gsd-pi 2.37.1 → 2.38.0-dev.96dc7fb

package/packages/pi-ai/src/providers/anthropic.ts

@@ -1,40 +1,29 @@
 // Lazy-loaded: Anthropic SDK (~500ms) is imported on first use, not at startup.
 // This avoids penalizing users who don't use Anthropic models.
 import type Anthropic from "@anthropic-ai/sdk";
-import type {
-	ContentBlockParam,
-	MessageCreateParamsStreaming,
-	MessageParam,
-} from "@anthropic-ai/sdk/resources/messages.js";
 import { getEnvApiKey } from "../env-api-keys.js";
-import { calculateCost } from "../models.js";
 import type {
-	Api,
-	AssistantMessage,
-	CacheRetention,
 	Context,
-	ImageContent,
-	Message,
 	Model,
-	ServerToolUseContent,
 	SimpleStreamOptions,
-	StopReason,
 	StreamFunction,
-	StreamOptions,
-	TextContent,
-	ThinkingContent,
-	Tool,
-	ToolCall,
-	ToolResultMessage,
-	WebSearchResultContent,
 } from "../types.js";
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
-import { parseStreamingJson } from "../utils/json-parse.js";
-import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
 
 import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.js";
 import { adjustMaxTokensForThinking, buildBaseOptions } from "./simple-options.js";
-import { transformMessages } from "./transform-messages.js";
+import {
+	type AnthropicEffort,
+	type AnthropicOptions,
+	extractRetryAfterMs,
+	mapThinkingLevelToEffort,
+	processAnthropicStream,
+	supportsAdaptiveThinking,
+} from "./anthropic-shared.js";
+
+// Re-export types used by other modules
+export type { AnthropicEffort, AnthropicOptions };
+export { extractRetryAfterMs };
 
 let _AnthropicClass: typeof Anthropic | undefined;
 async function getAnthropicClass(): Promise<typeof Anthropic> {
@@ -45,154 +34,9 @@ async function getAnthropicClass(): Promise<typeof Anthropic> {
 	return _AnthropicClass;
 }
 
-/**
- * Resolve cache retention preference.
- * Defaults to "short" and uses PI_CACHE_RETENTION for backward compatibility.
- */
-function resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {
-	if (cacheRetention) {
-		return cacheRetention;
-	}
-	if (typeof process !== "undefined" && process.env.PI_CACHE_RETENTION === "long") {
-		return "long";
-	}
-	return "short";
-}
-
-function getCacheControl(
-	baseUrl: string,
-	cacheRetention?: CacheRetention,
-): { retention: CacheRetention; cacheControl?: { type: "ephemeral"; ttl?: "1h" } } {
-	const retention = resolveCacheRetention(cacheRetention);
-	if (retention === "none") {
-		return { retention };
-	}
-	const ttl = retention === "long" && baseUrl.includes("api.anthropic.com") ? "1h" : undefined;
-	return {
-		retention,
-		cacheControl: { type: "ephemeral", ...(ttl && { ttl }) },
-	};
-}
-
 // Stealth mode: Mimic Claude Code's tool naming exactly
 const claudeCodeVersion = "2.1.62";
 
-// Claude Code 2.x tool names (canonical casing)
-// Source: https://cchistory.mariozechner.at/data/prompts-2.1.11.md
-// To update: https://github.com/badlogic/cchistory
-const claudeCodeTools = [
-	"Read",
-	"Write",
-	"Edit",
-	"Bash",
-	"Grep",
-	"Glob",
-	"AskUserQuestion",
-	"EnterPlanMode",
-	"ExitPlanMode",
-	"KillShell",
-	"NotebookEdit",
-	"Skill",
-	"Task",
-	"TaskOutput",
-	"TodoWrite",
-	"WebFetch",
-	"WebSearch",
-];
-
-const ccToolLookup = new Map(claudeCodeTools.map((t) => [t.toLowerCase(), t]));
-
-// Convert tool name to CC canonical casing if it matches (case-insensitive)
-const toClaudeCodeName = (name: string) => ccToolLookup.get(name.toLowerCase()) ?? name;
-const fromClaudeCodeName = (name: string, tools?: Tool[]) => {
-	if (tools && tools.length > 0) {
-		const lowerName = name.toLowerCase();
-		const matchedTool = tools.find((tool) => tool.name.toLowerCase() === lowerName);
-		if (matchedTool) return matchedTool.name;
-	}
-	return name;
-};
-
-/**
- * Convert content blocks to Anthropic API format
- */
-function convertContentBlocks(content: (TextContent | ImageContent)[]):
-	| string
-	| Array<
-			| { type: "text"; text: string }
-			| {
-					type: "image";
-					source: {
-						type: "base64";
-						media_type: "image/jpeg" | "image/png" | "image/gif" | "image/webp";
-						data: string;
-					};
-			  }
-	  > {
-	// If only text blocks, return as concatenated string for simplicity
-	const hasImages = content.some((c) => c.type === "image");
-	if (!hasImages) {
-		return sanitizeSurrogates(content.map((c) => (c as TextContent).text).join("\n"));
-	}
-
-	// If we have images, convert to content block array
-	const blocks = content.map((block) => {
-		if (block.type === "text") {
-			return {
-				type: "text" as const,
-				text: sanitizeSurrogates(block.text),
-			};
-		}
-		return {
-			type: "image" as const,
-			source: {
-				type: "base64" as const,
-				media_type: block.mimeType as "image/jpeg" | "image/png" | "image/gif" | "image/webp",
-				data: block.data,
-			},
-		};
-	});
-
-	// If only images (no text), add placeholder text block
-	const hasText = blocks.some((b) => b.type === "text");
-	if (!hasText) {
-		blocks.unshift({
-			type: "text" as const,
-			text: "(see attached image)",
-		});
-	}
-
-	return blocks;
-}
-
-export type AnthropicEffort = "low" | "medium" | "high" | "max";
-
-export interface AnthropicOptions extends StreamOptions {
-	/**
-	 * Enable extended thinking.
-	 * For Opus 4.6 and Sonnet 4.6: uses adaptive thinking (model decides when/how much to think).
-	 * For older models: uses budget-based thinking with thinkingBudgetTokens.
-	 */
-	thinkingEnabled?: boolean;
-	/**
-	 * Token budget for extended thinking (older models only).
-	 * Ignored for Opus 4.6 and Sonnet 4.6, which use adaptive thinking.
-	 */
-	thinkingBudgetTokens?: number;
-	/**
-	 * Effort level for adaptive thinking (Opus 4.6 and Sonnet 4.6).
-	 * Controls how much thinking Claude allocates:
-	 * - "max": Always thinks with no constraints (Opus 4.6 only)
-	 * - "high": Always thinks, deep reasoning (default)
-	 * - "medium": Moderate thinking, may skip for simple queries
-	 * - "low": Minimal thinking, skips for simple tasks
-	 * Ignored for older models.
-	 */
-	effort?: AnthropicEffort;
-	interleavedThinking?: boolean;
-	toolChoice?: "auto" | "any" | "none" | { type: "tool"; name: string };
-}
-
 function mergeHeaders(...headerSources: (Record<string, string> | undefined)[]): Record<string, string> {
 	const merged: Record<string, string> = {};
 	for (const headers of headerSources) {
@@ -203,410 +47,6 @@ function mergeHeaders(...headerSources: (Record<string, string> | undefined)[]):
 	return merged;
 }
 
-/**
- * Detect transient network errors that are likely to succeed on retry.
- * Covers WebSocket disconnects (Tailscale, VPN), TCP resets, and DNS failures.
- */
-function isTransientNetworkError(error: unknown): boolean {
-  if (!(error instanceof Error)) return false;
-  const msg = error.message.toLowerCase();
-  const code = (error as NodeJS.ErrnoException).code;
-  return (
-    code === 'ECONNRESET' ||
-    code === 'EPIPE' ||
-    code === 'ETIMEDOUT' ||
-    code === 'ENOTFOUND' ||
-    code === 'EAI_AGAIN' ||
-    msg.includes('connector_closed') ||
-    msg.includes('socket hang up') ||
-    msg.includes('network') ||
-    msg.includes('connection') && msg.includes('closed') ||
-    msg.includes('fetch failed')
-  );
-}
-
-/**
- * Extract retry delay from Anthropic error response headers (in milliseconds).
- * Checks: retry-after (seconds or RFC date), x-ratelimit-reset-requests, x-ratelimit-reset-tokens.
- * Returns undefined if no valid delay is found or if the delay is in the past.
- */
-function extractRetryAfterMs(headers: Headers | { get(name: string): string | null }, errorText = ""): number | undefined {
-	const normalizeDelay = (ms: number): number | undefined => (ms > 0 ? Math.ceil(ms + 1000) : undefined);
-
-	const retryAfter = headers.get("retry-after");
-	if (retryAfter) {
-		const seconds = Number(retryAfter);
-		if (Number.isFinite(seconds)) {
-			const delay = normalizeDelay(seconds * 1000);
-			if (delay !== undefined) return delay;
-		}
-		const asDate = new Date(retryAfter).getTime();
-		if (!Number.isNaN(asDate)) {
-			const delay = normalizeDelay(asDate - Date.now());
-			if (delay !== undefined) return delay;
-		}
-	}
-
-	// x-ratelimit-reset-requests / x-ratelimit-reset-tokens are Unix timestamps (seconds)
-	for (const header of ["x-ratelimit-reset-requests", "x-ratelimit-reset-tokens"]) {
-		const value = headers.get(header);
-		if (value) {
-			const resetSeconds = Number(value);
-			if (Number.isFinite(resetSeconds)) {
-				const delay = normalizeDelay(resetSeconds * 1000 - Date.now());
-				if (delay !== undefined) return delay;
-			}
-		}
-	}
-
-	return undefined;
-}
-
-export const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOptions> = (
-	model: Model<"anthropic-messages">,
-	context: Context,
-	options?: AnthropicOptions,
-): AssistantMessageEventStream => {
-	const stream = new AssistantMessageEventStream();
-
-	(async () => {
-		const output: AssistantMessage = {
-			role: "assistant",
-			content: [],
-			api: model.api as Api,
-			provider: model.provider,
-			model: model.id,
-			usage: {
-				input: 0,
-				output: 0,
-				cacheRead: 0,
-				cacheWrite: 0,
-				totalTokens: 0,
-				cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
-			},
-			stopReason: "stop",
-			timestamp: Date.now(),
-		};
-
-		try {
-			const apiKey = options?.apiKey ?? getEnvApiKey(model.provider) ?? "";
-
-			let copilotDynamicHeaders: Record<string, string> | undefined;
-			if (model.provider === "github-copilot") {
-				const hasImages = hasCopilotVisionInput(context.messages);
-				copilotDynamicHeaders = buildCopilotDynamicHeaders({
-					messages: context.messages,
-					hasImages,
-				});
-			}
-
-			const { client, isOAuthToken } = await createClient(
-				model,
-				apiKey,
-				options?.interleavedThinking ?? true,
-				options?.headers,
-				copilotDynamicHeaders,
-			);
-			let params = buildParams(model, context, isOAuthToken, options);
-			const nextParams = await options?.onPayload?.(params, model);
-			if (nextParams !== undefined) {
-				params = nextParams as MessageCreateParamsStreaming;
-			}
-			const anthropicStream = client.messages.stream({ ...params, stream: true }, { signal: options?.signal });
-			stream.push({ type: "start", partial: output });
-
-			type Block = (ThinkingContent | TextContent | (ToolCall & { partialJson: string }) | ServerToolUseContent | WebSearchResultContent) & { index: number };
-			const blocks = output.content as Block[];
-
-			for await (const event of anthropicStream) {
-				if (event.type === "message_start") {
-					// Capture initial token usage from message_start event
-					// This ensures we have input token counts even if the stream is aborted early
-					output.usage.input = event.message.usage.input_tokens || 0;
-					output.usage.output = event.message.usage.output_tokens || 0;
-					output.usage.cacheRead = event.message.usage.cache_read_input_tokens || 0;
-					output.usage.cacheWrite = event.message.usage.cache_creation_input_tokens || 0;
-					// Anthropic doesn't provide total_tokens, compute from components
-					output.usage.totalTokens =
-						output.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;
-					calculateCost(model, output.usage);
-				} else if (event.type === "content_block_start") {
-					if (event.content_block.type === "text") {
-						const block: Block = {
-							type: "text",
-							text: "",
-							index: event.index,
-						};
-						output.content.push(block);
-						stream.push({ type: "text_start", contentIndex: output.content.length - 1, partial: output });
-					} else if (event.content_block.type === "thinking") {
-						const block: Block = {
-							type: "thinking",
-							thinking: "",
-							thinkingSignature: "",
-							index: event.index,
-						};
-						output.content.push(block);
-						stream.push({ type: "thinking_start", contentIndex: output.content.length - 1, partial: output });
-					} else if (event.content_block.type === "redacted_thinking") {
-						const block: Block = {
-							type: "thinking",
-							thinking: "[Reasoning redacted]",
-							thinkingSignature: event.content_block.data,
-							redacted: true,
-							index: event.index,
-						};
-						output.content.push(block);
-						stream.push({ type: "thinking_start", contentIndex: output.content.length - 1, partial: output });
-					} else if (event.content_block.type === "tool_use") {
-						const block: Block = {
-							type: "toolCall",
-							id: event.content_block.id,
-							name: isOAuthToken
-								? fromClaudeCodeName(event.content_block.name, context.tools)
-								: event.content_block.name,
-							arguments: (event.content_block.input as Record<string, any>) ?? {},
-							partialJson: "",
-							index: event.index,
-						};
-						output.content.push(block);
-						stream.push({ type: "toolcall_start", contentIndex: output.content.length - 1, partial: output });
-					} else if ((event.content_block as any).type === "server_tool_use") {
-						const serverBlock = event.content_block as any;
-						const block: Block = {
-							type: "serverToolUse",
-							id: serverBlock.id,
-							name: serverBlock.name,
-							input: serverBlock.input,
-							index: event.index,
-						};
-						output.content.push(block);
-						stream.push({ type: "server_tool_use", contentIndex: output.content.length - 1, partial: output });
-					} else if ((event.content_block as any).type === "web_search_tool_result") {
-						const resultBlock = event.content_block as any;
-						const block: Block = {
-							type: "webSearchResult",
-							toolUseId: resultBlock.tool_use_id,
-							content: resultBlock.content,
-							index: event.index,
-						};
-						output.content.push(block);
-						stream.push({ type: "web_search_result", contentIndex: output.content.length - 1, partial: output });
-					}
-				} else if (event.type === "content_block_delta") {
-					if (event.delta.type === "text_delta") {
-						const index = blocks.findIndex((b) => b.index === event.index);
-						const block = blocks[index];
-						if (block && block.type === "text") {
-							block.text += event.delta.text;
-							stream.push({
-								type: "text_delta",
-								contentIndex: index,
-								delta: event.delta.text,
-								partial: output,
-							});
-						}
-					} else if (event.delta.type === "thinking_delta") {
-						const index = blocks.findIndex((b) => b.index === event.index);
-						const block = blocks[index];
-						if (block && block.type === "thinking") {
-							block.thinking += event.delta.thinking;
-							stream.push({
-								type: "thinking_delta",
-								contentIndex: index,
-								delta: event.delta.thinking,
-								partial: output,
-							});
-						}
-					} else if (event.delta.type === "input_json_delta") {
-						const index = blocks.findIndex((b) => b.index === event.index);
-						const block = blocks[index];
-						if (block && block.type === "toolCall") {
-							block.partialJson += event.delta.partial_json;
-							block.arguments = parseStreamingJson(block.partialJson);
-							stream.push({
-								type: "toolcall_delta",
-								contentIndex: index,
-								delta: event.delta.partial_json,
-								partial: output,
-							});
-						}
-					} else if (event.delta.type === "signature_delta") {
-						const index = blocks.findIndex((b) => b.index === event.index);
-						const block = blocks[index];
-						if (block && block.type === "thinking") {
-							block.thinkingSignature = block.thinkingSignature || "";
-							block.thinkingSignature += event.delta.signature;
-						}
-					}
-				} else if (event.type === "content_block_stop") {
-					const index = blocks.findIndex((b) => b.index === event.index);
-					const block = blocks[index];
-					if (block) {
-						delete (block as any).index;
-						if (block.type === "text") {
-							stream.push({
-								type: "text_end",
-								contentIndex: index,
-								content: block.text,
-								partial: output,
-							});
-						} else if (block.type === "thinking") {
-							stream.push({
-								type: "thinking_end",
-								contentIndex: index,
-								content: block.thinking,
-								partial: output,
-							});
-						} else if (block.type === "toolCall") {
-							block.arguments = parseStreamingJson(block.partialJson);
-							delete (block as any).partialJson;
-							stream.push({
-								type: "toolcall_end",
-								contentIndex: index,
-								toolCall: block,
-								partial: output,
-							});
-						}
-						// serverToolUse and webSearchResult blocks just need index cleanup (already emitted on start)
-					}
-				} else if (event.type === "message_delta") {
-					if (event.delta.stop_reason) {
-						output.stopReason = mapStopReason(event.delta.stop_reason);
-					}
-					// Only update usage fields if present (not null).
-					// Preserves input_tokens from message_start when proxies omit it in message_delta.
-					if (event.usage.input_tokens != null) {
-						output.usage.input = event.usage.input_tokens;
-					}
-					if (event.usage.output_tokens != null) {
-						output.usage.output = event.usage.output_tokens;
-					}
-					if (event.usage.cache_read_input_tokens != null) {
-						output.usage.cacheRead = event.usage.cache_read_input_tokens;
-					}
-					if (event.usage.cache_creation_input_tokens != null) {
-						output.usage.cacheWrite = event.usage.cache_creation_input_tokens;
-					}
-					// Anthropic doesn't provide total_tokens, compute from components
-					output.usage.totalTokens =
-						output.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;
-					calculateCost(model, output.usage);
-				}
-			}
-
-			if (options?.signal?.aborted) {
-				throw new Error("Request was aborted");
-			}
-
-			if (output.stopReason === "aborted" || output.stopReason === "error") {
-				throw new Error("An unknown error occurred");
-			}
-
-			stream.push({ type: "done", reason: output.stopReason, message: output });
-			stream.end();
-		} catch (error) {
-			for (const block of output.content) delete (block as any).index;
-			output.stopReason = options?.signal?.aborted ? "aborted" : "error";
-			output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
-			if (model.provider === "alibaba-coding-plan") {
-				output.errorMessage = `[alibaba-coding-plan] ${output.errorMessage}`;
-			}
-			const AnthropicSdk = _AnthropicClass;
-			if (AnthropicSdk && error instanceof AnthropicSdk.APIError && error.headers) {
-				const retryAfterMs = extractRetryAfterMs(error.headers, error.message);
-				if (retryAfterMs !== undefined) {
-					output.retryAfterMs = retryAfterMs;
-				}
-			}
-			// Mark transient network errors as retriable so auto-mode can
-			// detect them and retry instead of stopping (#833).
-			if (isTransientNetworkError(error)) {
-				output.retryAfterMs = output.retryAfterMs ?? 5000;
-			}
-			stream.push({ type: "error", reason: output.stopReason, error: output });
-			stream.end();
-		}
-	})();
-
-	return stream;
-};
-
-/**
- * Check if a model supports adaptive thinking (Opus 4.6 and Sonnet 4.6)
- */
-function supportsAdaptiveThinking(modelId: string): boolean {
-	// Opus 4.6 and Sonnet 4.6 model IDs (with or without date suffix)
-	return (
-		modelId.includes("opus-4-6") ||
-		modelId.includes("opus-4.6") ||
-		modelId.includes("sonnet-4-6") ||
-		modelId.includes("sonnet-4.6")
-	);
-}
-
-/**
- * Map ThinkingLevel to Anthropic effort levels for adaptive thinking.
- * Note: effort "max" is only valid on Opus 4.6.
- */
-function mapThinkingLevelToEffort(level: SimpleStreamOptions["reasoning"], modelId: string): AnthropicEffort {
-	switch (level) {
-		case "minimal":
-			return "low";
-		case "low":
-			return "low";
-		case "medium":
-			return "medium";
-		case "high":
-			return "high";
-		case "xhigh":
-			return modelId.includes("opus-4-6") || modelId.includes("opus-4.6") ? "max" : "high";
-		default:
-			return "high";
-	}
-}
-
-export const streamSimpleAnthropic: StreamFunction<"anthropic-messages", SimpleStreamOptions> = (
-	model: Model<"anthropic-messages">,
-	context: Context,
-	options?: SimpleStreamOptions,
-): AssistantMessageEventStream => {
-	const apiKey = options?.apiKey || getEnvApiKey(model.provider);
-	if (!apiKey) {
-		throw new Error(`No API key for provider: ${model.provider}`);
-	}
-
-	const base = buildBaseOptions(model, options, apiKey);
-	if (!options?.reasoning) {
-		return streamAnthropic(model, context, { ...base, thinkingEnabled: false } satisfies AnthropicOptions);
-	}
-
-	// For Opus 4.6 and Sonnet 4.6: use adaptive thinking with effort level
-	// For older models: use budget-based thinking
-	if (supportsAdaptiveThinking(model.id)) {
-		const effort = mapThinkingLevelToEffort(options.reasoning, model.id);
-		return streamAnthropic(model, context, {
-			...base,
-			thinkingEnabled: true,
-			effort,
-		} satisfies AnthropicOptions);
-	}
-
-	const adjusted = adjustMaxTokensForThinking(
-		base.maxTokens || 0,
-		model.maxTokens,
-		options.reasoning,
-		options.thinkingBudgets,
-	);
-
-	return streamAnthropic(model, context, {
-		...base,
-		maxTokens: adjusted.maxTokens,
-		thinkingEnabled: true,
-		thinkingBudgetTokens: adjusted.thinkingBudget,
-	} satisfies AnthropicOptions);
-};
-
 function isOAuthToken(apiKey: string): boolean {
 	return apiKey.includes("sk-ant-oat");
 }
@@ -702,315 +142,83 @@ async function createClient(
 	return { client, isOAuthToken: false };
 }
 
-function buildParams(
+export const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOptions> = (
 	model: Model<"anthropic-messages">,
 	context: Context,
-	isOAuthToken: boolean,
 	options?: AnthropicOptions,
-): MessageCreateParamsStreaming {
-	const { cacheControl } = getCacheControl(model.baseUrl, options?.cacheRetention);
-	// Strip variant suffixes like [1m] from model ID before sending to the API.
-	// The API only accepts the base model ID (e.g. "claude-opus-4-6"),
-	// not internal variant identifiers (e.g. "claude-opus-4-6[1m]").
-	// This applies to all auth methods — API keys, OAuth, and Copilot alike.
-	const apiModelId = model.id.replace(/\[.*\]$/, "");
-	const params: MessageCreateParamsStreaming = {
-		model: apiModelId,
-		messages: convertMessages(context.messages, model, isOAuthToken, cacheControl),
-		max_tokens: options?.maxTokens || (model.maxTokens / 3) | 0,
-		stream: true,
-	};
+): AssistantMessageEventStream => {
+	const stream = new AssistantMessageEventStream();
 
-	// For OAuth tokens, we MUST include Claude Code identity
-	if (isOAuthToken) {
-		params.system = [
-			{
-				type: "text",
-				text: "You are Claude Code, Anthropic's official CLI for Claude.",
-				...(cacheControl ? { cache_control: cacheControl } : {}),
-			},
-		];
-		if (context.systemPrompt) {
-			params.system.push({
-				type: "text",
-				text: sanitizeSurrogates(context.systemPrompt),
-				...(cacheControl ? { cache_control: cacheControl } : {}),
+	(async () => {
+		const apiKey = options?.apiKey ?? getEnvApiKey(model.provider) ?? "";
+
+		let copilotDynamicHeaders: Record<string, string> | undefined;
+		if (model.provider === "github-copilot") {
+			const hasImages = hasCopilotVisionInput(context.messages);
+			copilotDynamicHeaders = buildCopilotDynamicHeaders({
+				messages: context.messages,
+				hasImages,
 			});
 		}
-	} else if (context.systemPrompt) {
-		// Add cache control to system prompt for non-OAuth tokens
-		params.system = [
-			{
-				type: "text",
-				text: sanitizeSurrogates(context.systemPrompt),
-				...(cacheControl ? { cache_control: cacheControl } : {}),
-			},
-		];
-	}
-
-	// Temperature is incompatible with extended thinking (adaptive or budget-based).
-	if (options?.temperature !== undefined && !options?.thinkingEnabled) {
-		params.temperature = options.temperature;
-	}
-
-	if (context.tools) {
-		params.tools = convertTools(context.tools, isOAuthToken);
-	}
-
-	// Configure thinking mode: adaptive (Opus 4.6 and Sonnet 4.6) or budget-based (older models)
-	if (options?.thinkingEnabled && model.reasoning) {
-		if (supportsAdaptiveThinking(model.id)) {
-			// Adaptive thinking: Claude decides when and how much to think
-			params.thinking = { type: "adaptive" };
-			if (options.effort) {
-				params.output_config = { effort: options.effort };
-			}
-		} else {
-			// Budget-based thinking for older models
-			params.thinking = {
-				type: "enabled",
-				budget_tokens: options.thinkingBudgetTokens || 1024,
-			};
-		}
-	}
-
-	if (options?.metadata) {
-		const userId = options.metadata.user_id;
-		if (typeof userId === "string") {
-			params.metadata = { user_id: userId };
-		}
-	}
-
-	if (options?.toolChoice) {
-		if (typeof options.toolChoice === "string") {
-			params.tool_choice = { type: options.toolChoice };
-		} else {
-			params.tool_choice = options.toolChoice;
-		}
-	}
 
-	return params;
-}
+		const { client, isOAuthToken: isOAuth } = await createClient(
+			model,
+			apiKey,
+			options?.interleavedThinking ?? true,
+			options?.headers,
+			copilotDynamicHeaders,
+		);
+
+		processAnthropicStream(stream, {
+			client,
+			model,
+			context,
+			isOAuthToken: isOAuth,
+			options,
+			AnthropicSdkClass: _AnthropicClass,
+		});
+	})();
 
-// Normalize tool call IDs to match Anthropic's required pattern and length
-function normalizeToolCallId(id: string): string {
-	return id.replace(/[^a-zA-Z0-9_-]/g, "_").slice(0, 64);
-}
+	return stream;
+};
 
-function convertMessages(
-	messages: Message[],
+export const streamSimpleAnthropic: StreamFunction<"anthropic-messages", SimpleStreamOptions> = (
 	model: Model<"anthropic-messages">,
-	isOAuthToken: boolean,
-	cacheControl?: { type: "ephemeral"; ttl?: "1h" },
-): MessageParam[] {
-	const params: MessageParam[] = [];
-
-	// Transform messages for cross-provider compatibility
-	const transformedMessages = transformMessages(messages, model, normalizeToolCallId);
-
-	for (let i = 0; i < transformedMessages.length; i++) {
-		const msg = transformedMessages[i];
-
-		if (msg.role === "user") {
-			if (typeof msg.content === "string") {
-				if (msg.content.trim().length > 0) {
-					params.push({
-						role: "user",
-						content: sanitizeSurrogates(msg.content),
-					});
-				}
-			} else {
-				const blocks: ContentBlockParam[] = msg.content.map((item) => {
-					if (item.type === "text") {
-						return {
-							type: "text",
-							text: sanitizeSurrogates(item.text),
-						};
-					} else {
-						return {
-							type: "image",
-							source: {
-								type: "base64",
-								media_type: item.mimeType as "image/jpeg" | "image/png" | "image/gif" | "image/webp",
-								data: item.data,
-							},
-						};
-					}
-				});
-				let filteredBlocks = !model?.input.includes("image") ? blocks.filter((b) => b.type !== "image") : blocks;
-				filteredBlocks = filteredBlocks.filter((b) => {
-					if (b.type === "text") {
-						return b.text.trim().length > 0;
-					}
-					return true;
-				});
-				if (filteredBlocks.length === 0) continue;
-				params.push({
-					role: "user",
-					content: filteredBlocks,
-				});
-			}
-		} else if (msg.role === "assistant") {
-			const blocks: ContentBlockParam[] = [];
-
-			for (const block of msg.content) {
-				if (block.type === "text") {
-					if (block.text.trim().length === 0) continue;
-					blocks.push({
-						type: "text",
-						text: sanitizeSurrogates(block.text),
-					});
-				} else if (block.type === "thinking") {
-					// Redacted thinking: pass the opaque payload back as redacted_thinking
-					if (block.redacted) {
-						blocks.push({
-							type: "redacted_thinking",
-							data: block.thinkingSignature!,
-						});
-						continue;
-					}
-					if (block.thinking.trim().length === 0) continue;
-					// If thinking signature is missing/empty (e.g., from aborted stream),
-					// convert to plain text block without <thinking> tags to avoid API rejection
-					// and prevent Claude from mimicking the tags in responses
-					if (!block.thinkingSignature || block.thinkingSignature.trim().length === 0) {
-						blocks.push({
-							type: "text",
-							text: sanitizeSurrogates(block.thinking),
-						});
-					} else {
-						blocks.push({
-							type: "thinking",
-							thinking: sanitizeSurrogates(block.thinking),
-							signature: block.thinkingSignature,
-						});
-					}
-				} else if (block.type === "toolCall") {
-					blocks.push({
-						type: "tool_use",
-						id: block.id,
-						name: isOAuthToken ? toClaudeCodeName(block.name) : block.name,
-						input: block.arguments ?? {},
-					});
-				} else if (block.type === "serverToolUse") {
-					blocks.push({
-						type: "server_tool_use",
-						id: block.id,
-						name: block.name,
-						input: block.input ?? {},
-					} as any);
-				} else if (block.type === "webSearchResult") {
-					blocks.push({
-						type: "web_search_tool_result",
-						tool_use_id: block.toolUseId,
-						content: block.content,
-					} as any);
-				}
-			}
-			if (blocks.length === 0) continue;
-			params.push({
-				role: "assistant",
-				content: blocks,
-			});
-		} else if (msg.role === "toolResult") {
-			// Collect all consecutive toolResult messages, needed for z.ai Anthropic endpoint
-			const toolResults: ContentBlockParam[] = [];
-
-			// Add the current tool result
-			toolResults.push({
-				type: "tool_result",
-				tool_use_id: msg.toolCallId,
-				content: convertContentBlocks(msg.content),
-				is_error: msg.isError,
-			});
-
-			// Look ahead for consecutive toolResult messages
-			let j = i + 1;
-			while (j < transformedMessages.length && transformedMessages[j].role === "toolResult") {
-				const nextMsg = transformedMessages[j] as ToolResultMessage; // We know it's a toolResult
-				toolResults.push({
-					type: "tool_result",
-					tool_use_id: nextMsg.toolCallId,
-					content: convertContentBlocks(nextMsg.content),
-					is_error: nextMsg.isError,
-				});
-				j++;
-			}
-
-			// Skip the messages we've already processed
-			i = j - 1;
-
-			// Add a single user message with all tool results
-			params.push({
-				role: "user",
-				content: toolResults,
-			});
-		}
+	context: Context,
+	options?: SimpleStreamOptions,
+): AssistantMessageEventStream => {
+	const apiKey = options?.apiKey || getEnvApiKey(model.provider);
+	if (!apiKey) {
+		throw new Error(`No API key for provider: ${model.provider}`);
 	}
 
-	// Add cache_control to the last user message to cache conversation history
-	if (cacheControl && params.length > 0) {
-		const lastMessage = params[params.length - 1];
-		if (lastMessage.role === "user") {
-			if (Array.isArray(lastMessage.content)) {
-				const lastBlock = lastMessage.content[lastMessage.content.length - 1];
-				if (
-					lastBlock &&
-					(lastBlock.type === "text" || lastBlock.type === "image" || lastBlock.type === "tool_result")
-				) {
-					(lastBlock as any).cache_control = cacheControl;
-				}
-			} else if (typeof lastMessage.content === "string") {
-				lastMessage.content = [
-					{
-						type: "text",
-						text: lastMessage.content,
-						cache_control: cacheControl,
-					},
-				] as any;
-			}
-		}
+	const base = buildBaseOptions(model, options, apiKey);
+	if (!options?.reasoning) {
+		return streamAnthropic(model, context, { ...base, thinkingEnabled: false } satisfies AnthropicOptions);
 	}
 
-	return params;
-}
-
-function convertTools(tools: Tool[], isOAuthToken: boolean): Anthropic.Messages.Tool[] {
-	if (!tools) return [];
-
-	return tools.map((tool) => {
-		const jsonSchema = tool.parameters as any; // TypeBox already generates JSON Schema
+	// For Opus 4.6 and Sonnet 4.6: use adaptive thinking with effort level
+	// For older models: use budget-based thinking
+	if (supportsAdaptiveThinking(model.id)) {
+		const effort = mapThinkingLevelToEffort(options.reasoning, model.id);
+		return streamAnthropic(model, context, {
+			...base,
+			thinkingEnabled: true,
+			effort,
+		} satisfies AnthropicOptions);
+	}
 
-		return {
-			name: isOAuthToken ? toClaudeCodeName(tool.name) : tool.name,
-			description: tool.description,
-			input_schema: {
-				type: "object" as const,
-				properties: jsonSchema.properties || {},
-				required: jsonSchema.required || [],
-			},
-		};
-	});
-}
+	const adjusted = adjustMaxTokensForThinking(
+		base.maxTokens || 0,
+		model.maxTokens,
+		options.reasoning,
+		options.thinkingBudgets,
+	);
 
-function mapStopReason(reason: Anthropic.Messages.StopReason | string): StopReason {
-	switch (reason) {
-		case "end_turn":
-			return "stop";
-		case "max_tokens":
-			return "length";
-		case "tool_use":
-			return "toolUse";
-		case "refusal":
-			return "error";
-		case "pause_turn": // Stop is good enough -> resubmit
-			return "stop";
-		case "stop_sequence":
-			return "stop"; // We don't supply stop sequences, so this should never happen
-		case "sensitive": // Content flagged by safety filters (not yet in SDK types)
-			return "error";
-		default:
-			// Handle unknown stop reasons gracefully (API may add new values)
-			throw new Error(`Unhandled stop reason: ${reason}`);
-	}
-}
+	return streamAnthropic(model, context, {
+		...base,
+		maxTokens: adjusted.maxTokens,
+		thinkingEnabled: true,
+		thinkingBudgetTokens: adjusted.thinkingBudget,
+	} satisfies AnthropicOptions);
+};