Merge branch '7836246:main' into main

config.yaml.example

@@ -27,6 +27,19 @@ timeout: 120
 # Cursor 使用的模型
 cursor_model: "anthropic/claude-sonnet-4.6"
 
+# ==================== 自动续写配置 ====================
+# 当模型输出被截断时，自动发起续写请求的最大次数
+# 设为 0 可完全禁用自动续写（由用户在对话中手动续写）
+# 环境变量: MAX_AUTO_CONTINUE=3
+max_auto_continue: 3
+
+# ==================== 历史消息条数硬限制 ====================
+# 输入消息条数上限，超出时删除最早的消息（保留工具 few-shot 示例）
+# 防止超长对话（800+ 条）导致请求体积过大、响应变慢
+# 设为 -1 不限制消息条数
+# 环境变量: MAX_HISTORY_MESSAGES=100
+max_history_messages: 100
+
 # ==================== Thinking 开关（最高优先级） ====================
 # 控制是否向 Cursor 发送 thinking 请求，优先级高于客户端传入的 thinking 参数
 # 设为 true: 强制启用 thinking（即使客户端没请求也注入）

docker-compose.yml

@@ -36,6 +36,10 @@ services:
       # - COMPRESSION_ENABLED=true
       # - COMPRESSION_LEVEL=2
 
+      # ── 自动续写 & 历史消息限制 ──
+      # - MAX_AUTO_CONTINUE=3          # 截断后自动续写次数，0=禁用
+      # - MAX_HISTORY_MESSAGES=100     # 历史消息条数上限，-1=不限制
+
       # ── 日志持久化 ──
       # - LOG_FILE_ENABLED=true
       # - LOG_DIR=./logs

src/config.ts

@@ -12,6 +12,8 @@ export function getConfig(): AppConfig {
         port: 3010,
         timeout: 120,
         cursorModel: 'anthropic/claude-sonnet-4.6',
+        maxAutoContinue: 3,
+        maxHistoryMessages: 100,
         fingerprint: {
             userAgent: 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/140.0.0.0 Safari/537.36',
         },
@@ -26,6 +28,8 @@ export function getConfig(): AppConfig {
             if (yaml.timeout) config.timeout = yaml.timeout;
             if (yaml.proxy) config.proxy = yaml.proxy;
             if (yaml.cursor_model) config.cursorModel = yaml.cursor_model;
+            if (typeof yaml.max_auto_continue === 'number') config.maxAutoContinue = yaml.max_auto_continue;
+            if (typeof yaml.max_history_messages === 'number') config.maxHistoryMessages = yaml.max_history_messages;
             if (yaml.fingerprint) {
                 if (yaml.fingerprint.user_agent) config.fingerprint.userAgent = yaml.fingerprint.user_agent;
             }
@@ -90,6 +94,8 @@ export function getConfig(): AppConfig {
     if (process.env.TIMEOUT) config.timeout = parseInt(process.env.TIMEOUT);
     if (process.env.PROXY) config.proxy = process.env.PROXY;
     if (process.env.CURSOR_MODEL) config.cursorModel = process.env.CURSOR_MODEL;
+    if (process.env.MAX_AUTO_CONTINUE !== undefined) config.maxAutoContinue = parseInt(process.env.MAX_AUTO_CONTINUE);
+    if (process.env.MAX_HISTORY_MESSAGES !== undefined) config.maxHistoryMessages = parseInt(process.env.MAX_HISTORY_MESSAGES);
     if (process.env.AUTH_TOKEN) {
         config.authTokens = process.env.AUTH_TOKEN.split(',').map(s => s.trim()).filter(Boolean);
     }

src/converter.ts

@@ -399,6 +399,19 @@ export async function convertToCursorRequest(req: AnthropicRequest): Promise<Cur
         }
     }
 
+    // ★ 历史消息条数硬限制
+    // 超出 max_history_messages 时，删除最早的消息（保留 few-shot 示例）
+    const maxHistoryMessages = config.maxHistoryMessages;
+    if (maxHistoryMessages >= 0) {
+        const fewShotOffset = hasTools ? 2 : 0; // 工具模式有2条 few-shot 消息需跳过
+        const userMessages = messages.length - fewShotOffset;
+        if (userMessages > maxHistoryMessages) {
+            const toRemove = userMessages - maxHistoryMessages;
+            messages.splice(fewShotOffset, toRemove);
+            console.log(`[Converter] 历史消息裁剪: ${userMessages} → ${maxHistoryMessages} 条 (移除了最早的 ${toRemove} 条)`);
+        }
+    }
+
     // ★ 渐进式历史压缩（智能压缩，不破坏结构）
     // 可通过 config.yaml 的 compression 配置控制开关和级别
     // 策略：保留最近 KEEP_RECENT 条消息完整，对早期消息进行结构感知压缩

src/handler.ts

@@ -693,12 +693,12 @@ export async function autoContinueCursorToolResponseStream(
     hasTools: boolean,
 ): Promise<string> {
     let fullResponse = initialResponse;
-    const MAX_AUTO_CONTINUE = 3;
+    const MAX_AUTO_CONTINUE = getConfig().maxAutoContinue;
     let continueCount = 0;
     let consecutiveSmallAdds = 0;
     const originalMessages = [...cursorReq.messages];
 
-    while (shouldAutoContinueTruncatedToolResponse(fullResponse, hasTools) && continueCount < MAX_AUTO_CONTINUE) {
+    while (MAX_AUTO_CONTINUE > 0 && shouldAutoContinueTruncatedToolResponse(fullResponse, hasTools) && continueCount < MAX_AUTO_CONTINUE) {
         continueCount++;
 
         const anchorLength = Math.min(300, fullResponse.length);
@@ -760,12 +760,12 @@ export async function autoContinueCursorToolResponseFull(
     hasTools: boolean,
 ): Promise<string> {
     let fullText = initialText;
-    const MAX_AUTO_CONTINUE = 3;
+    const MAX_AUTO_CONTINUE = getConfig().maxAutoContinue;
     let continueCount = 0;
     let consecutiveSmallAdds = 0;
     const originalMessages = [...cursorReq.messages];
 
-    while (shouldAutoContinueTruncatedToolResponse(fullText, hasTools) && continueCount < MAX_AUTO_CONTINUE) {
+    while (MAX_AUTO_CONTINUE > 0 && shouldAutoContinueTruncatedToolResponse(fullText, hasTools) && continueCount < MAX_AUTO_CONTINUE) {
         continueCount++;
 
         const anchorLength = Math.min(300, fullText.length);
@@ -1279,14 +1279,14 @@ async function handleStream(res: Response, cursorReq: CursorChatRequest, body: A
         // 流完成后，处理完整响应
         // ★ 内部截断续写：如果模型输出过长被截断（常见于写大文件），Proxy 内部分段续写，然后拼接成完整响应
         // 这样可以确保工具调用（如 Write）不会横跨两次 API 响应而退化为纯文本
-        const MAX_AUTO_CONTINUE = 3;
+        const MAX_AUTO_CONTINUE = getConfig().maxAutoContinue;
         let continueCount = 0;
         let consecutiveSmallAdds = 0; // 连续小增量计数
         
         // 保存原始请求的消息快照（不含续写追加的消息）
         const originalMessages = [...activeCursorReq.messages];
         
-        while (shouldAutoContinueTruncatedToolResponse(fullResponse, hasTools) && continueCount < MAX_AUTO_CONTINUE) {
+        while (MAX_AUTO_CONTINUE > 0 && shouldAutoContinueTruncatedToolResponse(fullResponse, hasTools) && continueCount < MAX_AUTO_CONTINUE) {
             continueCount++;
             const prevLength = fullResponse.length;
             log.warn('Handler', 'continuation', `内部检测到截断 (${fullResponse.length} chars)，隐式续写 (第${continueCount}次)`);
@@ -1671,12 +1671,12 @@ async function handleNonStream(res: Response, cursorReq: CursorChatRequest, body
     // ★ 内部截断续写（与流式路径对齐）
     // Claude CLI 使用非流式模式时，写大文件最容易被截断
     // 在 proxy 内部完成续写，确保工具调用参数完整
-    const MAX_AUTO_CONTINUE = 3;
+    const MAX_AUTO_CONTINUE = getConfig().maxAutoContinue;
     let continueCount = 0;
     let consecutiveSmallAdds = 0; // 连续小增量计数
     const originalMessages = [...activeCursorReq.messages];
 
-    while (shouldAutoContinueTruncatedToolResponse(fullText, hasTools) && continueCount < MAX_AUTO_CONTINUE) {
+    while (MAX_AUTO_CONTINUE > 0 && shouldAutoContinueTruncatedToolResponse(fullText, hasTools) && continueCount < MAX_AUTO_CONTINUE) {
         continueCount++;
         const prevLength = fullText.length;
         log.warn('Handler', 'continuation', `非流式检测到截断 (${fullText.length} chars)，隐式续写 (第${continueCount}次)`);

src/types.ts

@@ -106,6 +106,8 @@ export interface AppConfig {
     proxy?: string;
     cursorModel: string;
     authTokens?: string[];  // API 鉴权 token 列表，为空则不鉴权
+    maxAutoContinue: number;        // 自动续写最大次数，默认 3，设 0 禁用
+    maxHistoryMessages: number;     // 历史消息条数硬限制，默认 100，-1 不限制
     vision?: {
         enabled: boolean;
         mode: 'ocr' | 'api';