy.mode === "direct" ? options?.maxTokens : clampedMaxTokens) {
@@ -1078,34 +1088,34 @@ function buildOpenAICompletionsRequest(model, context, options, policy) {
 			else params.max_completion_tokens = clampedMaxTokens;
 		}
 	}
-	if (policy.mode === "direct") {
-		applyDirectCompletionsReasoningAndRouting(params, model, options, compat);
-		return params;
+	const isOpenRouter = compat.thinkingFormat === "openrouter";
+	const usesBinaryOpenRouterThinking = isOpenRouter && (model.compat?.supportsReasoningEffort === false || model.compat?.supportedReasoningEfforts?.length === 0);
+	const simpleReasoning = options?.reasoning;
+	const requestedEffort = policy.mode === "direct" ? options?.reasoningEffort : options?.reasoningEffort ?? (simpleReasoning === "none" ? "none" : resolveOpenAISimpleReasoningEffort({
+		...model,
+		compat: model.compat ?? void 0
+	}, simpleReasoning)) ?? (usesBinaryOpenRouterThinking ? void 0 : "high");
+	const reasoning = resolveOpenAIRequestReasoning(model, requestedEffort);
+	const { effort, thinkingEnabled } = reasoning;
+	if (isOpenRouter && model.reasoning) {
+		if (usesBinaryOpenRouterThinking && thinkingEnabled !== void 0) params.reasoning = { enabled: thinkingEnabled };
+		else if (effort !== void 0) params.reasoning = { effort };
+	}
+	if (policy.mode === "direct") applyDirectCompletionsReasoningAndRouting(params, model, reasoning, compat);
+	else {
+		const suppressScalarEffort = applyBinaryCompletionsThinkingParams({
+			compatThinkingFormat: compat.thinkingFormat,
+			modelReasoning: model.reasoning,
+			payload: params,
+			thinkingEnabled: thinkingEnabled ?? false
+		});
+		if (!isOpenRouter && effort && model.reasoning && compat.supportsReasoningEffort && !suppressScalarEffort) params.reasoning_effort = effort;
+		if (compat.cacheControlFormat === "anthropic") applyCompletionsAnthropicCacheControl(params, cacheControl ?? null, cacheOptOutIndexes, markTools, !managedCompat?.requiresStringContent);
+	}
+	if (params.tools?.length && isKnownOpenAICompletionsEndpoint(model)) {
+		if (isOpenAIGpt56Model(model)) params.reasoning_effort = "none";
+		else if (isOpenAIGpt54MiniModel(model) || isOpenAIGpt55Model(model)) delete params.reasoning_effort;
 	}
-	const completionsReasoningEffort = resolveOpenAICompletionsReasoningEffort(options);
-	const resolvedCompletionsReasoningEffort = completionsReasoningEffort ? resolveOpenAIReasoningEffortForModel({
-		model,
-		effort: completionsReasoningEffort,
-		fallbackMap: managedCompat?.reasoningEffortMap
-	}) : void 0;
-	const omitChatCompletionsToolReasoningEffort = Array.isArray(params.tools) && params.tools.length > 0 && (isOpenAIGpt54MiniModel(model) || isOpenAIGpt55Model(model) && isKnownOpenAICompletionsEndpoint(model));
-	const disableChatCompletionsToolReasoning = Array.isArray(params.tools) && params.tools.length > 0 && isOpenAIGpt56Model(model) && isKnownOpenAICompletionsEndpoint(model);
-	const handledQwenThinkingFormat = applyQwenOpenAICompletionsThinkingParams({
-		compatThinkingFormat: compat.thinkingFormat,
-		modelReasoning: model.reasoning,
-		payload: params,
-		requestedEffort: completionsReasoningEffort
-	});
-	applyTogetherOpenAICompletionsThinkingParams({
-		compatThinkingFormat: compat.thinkingFormat,
-		modelReasoning: model.reasoning,
-		payload: params,
-		requestedEffort: completionsReasoningEffort
-	});
-	if (disableChatCompletionsToolReasoning) params.reasoning_effort = "none";
-	else if (compat.thinkingFormat === "openrouter" && model.reasoning && resolvedCompletionsReasoningEffort) params.reasoning = { effort: resolvedCompletionsReasoningEffort };
-	else if (resolvedCompletionsReasoningEffort && model.reasoning && compat.supportsReasoningEffort && !handledQwenThinkingFormat && !omitChatCompletionsToolReasoningEffort) params.reasoning_effort = resolvedCompletionsReasoningEffort;
-	if (compat.cacheControlFormat === "anthropic") applyCompletionsAnthropicCacheControl(params, cacheControl ?? null, cacheOptOutIndexes, markTools, !managedCompat?.requiresStringContent);
 	return params;
 }
 //#endregion
@@ -1416,7 +1426,7 @@ async function processCompletionsStream(responseStream, output, model, stream, o
 	let isFlushingPendingPostToolCallDeltas = false;
 	const toolCallBlocksByIndex = /* @__PURE__ */ new Map();
 	const toolCallBlocksById = /* @__PURE__ */ new Map();
-	const encryptedReasoning = directMode ? createOpenAIEncryptedToolCallReasoningTracker() : void 0;
+	const encryptedReasoning = createOpenAIEncryptedToolCallReasoningTracker();
 	const toolArgumentPreviewSchedules = /* @__PURE__ */ new WeakMap();
 	const provisionalCommentaryTags = directMode ? options.provisionalCommentaryTags : /* @__PURE__ */ new Map();
 	const contentBlockIndices = /* @__PURE__ */ new WeakMap();
@@ -1751,7 +1761,7 @@ async function processCompletionsStream(responseStream, output, model, stream, o
 							partialArgs: "",
 							...initialSig ? { thoughtSignature: initialSig } : {}
 						};
-						encryptedReasoning?.rememberToolCall(block.id, block);
+						encryptedReasoning.rememberToolCall(block.id, block);
 						toolArgumentPreviewSchedules.set(block, createToolArgumentPreviewSchedule());
 						output.content.push(block);
 						toolCallBlockIndices.set(block, output.content.length - 1);
@@ -1763,9 +1773,10 @@ async function processCompletionsStream(responseStream, output, model, stream, o
 					}
 					if (streamIndex !== void 0 && !toolCallBlocksByIndex.has(streamIndex)) toolCallBlocksByIndex.set(streamIndex, block);
 					if (toolCall.id) {
+						const previousId = block.id;
 						if (!directMode || !block.id) block.id = toolCall.id;
 						toolCallBlocksById.set(toolCall.id, block);
-						if (block.id === toolCall.id) encryptedReasoning?.rememberToolCall(toolCall.id, block);
+						if (block.id === toolCall.id) encryptedReasoning.rememberToolCall(toolCall.id, block, previousId);
 					}
 					currentBlock = block;
 					const conflictingId = directMode && block.id && toolCall.id && block.id !== toolCall.id;
@@ -1785,7 +1796,7 @@ async function processCompletionsStream(responseStream, output, model, stream, o
 					});
 				}
 			}
-			encryptedReasoning?.consumeDetails(deltaFields.reasoning_details);
+			encryptedReasoning.consumeDetails(deltaFields.reasoning_details);
 		}
 		flushPendingPostToolCallDeltas();
 		emitReasoningUsageActivity(hasReasoningUsageActivity);

package/dist/{openai-prompt-cache-B4eYo2-I.d.mts → openai-prompt-cache-C4RHdN4n.d.mts}

@@ -1,5 +1,5 @@
-import { M as OpenAICompletionsCompat, O as Model, l as CacheRetention } from "./types-Ntv5z2g2.mjs";
-import "./types-DlfwzH3T.mjs";
+import { M as OpenAICompletionsCompat, O as Model, l as CacheRetention } from "./types-qZFbaDbK.mjs";
+import "./types-DTvddOfm.mjs";
 //#region packages/ai/src/providers/openai-prompt-cache.d.ts
 /** Selects documented lifetime fields shared by Responses and Chat Completions. */
 declare function resolveOpenAIPromptCacheParams(model: Pick<Model, "id" | "provider" | "baseUrl">, cacheRetention: CacheRetention, compat: Required<Pick<OpenAICompletionsCompat, "supportsPromptCacheKey" | "supportsLongCacheRetention">>): {

package/dist/openai-prompt-cache-CjV_2IRI.mjs

@@ -0,0 +1,46 @@
+import { n as isNativeOpenAIEndpoint } from "./openai-completions-compat-CUollwl6.mjs";
+//#region packages/normalization-core/src/code-points.ts
+/** Truncates to a nonnegative code-point budget; grapheme clusters may be split. */
+function truncateCodePoints(text, maxCodePoints) {
+	const limit = Math.max(0, Math.trunc(maxCodePoints) || 0);
+	if (text.length <= limit) return text;
+	const prefix = [];
+	for (const codePoint of text) {
+		if (prefix.length >= limit) break;
+		prefix.push(codePoint);
+	}
+	return prefix.join("");
+}
+//#endregion
+//#region packages/ai/src/providers/openai-prompt-cache.ts
+const EXTENDED_RETENTION_MODELS = /* @__PURE__ */ new Set([
+	"gpt-5.5",
+	"gpt-5.5-pro",
+	"gpt-5.4",
+	"gpt-5.2",
+	"gpt-5.1-codex-max",
+	"gpt-5.1",
+	"gpt-5.1-codex",
+	"gpt-5.1-codex-mini",
+	"gpt-5.1-chat-latest",
+	"gpt-5",
+	"gpt-5-codex",
+	"gpt-4.1"
+]);
+/** Selects documented lifetime fields shared by Responses and Chat Completions. */
+function resolveOpenAIPromptCacheParams(model, cacheRetention, compat) {
+	if (cacheRetention !== "long" || !compat.supportsPromptCacheKey || !compat.supportsLongCacheRetention) return {};
+	const version = /^gpt-(\d+)(?:\.(\d+))?(?:-|$)/.exec(model.id);
+	if (version && (Number(version[1]) > 5 || Number(version[1]) === 5 && Number(version[2]) >= 6)) return { prompt_cache_options: { ttl: "30m" } };
+	const modelId = model.id.replace(/-\d{4}-\d{2}-\d{2}$/, "");
+	return !isNativeOpenAIEndpoint(model) || EXTENDED_RETENTION_MODELS.has(modelId) ? { prompt_cache_retention: "24h" } : {};
+}
+/** Maximum prompt cache key length accepted by OpenAI-compatible request metadata. */
+const OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH = 64;
+/** Truncates a prompt cache key by Unicode code point count. */
+function clampOpenAIPromptCacheKey(key) {
+	if (key === void 0) return;
+	return truncateCodePoints(key, 64);
+}
+//#endregion
+export { clampOpenAIPromptCacheKey as n, resolveOpenAIPromptCacheParams as r, OPENAI_PROMPT_CACHE_KEY_MAX_LENGTH as t };

package/dist/{openai-provider-client-S2gCrM2Z.mjs → openai-provider-client-BP8J1aeg.mjs}

@@ -1,5 +1,5 @@
-import { n as getAiTransportHost } from "./host-B8YfDGd4.mjs";
-import { T as resolveOpenAIClientBaseUrl } from "./openai-transport-params-9aPuV5YY.mjs";
+import { n as getAiTransportHost } from "./host-yTvBrYM2.mjs";
+import { k as resolveOpenAIClientBaseUrl } from "./openai-transport-params-DZKOTSfH.mjs";
 import { i as resolveCloudflareBaseUrl, r as isCloudflareProvider } from "./github-copilot-headers-B37TB3rB.mjs";
 import OpenAI from "openai";
 //#region packages/ai/src/providers/openai-provider-client.ts