npm - @dianshuv/copilot-api - Versions diffs - 0.2.0 → 0.2.2 - Mend

@dianshuv/copilot-api 0.2.0 → 0.2.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (2) hide show

package/dist/main.mjs +396 -363
package/package.json +2 -2

package/dist/main.mjs CHANGED Viewed

@@ -1017,7 +1017,7 @@ const patchClaude = defineCommand({
 //#endregion
 //#region package.json
-var version = "0.2.0";
+var version = "0.2.2";
 //#endregion
 //#region src/lib/adaptive-rate-limiter.ts
@@ -1406,7 +1406,7 @@ function recordResponse(id, response, durationMs) {
 	}
 }
 function getHistory(options = {}) {
-	const { page = 1, limit = 50, model, endpoint, success, from, to, search, sessionId } = options;
+	const { page = 1, limit = 50, model, endpoint, status, from, to, search, sessionId } = options;
 	let filtered = [...historyState.entries];
 	if (sessionId) filtered = filtered.filter((e) => e.sessionId === sessionId);
 	if (model) {
@@ -1414,7 +1414,21 @@ function getHistory(options = {}) {
 		filtered = filtered.filter((e) => e.request.model.toLowerCase().includes(modelLower) || e.response?.model.toLowerCase().includes(modelLower));
 	}
 	if (endpoint) filtered = filtered.filter((e) => e.endpoint === endpoint);
-	if (success !== void 0) filtered = filtered.filter((e) => e.response?.success === success);
+	let effectiveStatus = status;
+	const legacySuccess = options.success;
+	if (!effectiveStatus && legacySuccess !== void 0) effectiveStatus = legacySuccess ? "success" : "error";
+	switch (effectiveStatus) {
+		case "success":
+			filtered = filtered.filter((e) => e.response?.success === true);
+			break;
+		case "error":
+			filtered = filtered.filter((e) => e.response !== void 0 && !e.response.success);
+			break;
+		case "pending":
+			filtered = filtered.filter((e) => !e.response);
+			break;
+		default: break;
+	}
 	if (from) filtered = filtered.filter((e) => e.timestamp >= from);
 	if (to) filtered = filtered.filter((e) => e.timestamp <= to);
 	if (search) {
@@ -3278,6 +3292,11 @@ function handleGetEntries(c) {
 		model: query.model || void 0,
 		endpoint: query.endpoint,
 		success: query.success ? query.success === "true" : void 0,
+		status: [
+			"success",
+			"error",
+			"pending"
+		].includes(query.status) ? query.status : void 0,
 		from: query.from ? Number.parseInt(query.from, 10) : void 0,
 		to: query.to ? Number.parseInt(query.to, 10) : void 0,
 		search: query.search || void 0,
@@ -3554,11 +3573,11 @@ async function loadEntries() {
   if (currentSessionId) params.set('sessionId', currentSessionId);
   const endpoint = document.getElementById('filter-endpoint').value;
-  const success = document.getElementById('filter-success').value;
+  const status = document.getElementById('filter-status').value;
   const search = document.getElementById('filter-search').value;
   if (endpoint) params.set('endpoint', endpoint);
-  if (success) params.set('success', success);
+  if (status) params.set('status', status);
   if (search) params.set('search', search);
   try {
@@ -4311,10 +4330,11 @@ const template = `
         <option value="anthropic">Anthropic</option>
         <option value="openai">OpenAI</option>
       </select>
-      <select id="filter-success" onchange="loadEntries()">
+      <select id="filter-status" onchange="loadEntries()">
         <option value="">All Status</option>
-        <option value="true">Success</option>
-        <option value="false">Failed</option>
+        <option value="success">Success</option>
+        <option value="error">Failed</option>
+        <option value="pending">Pending</option>
       </select>
     </div>
@@ -4361,6 +4381,7 @@ function getHistoryUI() {
   <meta charset="UTF-8">
   <meta name="viewport" content="width=device-width, initial-scale=1.0">
   <title>Copilot API - Request History</title>
+  <link rel="icon" href="data:,">
   <style>${styles}</style>
 </head>
 <body>
@@ -4907,6 +4928,211 @@ async function checkNeedsCompactionAnthropic(payload, model, config = {}) {
 	};
 }
+//#endregion
+//#region src/services/copilot/create-anthropic-messages.ts
+/**
+* Direct Anthropic-style message API for Copilot.
+* Used when the model vendor is Anthropic and supports /v1/messages endpoint.
+*/
+/**
+* Fields that are supported by Copilot's Anthropic API endpoint.
+* Any other fields in the incoming request will be stripped.
+*/
+const COPILOT_SUPPORTED_FIELDS = new Set([
+	"model",
+	"messages",
+	"max_tokens",
+	"system",
+	"metadata",
+	"stop_sequences",
+	"stream",
+	"temperature",
+	"top_p",
+	"top_k",
+	"tools",
+	"tool_choice",
+	"thinking",
+	"service_tier"
+]);
+/**
+* Filter payload to only include fields supported by Copilot's Anthropic API.
+* This prevents errors like "Extra inputs are not permitted" for unsupported
+* fields like `output_config`.
+*
+* Also converts server-side tools (web_search, etc.) to custom tools.
+*/
+function filterPayloadForCopilot(payload) {
+	const filtered = {};
+	const unsupportedFields = [];
+	for (const [key, value] of Object.entries(payload)) if (COPILOT_SUPPORTED_FIELDS.has(key)) filtered[key] = value;
+	else unsupportedFields.push(key);
+	if (unsupportedFields.length > 0) consola.debug(`[DirectAnthropic] Filtered unsupported fields: ${unsupportedFields.join(", ")}`);
+	if (filtered.tools) filtered.tools = convertServerToolsToCustom(filtered.tools);
+	return filtered;
+}
+/**
+* Adjust max_tokens if thinking is enabled.
+* According to Anthropic docs, max_tokens must be greater than thinking.budget_tokens.
+* max_tokens = thinking_budget + response_tokens
+*/
+function adjustMaxTokensForThinking(payload) {
+	const thinking = payload.thinking;
+	if (!thinking) return payload;
+	const budgetTokens = thinking.budget_tokens;
+	if (!budgetTokens) return payload;
+	if (payload.max_tokens <= budgetTokens) {
+		const newMaxTokens = budgetTokens + Math.min(16384, budgetTokens);
+		consola.debug(`[DirectAnthropic] Adjusted max_tokens: ${payload.max_tokens} → ${newMaxTokens} (thinking.budget_tokens=${budgetTokens})`);
+		return {
+			...payload,
+			max_tokens: newMaxTokens
+		};
+	}
+	return payload;
+}
+/**
+* Create messages using Anthropic-style API directly.
+* This bypasses the OpenAI translation layer for Anthropic models.
+*/
+async function createAnthropicMessages(payload, options) {
+	if (!state.copilotToken) throw new Error("Copilot token not found");
+	let filteredPayload = filterPayloadForCopilot(payload);
+	filteredPayload = adjustMaxTokensForThinking(filteredPayload);
+	const enableVision = filteredPayload.messages.some((msg) => {
+		if (typeof msg.content === "string") return false;
+		return msg.content.some((block) => block.type === "image");
+	});
+	const isAgentCall = filteredPayload.messages.some((msg) => msg.role === "assistant");
+	const headers = {
+		...copilotHeaders(state, enableVision),
+		"X-Initiator": options?.initiator ?? (isAgentCall ? "agent" : "user"),
+		"anthropic-version": "2023-06-01"
+	};
+	consola.debug("Sending direct Anthropic request to Copilot /v1/messages");
+	const response = await fetch(`${copilotBaseUrl(state)}/v1/messages`, {
+		method: "POST",
+		headers,
+		body: JSON.stringify(filteredPayload)
+	});
+	if (!response.ok) {
+		consola.debug("Request failed:", {
+			model: filteredPayload.model,
+			max_tokens: filteredPayload.max_tokens,
+			stream: filteredPayload.stream,
+			tools: filteredPayload.tools?.map((t) => ({
+				name: t.name,
+				type: t.type
+			})),
+			thinking: filteredPayload.thinking,
+			messageCount: filteredPayload.messages.length
+		});
+		throw await HTTPError.fromResponse("Failed to create Anthropic messages", response, filteredPayload.model);
+	}
+	if (payload.stream) return events(response);
+	return await response.json();
+}
+const SERVER_TOOL_CONFIGS = {
+	web_search: {
+		description: "Search the web for current information. Returns web search results that can help answer questions about recent events, current data, or information that may have changed since your knowledge cutoff.",
+		input_schema: {
+			type: "object",
+			properties: { query: {
+				type: "string",
+				description: "The search query"
+			} },
+			required: ["query"]
+		}
+	},
+	web_fetch: {
+		description: "Fetch content from a URL. NOTE: This is a client-side tool - the client must fetch the URL and return the content.",
+		input_schema: {
+			type: "object",
+			properties: { url: {
+				type: "string",
+				description: "The URL to fetch"
+			} },
+			required: ["url"]
+		}
+	},
+	code_execution: {
+		description: "Execute code in a sandbox. NOTE: This is a client-side tool - the client must execute the code.",
+		input_schema: {
+			type: "object",
+			properties: {
+				code: {
+					type: "string",
+					description: "The code to execute"
+				},
+				language: {
+					type: "string",
+					description: "The programming language"
+				}
+			},
+			required: ["code"]
+		}
+	},
+	computer: {
+		description: "Control computer desktop. NOTE: This is a client-side tool - the client must handle computer control.",
+		input_schema: {
+			type: "object",
+			properties: { action: {
+				type: "string",
+				description: "The action to perform"
+			} },
+			required: ["action"]
+		}
+	}
+};
+/**
+* Check if a tool is a server-side tool that needs conversion.
+*/
+function getServerToolPrefix(tool) {
+	if (tool.type) {
+		for (const prefix of Object.keys(SERVER_TOOL_CONFIGS)) if (tool.type.startsWith(prefix)) return prefix;
+	}
+	return null;
+}
+/**
+* Convert server-side tools to custom tools, or pass them through unchanged.
+* This allows them to be passed to the API and handled by the client.
+*
+* Note: Server-side tools are only converted if state.rewriteAnthropicTools is enabled.
+*/
+function convertServerToolsToCustom(tools) {
+	if (!tools) return;
+	const result = [];
+	for (const tool of tools) {
+		const serverToolPrefix = getServerToolPrefix(tool);
+		if (serverToolPrefix) {
+			const config = SERVER_TOOL_CONFIGS[serverToolPrefix];
+			if (!state.rewriteAnthropicTools) {
+				consola.debug(`[DirectAnthropic] Passing ${serverToolPrefix} through unchanged (use --rewrite-anthropic-tools to convert)`);
+				result.push(tool);
+				continue;
+			}
+			if (config.remove) {
+				consola.warn(`[DirectAnthropic] Removing unsupported server tool: ${tool.name}. Reason: ${config.removalReason}`);
+				continue;
+			}
+			consola.debug(`[DirectAnthropic] Converting server tool to custom: ${tool.name} (type: ${tool.type})`);
+			result.push({
+				name: tool.name,
+				description: config.description,
+				input_schema: config.input_schema
+			});
+		} else result.push(tool);
+	}
+	return result.length > 0 ? result : void 0;
+}
+/**
+* Check if a model supports direct Anthropic API.
+* Returns true if redirect is disabled (direct API is on) and the model is from Anthropic vendor.
+*/
+function supportsDirectAnthropicApi(modelId) {
+	if (state.redirectAnthropic) return false;
+	return (state.models?.data.find((m) => m.id === modelId))?.vendor === "Anthropic";
+}
 //#endregion
 //#region src/routes/messages/message-utils.ts
 function convertAnthropicMessages(messages) {
@@ -4976,51 +5202,106 @@ function mapOpenAIStopReasonToAnthropic(finishReason) {
 }
 //#endregion
-//#region src/routes/messages/non-stream-translation.ts
-const OPENAI_TOOL_NAME_LIMIT = 64;
-/**
-* Ensure all tool_use blocks have corresponding tool_result responses.
-* This handles edge cases where conversation history may be incomplete:
-* - Session interruptions where tool execution was cut off
-* - Previous request failures
-* - Client sending truncated history
-*
-* Adding placeholder responses prevents API errors and maintains protocol compliance.
-*/
-function fixMessageSequence(messages) {
-	const fixedMessages = [];
-	for (let i = 0; i < messages.length; i++) {
-		const message = messages[i];
-		fixedMessages.push(message);
-		if (message.role === "assistant" && message.tool_calls && message.tool_calls.length > 0) {
-			const foundToolResponses = /* @__PURE__ */ new Set();
-			let j = i + 1;
-			while (j < messages.length && messages[j].role === "tool") {
-				const toolMessage = messages[j];
-				if (toolMessage.tool_call_id) foundToolResponses.add(toolMessage.tool_call_id);
-				j++;
-			}
-			for (const toolCall of message.tool_calls) if (!foundToolResponses.has(toolCall.id)) {
-				consola.debug(`Adding placeholder tool_result for ${toolCall.id}`);
-				fixedMessages.push({
-					role: "tool",
-					tool_call_id: toolCall.id,
-					content: "Tool execution was interrupted or failed."
-				});
-			}
-		}
-	}
-	return fixedMessages;
-}
-function translateToOpenAI(payload) {
-	const toolNameMapping = {
-		truncatedToOriginal: /* @__PURE__ */ new Map(),
-		originalToTruncated: /* @__PURE__ */ new Map()
-	};
-	const messages = translateAnthropicMessagesToOpenAI(payload.messages, payload.system, toolNameMapping);
+//#region src/routes/messages/stream-accumulator.ts
+function createAnthropicStreamAccumulator() {
 	return {
-		payload: {
-			model: translateModelName(payload.model),
+		model: "",
+		inputTokens: 0,
+		outputTokens: 0,
+		stopReason: "",
+		content: "",
+		toolCalls: [],
+		currentToolCall: null
+	};
+}
+function processAnthropicEvent(event, acc) {
+	switch (event.type) {
+		case "content_block_delta":
+			handleContentBlockDelta(event.delta, acc);
+			break;
+		case "content_block_start":
+			handleContentBlockStart(event.content_block, acc);
+			break;
+		case "content_block_stop":
+			handleContentBlockStop(acc);
+			break;
+		case "message_delta":
+			handleMessageDelta(event.delta, event.usage, acc);
+			break;
+		default: break;
+	}
+}
+function handleContentBlockDelta(delta, acc) {
+	if (delta.type === "text_delta") acc.content += delta.text;
+	else if (delta.type === "input_json_delta" && acc.currentToolCall) acc.currentToolCall.input += delta.partial_json;
+}
+function handleContentBlockStart(block, acc) {
+	if (block.type === "tool_use") acc.currentToolCall = {
+		id: block.id,
+		name: block.name,
+		input: ""
+	};
+}
+function handleContentBlockStop(acc) {
+	if (acc.currentToolCall) {
+		acc.toolCalls.push(acc.currentToolCall);
+		acc.currentToolCall = null;
+	}
+}
+function handleMessageDelta(delta, usage, acc) {
+	if (delta.stop_reason) acc.stopReason = delta.stop_reason;
+	if (usage) {
+		acc.inputTokens = usage.input_tokens ?? 0;
+		acc.outputTokens = usage.output_tokens;
+	}
+}
+//#endregion
+//#region src/routes/messages/non-stream-translation.ts
+const OPENAI_TOOL_NAME_LIMIT = 64;
+/**
+* Ensure all tool_use blocks have corresponding tool_result responses.
+* This handles edge cases where conversation history may be incomplete:
+* - Session interruptions where tool execution was cut off
+* - Previous request failures
+* - Client sending truncated history
+*
+* Adding placeholder responses prevents API errors and maintains protocol compliance.
+*/
+function fixMessageSequence(messages) {
+	const fixedMessages = [];
+	for (let i = 0; i < messages.length; i++) {
+		const message = messages[i];
+		fixedMessages.push(message);
+		if (message.role === "assistant" && message.tool_calls && message.tool_calls.length > 0) {
+			const foundToolResponses = /* @__PURE__ */ new Set();
+			let j = i + 1;
+			while (j < messages.length && messages[j].role === "tool") {
+				const toolMessage = messages[j];
+				if (toolMessage.tool_call_id) foundToolResponses.add(toolMessage.tool_call_id);
+				j++;
+			}
+			for (const toolCall of message.tool_calls) if (!foundToolResponses.has(toolCall.id)) {
+				consola.debug(`Adding placeholder tool_result for ${toolCall.id}`);
+				fixedMessages.push({
+					role: "tool",
+					tool_call_id: toolCall.id,
+					content: "Tool execution was interrupted or failed."
+				});
+			}
+		}
+	}
+	return fixedMessages;
+}
+function translateToOpenAI(payload) {
+	const toolNameMapping = {
+		truncatedToOriginal: /* @__PURE__ */ new Map(),
+		originalToTruncated: /* @__PURE__ */ new Map()
+	};
+	const messages = translateAnthropicMessagesToOpenAI(payload.messages, payload.system, toolNameMapping);
+	return {
+		payload: {
+			model: translateModelName(payload.model),
 			messages: fixMessageSequence(messages),
 			max_tokens: payload.max_tokens,
 			stop: payload.stop_sequences,
@@ -5073,7 +5354,8 @@ function translateModelName(model) {
 	}
 	if (/^claude-sonnet-4-5-\d+$/.test(model)) return "claude-sonnet-4.5";
 	if (/^claude-sonnet-4-\d+$/.test(model)) return "claude-sonnet-4";
-	if (/^claude-opus-4-6$/.test(model)) return findLatestModel("claude-opus-4.6", "claude-opus-4.6");
+	if (model === "claude-opus-4-6-1m") return "claude-opus-4.6-1m";
+	if (/^claude-opus-4-6$/.test(model)) return "claude-opus-4.6";
 	if (/^claude-opus-4-5-\d+$/.test(model)) return "claude-opus-4.5";
 	if (/^claude-opus-4-\d+$/.test(model)) return findLatestModel("claude-opus", "claude-opus-4.5");
 	if (/^claude-haiku-4-5-\d+$/.test(model)) return "claude-haiku-4.5";
@@ -5304,316 +5586,6 @@ function getAnthropicToolUseBlocks(toolCalls, toolNameMapping) {
 	});
 }
-//#endregion
-//#region src/routes/messages/count-tokens-handler.ts
-/**
-* Handles token counting for Anthropic messages.
-*
-* For Anthropic models (vendor === "Anthropic"), uses the official Anthropic tokenizer.
-* For other models, uses GPT tokenizers with appropriate buffers.
-*
-* When auto-truncate is enabled and the request would exceed limits,
-* returns an inflated token count to trigger Claude Code's auto-compact mechanism.
-*/
-async function handleCountTokens(c) {
-	try {
-		const anthropicBeta = c.req.header("anthropic-beta");
-		const anthropicPayload = await c.req.json();
-		const { payload: openAIPayload } = translateToOpenAI(anthropicPayload);
-		const selectedModel = state.models?.data.find((model) => model.id === openAIPayload.model);
-		if (!selectedModel) {
-			consola.warn("Model not found, returning default token count");
-			return c.json({ input_tokens: 1 });
-		}
-		if (state.autoTruncate) {
-			const truncateCheck = await checkNeedsCompactionAnthropic(anthropicPayload, selectedModel);
-			if (truncateCheck.needed) {
-				const contextWindow = selectedModel.capabilities?.limits?.max_context_window_tokens ?? 2e5;
-				const inflatedTokens = Math.floor(contextWindow * .95);
-				consola.debug(`[count_tokens] Would trigger auto-truncate: ${truncateCheck.currentTokens} tokens > ${truncateCheck.tokenLimit}, returning inflated count: ${inflatedTokens}`);
-				return c.json({ input_tokens: inflatedTokens });
-			}
-		}
-		const tokenizerName = selectedModel.capabilities?.tokenizer ?? "o200k_base";
-		const tokenCount = await getTokenCount(openAIPayload, selectedModel);
-		if (anthropicPayload.tools && anthropicPayload.tools.length > 0) {
-			let mcpToolExist = false;
-			if (anthropicBeta?.startsWith("claude-code")) mcpToolExist = anthropicPayload.tools.some((tool) => tool.name.startsWith("mcp__"));
-			if (!mcpToolExist) {
-				if (anthropicPayload.model.startsWith("claude")) tokenCount.input = tokenCount.input + 346;
-				else if (anthropicPayload.model.startsWith("grok")) tokenCount.input = tokenCount.input + 480;
-			}
-		}
-		let finalTokenCount = tokenCount.input + tokenCount.output;
-		if (!(selectedModel.vendor === "Anthropic")) finalTokenCount = anthropicPayload.model.startsWith("grok") ? Math.round(finalTokenCount * 1.03) : Math.round(finalTokenCount * 1.05);
-		consola.debug(`Token count: ${finalTokenCount} (tokenizer: ${tokenizerName})`);
-		return c.json({ input_tokens: finalTokenCount });
-	} catch (error) {
-		consola.error("Error counting tokens:", error);
-		return c.json({ input_tokens: 1 });
-	}
-}
-//#endregion
-//#region src/services/copilot/create-anthropic-messages.ts
-/**
-* Direct Anthropic-style message API for Copilot.
-* Used when the model vendor is Anthropic and supports /v1/messages endpoint.
-*/
-/**
-* Fields that are supported by Copilot's Anthropic API endpoint.
-* Any other fields in the incoming request will be stripped.
-*/
-const COPILOT_SUPPORTED_FIELDS = new Set([
-	"model",
-	"messages",
-	"max_tokens",
-	"system",
-	"metadata",
-	"stop_sequences",
-	"stream",
-	"temperature",
-	"top_p",
-	"top_k",
-	"tools",
-	"tool_choice",
-	"thinking",
-	"service_tier"
-]);
-/**
-* Filter payload to only include fields supported by Copilot's Anthropic API.
-* This prevents errors like "Extra inputs are not permitted" for unsupported
-* fields like `output_config`.
-*
-* Also converts server-side tools (web_search, etc.) to custom tools.
-*/
-function filterPayloadForCopilot(payload) {
-	const filtered = {};
-	const unsupportedFields = [];
-	for (const [key, value] of Object.entries(payload)) if (COPILOT_SUPPORTED_FIELDS.has(key)) filtered[key] = value;
-	else unsupportedFields.push(key);
-	if (unsupportedFields.length > 0) consola.debug(`[DirectAnthropic] Filtered unsupported fields: ${unsupportedFields.join(", ")}`);
-	if (filtered.tools) filtered.tools = convertServerToolsToCustom(filtered.tools);
-	return filtered;
-}
-/**
-* Adjust max_tokens if thinking is enabled.
-* According to Anthropic docs, max_tokens must be greater than thinking.budget_tokens.
-* max_tokens = thinking_budget + response_tokens
-*/
-function adjustMaxTokensForThinking(payload) {
-	const thinking = payload.thinking;
-	if (!thinking) return payload;
-	const budgetTokens = thinking.budget_tokens;
-	if (!budgetTokens) return payload;
-	if (payload.max_tokens <= budgetTokens) {
-		const newMaxTokens = budgetTokens + Math.min(16384, budgetTokens);
-		consola.debug(`[DirectAnthropic] Adjusted max_tokens: ${payload.max_tokens} → ${newMaxTokens} (thinking.budget_tokens=${budgetTokens})`);
-		return {
-			...payload,
-			max_tokens: newMaxTokens
-		};
-	}
-	return payload;
-}
-/**
-* Create messages using Anthropic-style API directly.
-* This bypasses the OpenAI translation layer for Anthropic models.
-*/
-async function createAnthropicMessages(payload, options) {
-	if (!state.copilotToken) throw new Error("Copilot token not found");
-	let filteredPayload = filterPayloadForCopilot(payload);
-	filteredPayload = adjustMaxTokensForThinking(filteredPayload);
-	const enableVision = filteredPayload.messages.some((msg) => {
-		if (typeof msg.content === "string") return false;
-		return msg.content.some((block) => block.type === "image");
-	});
-	const isAgentCall = filteredPayload.messages.some((msg) => msg.role === "assistant");
-	const headers = {
-		...copilotHeaders(state, enableVision),
-		"X-Initiator": options?.initiator ?? (isAgentCall ? "agent" : "user"),
-		"anthropic-version": "2023-06-01"
-	};
-	consola.debug("Sending direct Anthropic request to Copilot /v1/messages");
-	const response = await fetch(`${copilotBaseUrl(state)}/v1/messages`, {
-		method: "POST",
-		headers,
-		body: JSON.stringify(filteredPayload)
-	});
-	if (!response.ok) {
-		consola.debug("Request failed:", {
-			model: filteredPayload.model,
-			max_tokens: filteredPayload.max_tokens,
-			stream: filteredPayload.stream,
-			tools: filteredPayload.tools?.map((t) => ({
-				name: t.name,
-				type: t.type
-			})),
-			thinking: filteredPayload.thinking,
-			messageCount: filteredPayload.messages.length
-		});
-		throw await HTTPError.fromResponse("Failed to create Anthropic messages", response, filteredPayload.model);
-	}
-	if (payload.stream) return events(response);
-	return await response.json();
-}
-const SERVER_TOOL_CONFIGS = {
-	web_search: {
-		description: "Search the web for current information. Returns web search results that can help answer questions about recent events, current data, or information that may have changed since your knowledge cutoff.",
-		input_schema: {
-			type: "object",
-			properties: { query: {
-				type: "string",
-				description: "The search query"
-			} },
-			required: ["query"]
-		}
-	},
-	web_fetch: {
-		description: "Fetch content from a URL. NOTE: This is a client-side tool - the client must fetch the URL and return the content.",
-		input_schema: {
-			type: "object",
-			properties: { url: {
-				type: "string",
-				description: "The URL to fetch"
-			} },
-			required: ["url"]
-		}
-	},
-	code_execution: {
-		description: "Execute code in a sandbox. NOTE: This is a client-side tool - the client must execute the code.",
-		input_schema: {
-			type: "object",
-			properties: {
-				code: {
-					type: "string",
-					description: "The code to execute"
-				},
-				language: {
-					type: "string",
-					description: "The programming language"
-				}
-			},
-			required: ["code"]
-		}
-	},
-	computer: {
-		description: "Control computer desktop. NOTE: This is a client-side tool - the client must handle computer control.",
-		input_schema: {
-			type: "object",
-			properties: { action: {
-				type: "string",
-				description: "The action to perform"
-			} },
-			required: ["action"]
-		}
-	}
-};
-/**
-* Check if a tool is a server-side tool that needs conversion.
-*/
-function getServerToolPrefix(tool) {
-	if (tool.type) {
-		for (const prefix of Object.keys(SERVER_TOOL_CONFIGS)) if (tool.type.startsWith(prefix)) return prefix;
-	}
-	return null;
-}
-/**
-* Convert server-side tools to custom tools, or pass them through unchanged.
-* This allows them to be passed to the API and handled by the client.
-*
-* Note: Server-side tools are only converted if state.rewriteAnthropicTools is enabled.
-*/
-function convertServerToolsToCustom(tools) {
-	if (!tools) return;
-	const result = [];
-	for (const tool of tools) {
-		const serverToolPrefix = getServerToolPrefix(tool);
-		if (serverToolPrefix) {
-			const config = SERVER_TOOL_CONFIGS[serverToolPrefix];
-			if (!state.rewriteAnthropicTools) {
-				consola.debug(`[DirectAnthropic] Passing ${serverToolPrefix} through unchanged (use --rewrite-anthropic-tools to convert)`);
-				result.push(tool);
-				continue;
-			}
-			if (config.remove) {
-				consola.warn(`[DirectAnthropic] Removing unsupported server tool: ${tool.name}. Reason: ${config.removalReason}`);
-				continue;
-			}
-			consola.debug(`[DirectAnthropic] Converting server tool to custom: ${tool.name} (type: ${tool.type})`);
-			result.push({
-				name: tool.name,
-				description: config.description,
-				input_schema: config.input_schema
-			});
-		} else result.push(tool);
-	}
-	return result.length > 0 ? result : void 0;
-}
-/**
-* Check if a model supports direct Anthropic API.
-* Returns true if redirect is disabled (direct API is on) and the model is from Anthropic vendor.
-*/
-function supportsDirectAnthropicApi(modelId) {
-	if (state.redirectAnthropic) return false;
-	return (state.models?.data.find((m) => m.id === modelId))?.vendor === "Anthropic";
-}
-//#endregion
-//#region src/routes/messages/stream-accumulator.ts
-function createAnthropicStreamAccumulator() {
-	return {
-		model: "",
-		inputTokens: 0,
-		outputTokens: 0,
-		stopReason: "",
-		content: "",
-		toolCalls: [],
-		currentToolCall: null
-	};
-}
-function processAnthropicEvent(event, acc) {
-	switch (event.type) {
-		case "content_block_delta":
-			handleContentBlockDelta(event.delta, acc);
-			break;
-		case "content_block_start":
-			handleContentBlockStart(event.content_block, acc);
-			break;
-		case "content_block_stop":
-			handleContentBlockStop(acc);
-			break;
-		case "message_delta":
-			handleMessageDelta(event.delta, event.usage, acc);
-			break;
-		default: break;
-	}
-}
-function handleContentBlockDelta(delta, acc) {
-	if (delta.type === "text_delta") acc.content += delta.text;
-	else if (delta.type === "input_json_delta" && acc.currentToolCall) acc.currentToolCall.input += delta.partial_json;
-}
-function handleContentBlockStart(block, acc) {
-	if (block.type === "tool_use") acc.currentToolCall = {
-		id: block.id,
-		name: block.name,
-		input: ""
-	};
-}
-function handleContentBlockStop(acc) {
-	if (acc.currentToolCall) {
-		acc.toolCalls.push(acc.currentToolCall);
-		acc.currentToolCall = null;
-	}
-}
-function handleMessageDelta(delta, usage, acc) {
-	if (delta.stop_reason) acc.stopReason = delta.stop_reason;
-	if (usage) {
-		acc.inputTokens = usage.input_tokens ?? 0;
-		acc.outputTokens = usage.output_tokens;
-	}
-}
 //#endregion
 //#region src/routes/messages/stream-translation.ts
 function isToolBlockOpen(state) {
@@ -6212,9 +6184,19 @@ function recordStreamingResponse(acc, fallbackModel, ctx) {
 //#endregion
 //#region src/routes/messages/handler.ts
+function resolveModelFromBetaHeader(model, betaHeader) {
+	if (!betaHeader || !/\bcontext-1m\b/.test(betaHeader)) return model;
+	if (!model.startsWith("claude-")) return model;
+	if (model.endsWith("-1m")) return model;
+	const resolved = `${model}-1m`;
+	consola.debug(`Detected context-1m in anthropic-beta header, resolving model: ${model} → ${resolved}`);
+	return resolved;
+}
 async function handleCompletion(c) {
 	const anthropicPayload = await c.req.json();
 	consola.debug("Anthropic request payload:", JSON.stringify(anthropicPayload));
+	const betaHeader = c.req.header("anthropic-beta");
+	anthropicPayload.model = resolveModelFromBetaHeader(anthropicPayload.model, betaHeader);
 	logToolInfo(anthropicPayload);
 	const subagentMarker = parseSubagentMarkerFromFirstUser(anthropicPayload);
 	const initiatorOverride = subagentMarker ? "agent" : void 0;
@@ -6259,6 +6241,57 @@ function logToolInfo(anthropicPayload) {
 	}
 }
+//#endregion
+//#region src/routes/messages/count-tokens-handler.ts
+/**
+* Handles token counting for Anthropic messages.
+*
+* For Anthropic models (vendor === "Anthropic"), uses the official Anthropic tokenizer.
+* For other models, uses GPT tokenizers with appropriate buffers.
+*
+* When auto-truncate is enabled and the request would exceed limits,
+* returns an inflated token count to trigger Claude Code's auto-compact mechanism.
+*/
+async function handleCountTokens(c) {
+	try {
+		const anthropicBeta = c.req.header("anthropic-beta");
+		const anthropicPayload = await c.req.json();
+		anthropicPayload.model = resolveModelFromBetaHeader(anthropicPayload.model, anthropicBeta);
+		const { payload: openAIPayload } = translateToOpenAI(anthropicPayload);
+		const selectedModel = state.models?.data.find((model) => model.id === openAIPayload.model);
+		if (!selectedModel) {
+			consola.warn("Model not found, returning default token count");
+			return c.json({ input_tokens: 1 });
+		}
+		if (state.autoTruncate) {
+			const truncateCheck = await checkNeedsCompactionAnthropic(anthropicPayload, selectedModel);
+			if (truncateCheck.needed) {
+				const contextWindow = selectedModel.capabilities?.limits?.max_context_window_tokens ?? 2e5;
+				const inflatedTokens = Math.floor(contextWindow * .95);
+				consola.debug(`[count_tokens] Would trigger auto-truncate: ${truncateCheck.currentTokens} tokens > ${truncateCheck.tokenLimit}, returning inflated count: ${inflatedTokens}`);
+				return c.json({ input_tokens: inflatedTokens });
+			}
+		}
+		const tokenizerName = selectedModel.capabilities?.tokenizer ?? "o200k_base";
+		const tokenCount = await getTokenCount(openAIPayload, selectedModel);
+		if (anthropicPayload.tools && anthropicPayload.tools.length > 0) {
+			let mcpToolExist = false;
+			if (anthropicBeta?.startsWith("claude-code")) mcpToolExist = anthropicPayload.tools.some((tool) => tool.name.startsWith("mcp__"));
+			if (!mcpToolExist) {
+				if (anthropicPayload.model.startsWith("claude")) tokenCount.input = tokenCount.input + 346;
+				else if (anthropicPayload.model.startsWith("grok")) tokenCount.input = tokenCount.input + 480;
+			}
+		}
+		let finalTokenCount = tokenCount.input + tokenCount.output;
+		if (!(selectedModel.vendor === "Anthropic")) finalTokenCount = anthropicPayload.model.startsWith("grok") ? Math.round(finalTokenCount * 1.03) : Math.round(finalTokenCount * 1.05);
+		consola.debug(`Token count: ${finalTokenCount} (tokenizer: ${tokenizerName})`);
+		return c.json({ input_tokens: finalTokenCount });
+	} catch (error) {
+		consola.error("Error counting tokens:", error);
+		return c.json({ input_tokens: 1 });
+	}
+}
 //#endregion
 //#region src/routes/messages/route.ts
 const messageRoutes = new Hono();

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "@dianshuv/copilot-api",
-  "version": "0.2.0",
+  "version": "0.2.2",
   "description": "Turn GitHub Copilot into OpenAI/Anthropic API compatible server. Usable with Claude Code!",
   "author": "dianshuv",
   "type": "module",
@@ -19,7 +19,7 @@
     "prepack": "npm run build",
     "prepare": "npm run build && (command -v bun >/dev/null 2>&1 && simple-git-hooks || true)",
     "prepublishOnly": "npm run typecheck && npm run lint:all && npm run test",
-    "release": "npm publish --access public",
+    "release": "npm publish --access public --//registry.npmjs.org/:_authToken=$NPM_TOKEN",
     "start": "NODE_ENV=production bun run ./src/main.ts",
     "test": "bun test tests/*.test.ts",
     "test:all": "bun test tests/*.test.ts && bun test tests/integration/",