ework-qq-bridge 0.5.2 → 0.6.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ework-qq-bridge",
-  "version": "0.5.2",
+  "version": "0.6.0",
   "description": "QQ group <-> ework issue bridge (OneBot 11 reverse WebSocket)",
   "type": "module",
   "main": "src/index.ts",

package/src/config.ts

@@ -25,20 +25,12 @@ const Schema = z.object({
   // Runtime pin overrides written by the 绑定/解绑 commands (JSON, group -> issue).
   WORK_BINDINGS_FILE: z.string().default(""),
 
-  // Pure-API chat (@bot Q&A): OpenAI-compatible endpoint, no tools/session.
-  // Empty WORK_CHAT_API disables chat and falls back to usage guidance.
-  WORK_CHAT_API: z.string().default(""),
-  WORK_CHAT_API_KEY: z.string().default(""),
-  WORK_CHAT_MODEL: z.string().default(""),
-  WORK_CHAT_TIMEOUT_MS: z.coerce.number().int().default(90000),
-  WORK_CHAT_MAX_HISTORY: z.coerce.number().int().default(20),
-  // Token budget (heuristic estimate) for one chat request; when history
-  // exceeds it the oldest turns are dropped ("满了就清理").
-  WORK_CHAT_MAX_CONTEXT: z.coerce.number().int().default(50000),
-  // Disable the model's thinking pass for QQ chat replies only (default on:
-  // "仅仅这个机器人关掉"); "0" restores thinking. Daemon/opencode sessions
-  // are untouched — they use their own runtime.
-  WORK_CHAT_NO_THINK: z.preprocess((v) => (v === undefined || v === "" ? "1" : v), z.string()).transform((v) => v !== "0"),
+  // Pure chat (@bot Q&A) is delegated to the standalone ework-chat service
+  // (chat-only LLM + bili transparent compression + JSONL per-turn disk
+  // persistence). Empty WORK_CHAT_URL disables chat and falls back to usage
+  // guidance.
+  WORK_CHAT_URL: z.string().default(""),
+  WORK_CHAT_TOKEN: z.string().default(""),
 
   // QQ user_ids allowed to dispatch AI work (comma-separated). Messages from
   // other members are logged and ignored — same trust model as the GitHub

package/src/router.ts

@@ -3,7 +3,29 @@ import type { EworkClient } from "./ework";
 import type { GroupMessageEvent } from "./onebot";
 import type { BridgeStore } from "./db";
 import type { BindingStore } from "./bindings";
-import { buildChatMessages, capContent, chatComplete, splitForQQ, trimStored, type ChatTurn } from "./chat";
+
+export const QQ_CHAT_SYSTEM = [
+  "你是 QQ 群里的即时问答助手，背后是 ework 开发平台。",
+  "风格：简短直接，能用一两句话说清的就别铺开；技术问题给结论和关键理由，需要展开再展开。",
+  "群里成员通过 @你 提问。你看到的多轮对话里每条 user 消息前缀了提问者的昵称，注意区分不同人。",
+  "如果请求明显是需要长时间执行的开发任务（改代码、查仓库、提交 PR），不要假装去做——建议对方发「任务 <标题>」创建 issue，AI agent 会接单处理。",
+  "不知道就直说，不要编造。",
+].join("\n");
+
+export function splitForQQ(text: string, maxLen = 1500): string[] {
+  if (text.length <= maxLen) return [text];
+  const parts: string[] = [];
+  let remaining = text;
+  while (remaining.length > maxLen) {
+    let cut = remaining.lastIndexOf("\n", maxLen);
+    if (cut < maxLen * 0.5) cut = remaining.lastIndexOf("。", maxLen);
+    if (cut < maxLen * 0.5) cut = maxLen;
+    parts.push(remaining.slice(0, cut + 1));
+    remaining = remaining.slice(cut + 1);
+  }
+  if (remaining) parts.push(remaining);
+  return parts;
+}
 
 const HELP_TEXT = [
   "用法：",
@@ -12,7 +34,7 @@ const HELP_TEXT = [
   "  绑定 #<编号> —— 把本群绑定到该 issue（长记忆模式：此后发言都进这个 issue）",
   "  解绑 —— 恢复为项目模式（接收整个项目的回复）",
   "  查询 —— 列出最近 issue",
-  "  @我 <问题> —— 即时问答（纯 API，不留 issue，上下文满自动清理）",
+  "  @我 <问题> —— 即时问答（透明压缩长记忆，历史落盘）",
   "  （绑定后：普通发言进绑定的 issue，AI 回复自动回群）",
 ].join("\n");
 
@@ -48,17 +70,26 @@ export function parseCommand(raw: string): ParsedCommand | null {
 
 export function createRouter(deps: RouterDeps) {
   const { cfg, bindings, wakeList, ework, store } = deps;
-  const chatHistory = new Map<number, ChatTurn[]>();
 
   async function answerChat(ev: GroupMessageEvent, question: string): Promise<void> {
     try {
-      const turn: ChatTurn = { role: "user", name: ev.nickname, content: capContent(question) };
-      const stored = trimStored(chatHistory.get(ev.groupId) ?? [], cfg.WORK_CHAT_MAX_HISTORY, cfg.WORK_CHAT_MAX_CONTEXT);
-      chatHistory.set(ev.groupId, stored);
-      const messages = buildChatMessages(stored, turn);
-      const answer = await chatComplete(cfg.WORK_CHAT_API, cfg.WORK_CHAT_API_KEY, cfg.WORK_CHAT_MODEL, messages, cfg.WORK_CHAT_TIMEOUT_MS, cfg.WORK_CHAT_NO_THINK);
-      chatHistory.set(ev.groupId, [...stored, turn, { role: "assistant", name: "bot", content: capContent(answer) }]);
-      for (const part of splitForQQ(answer)) {
+      const res = await fetch(`${cfg.WORK_CHAT_URL.replace(/\/+$/, "")}/v1/chat`, {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          ...(cfg.WORK_CHAT_TOKEN ? { Authorization: `Bearer ${cfg.WORK_CHAT_TOKEN}` } : {}),
+        },
+        body: JSON.stringify({
+          conversation: String(ev.groupId),
+          message: question,
+          user: ev.nickname,
+          system: QQ_CHAT_SYSTEM,
+        }),
+      });
+      if (!res.ok) throw new Error(`chat service ${res.status}: ${(await res.text()).slice(0, 120)}`);
+      const data = (await res.json()) as { reply?: string };
+      if (!data.reply) throw new Error("chat service returned no reply");
+      for (const part of splitForQQ(data.reply)) {
         await deps.reply(ev.groupId, part);
       }
     } catch (e) {
@@ -81,7 +112,7 @@ export function createRouter(deps: RouterDeps) {
     const cmd = parseCommand(ev.rawMessage);
     if (!cmd) {
       const text = ev.rawMessage.replace(/\[CQ:[^\]]*\]/g, "").trim();
-      if (atBot && text && cfg.WORK_CHAT_API) {
+      if (atBot && text && cfg.WORK_CHAT_URL) {
         await answerChat(ev, text);
         return;
       }

package/tests/bridge.test.ts

@@ -4,6 +4,7 @@ import { BindingStore } from "../src/bindings";
 import { tmpdir } from "node:os";
 import { randomUUID } from "node:crypto";
 const pinFile = () => `${tmpdir()}/qqb-test-${randomUUID()}.json`;
+const histFile = () => `${tmpdir()}/qqb-hist-${randomUUID()}.json`;
 import { parseGroupMap, parseList } from "../src/config";
 import { buildScrubber } from "../src/scrub";
 import { verifySignature } from "../src/ingest";
@@ -230,111 +231,37 @@ describe("issue pinning", () => {
   });
 });
 
-describe("pure-API chat", () => {
-  const { estimateTokens, trimStored, buildChatMessages, CHAT_SYSTEM_PROMPT, CHAT_MSG_CHAR_CAP } = require("../src/chat");
-  const turn = (content: string, role: "user" | "assistant" = "user") => ({ role, name: "u", content });
 
-  test("estimateTokens grows with length", () => {
-    expect(estimateTokens("ab")).toBeGreaterThan(0);
-    expect(estimateTokens("abcd".repeat(100))).toBeGreaterThan(estimateTokens("abcd"));
-  });
-
-  test("trimStored is a no-op under both limits", () => {
-    const hist = [turn("hi"), turn("yo"), turn("lo")];
-    expect(trimStored(hist, 20, 50000)).toBe(hist);
-  });
-
-  test("trimStored evicts in bulk on count overflow and is sticky", () => {
-    const mk = (n: number) => Array.from({ length: n }, (_, i) => turn("m" + i));
-    const over = trimStored(mk(41), 20, 5000000);
-    expect(over.length).toBe(Math.max(2, Math.floor(40 * 0.7)));
-    expect(over[over.length - 1].content).toBe("m40");
-    expect(trimStored(over, 20, 5000000)).toBe(over);
-  });
-
-  test("trimStored evicts in bulk on token overflow", () => {
-    const hist = Array.from({ length: 10 }, (_, i) => turn("x".repeat(1000) + i));
-    const per = estimateTokens("x".repeat(1000));
-    const reserve = estimateTokens(CHAT_SYSTEM_PROMPT) + estimateTokens("x".repeat(CHAT_MSG_CHAR_CAP));
-    const budget = reserve + per * 6;
-    const kept = trimStored(hist, 100, budget);
-    expect(kept.length).toBeLessThan(10);
-    const keptTotal = kept.reduce((n: number, t: { content: string }) => n + estimateTokens(t.content), 0);
-    const tokenFloor = Math.floor(per * 6 * 0.7);
-    expect(keptTotal).toBeLessThanOrEqual(tokenFloor + per);
-    expect(kept[kept.length - 1].content.endsWith("9")).toBe(true);
-  });
-
-  test("count cap evicts in bulk too (no sliding window)", () => {
-    const mk = (n: number) => Array.from({ length: n }, (_, i) => turn("m" + i));
-    expect(buildChatMessages(mk(40), turn("q")).length).toBe(42);
-    const over = trimStored(mk(41), 20, 5000000);
-    expect(buildChatMessages(over, turn("q")).length).toBe(Math.floor(40 * 0.7) + 2);
-  });
-
-  test("prompt grows append-only between evictions (prefix cache friendly)", () => {
-    const maxHistory = 3;
-    let hist: { role: "user" | "assistant"; name: string; content: string }[] = [];
-    let prev: { role: string; content: string }[] | null = null;
-    let evictions = 0;
-    for (let i = 0; i < 12; i++) {
-      const turnU = { role: "user" as const, name: "u", content: "q" + i };
-      const trimmed = trimStored(hist, maxHistory, 5000000);
-      if (trimmed !== hist) evictions++;
-      hist = trimmed;
-      const msgs = buildChatMessages(hist, turnU).map((m: { role: string; content: string }) => ({ role: m.role, content: m.content }));
-      if (prev) {
-        const prefixIntact = prev.every((pm, idx) => msgs[idx] && msgs[idx].role === pm.role && msgs[idx].content === pm.content);
-        const grewAppendOnly = msgs.length >= prev.length;
-        if (!grewAppendOnly) evictions++;
-        expect(prefixIntact || !grewAppendOnly).toBe(true);
-      }
-      prev = msgs;
-      hist = [...hist, turnU, { role: "assistant", name: "bot", content: "a" + i }];
-    }
-    expect(evictions).toBeGreaterThan(0);
-  });
-
-  test("buildChatMessages truncates oversized single message", () => {
-    const msgs = buildChatMessages([], turn("x".repeat(CHAT_MSG_CHAR_CAP + 500)));
-    expect(msgs[msgs.length - 1].content.length).toBeLessThanOrEqual(CHAT_MSG_CHAR_CAP + 10);
-    expect(msgs[msgs.length - 1].content).toContain("已截断");
-  });
 
-  test("buildChatMessages keeps system prompt first and reserves question budget", () => {
-    const hist = [turn("a".repeat(40000)), turn("b")];
-    const msgs = buildChatMessages(trimStored(hist, 20, 50000), turn("q"));
-    expect(msgs[0].role).toBe("system");
-    expect(msgs[0].content).toBe(CHAT_SYSTEM_PROMPT);
-    const total = msgs.reduce((n: number, m: { content: string }) => n + estimateTokens(m.content), 0);
-    expect(total).toBeLessThanOrEqual(50000 + estimateTokens("q") + estimateTokens(CHAT_SYSTEM_PROMPT));
-  });
 
-  test("router: @bot routes to chat when WORK_CHAT_API set", async () => {
+describe("chat delegation to ework-chat", () => {
+  test("router: @bot delegates to ework-chat service", async () => {
     const origFetch = globalThis.fetch;
-    const bodies: unknown[] = [];
-    globalThis.fetch = (async (_url: unknown, init?: { body?: string }) => {
-      bodies.push(JSON.parse(String(init?.body)));
-      return new Response(JSON.stringify({ choices: [{ message: { content: "秒回的答案" } }] }), { status: 200 });
+    const calls: Array<{ url: string; body: Record<string, unknown> }> = [];
+    globalThis.fetch = (async (url: unknown, init?: { body?: string; headers?: Record<string, string> }) => {
+      calls.push({ url: String(url), body: JSON.parse(String(init?.body)) });
+      return new Response(JSON.stringify({ conversation: "1", reply: "秒回的答案" }), { status: 200 });
     }) as typeof fetch;
     try {
       const { createRouter } = require("../src/router");
       const replies: string[] = [];
       const comments: unknown[] = [];
       const router = createRouter({
-        cfg: { VERBOSE: false, WORK_CHAT_API: "http://x/v1", WORK_CHAT_API_KEY: "k", WORK_CHAT_MODEL: "m", WORK_CHAT_TIMEOUT_MS: 1000, WORK_CHAT_MAX_HISTORY: 20, WORK_CHAT_MAX_CONTEXT: 50000 },
+        cfg: { VERBOSE: false, WORK_CHAT_URL: "http://127.0.0.1:8210", WORK_CHAT_TOKEN: "t0" },
         bindings: new BindingStore([{ groupId: 1, owner: "o", repo: "r", issue: 7 }], pinFile()),
         wakeList: new Set(["1"]),
         ework: { createIssue: async () => 9, addComment: async (...a: unknown[]) => { comments.push(a); } },
         store: { seenPost: () => false },
         reply: async (_g: number, x: string) => { replies.push(x); },
       });
-      await router.handleGroupMessage({ groupId: 1, userId: 1, nickname: "u", postId: "c1", rawMessage: "[CQ:at,qq=2661222094] 快问快答" });
+      await router.handleGroupMessage({ groupId: 1, userId: 1, nickname: "小狗", postId: "c1", rawMessage: "[CQ:at,qq=2661222094] 快问快答" });
       expect(replies).toEqual(["秒回的答案"]);
       expect(comments).toEqual([]);
-      expect((bodies[0] as { messages: { content: string }[] }).messages[0].role).toBe("system");
-      await router.handleGroupMessage({ groupId: 1, userId: 1, nickname: "u", postId: "c2", rawMessage: "[CQ:at,qq=2661222094] 追问一句" });
-      expect((bodies[1] as { messages: { content: string }[] }).messages.length).toBe(4);
+      expect(calls[0]?.url).toBe("http://127.0.0.1:8210/v1/chat");
+      expect(calls[0]?.body.conversation).toBe("1");
+      expect(calls[0]?.body.message).toBe("快问快答");
+      expect(calls[0]?.body.user).toBe("小狗");
+      expect(typeof calls[0]?.body.system).toBe("string");
     } finally {
       globalThis.fetch = origFetch;
     }
@@ -347,65 +274,15 @@ describe("pure-API chat", () => {
       const { createRouter } = require("../src/router");
       const replies: string[] = [];
       const router = createRouter({
-        cfg: { VERBOSE: false, WORK_CHAT_API: "http://x/v1", WORK_CHAT_API_KEY: "k", WORK_CHAT_MODEL: "m", WORK_CHAT_TIMEOUT_MS: 1000, WORK_CHAT_MAX_HISTORY: 20, WORK_CHAT_MAX_CONTEXT: 50000 },
-        bindings: new BindingStore([{ groupId: 1, owner: "o", repo: "r" }], pinFile()),
-        wakeList: new Set(["1"]),
-        ework: { createIssue: async () => 9, addComment: async () => {} },
-        store: { seenPost: () => false },
-        reply: async (_g: number, x: string) => { replies.push(x); },
-      });
-      await router.handleGroupMessage({ groupId: 1, userId: 1, nickname: "u", postId: "c3", rawMessage: "[CQ:at,qq=2661222094] 会失败吗" });
-      expect(replies[0]).toContain("问答失败");
-    } finally {
-      globalThis.fetch = origFetch;
-    }
-  });
-});
-
-describe("no-think", () => {
-  const { stripThink } = require("../src/chat");
-
-  test("stripThink removes inline think blocks", () => {
-    expect(stripThink("<think>internal</think>答案")).toBe("答案");
-    expect(stripThink("<think>a</think>前<think>b</think>后")).toBe("前后");
-    expect(stripThink("普通回复")).toBe("普通回复");
-  });
-
-  test("chat request carries enable_thinking:false by default", async () => {
-    const origFetch = globalThis.fetch;
-    let sent: Record<string, unknown> = {};
-    globalThis.fetch = (async (_u: unknown, init?: { body?: string }) => {
-      sent = JSON.parse(String(init?.body));
-      return new Response(JSON.stringify({ choices: [{ message: { content: "ok" } }] }), { status: 200 });
-    }) as typeof fetch;
-    try {
-      const { chatComplete } = require("../src/chat");
-      const out = await chatComplete("http://x/v1", "k", "m", [{ role: "user", content: "q" }], 1000);
-      expect(out).toBe("ok");
-      expect((sent.chat_template_kwargs as { enable_thinking?: boolean })?.enable_thinking).toBe(false);
-      await chatComplete("http://x/v1", "k", "m", [{ role: "user", content: "q" }], 1000, false);
-      expect(sent.chat_template_kwargs).toBeUndefined();
-    } finally {
-      globalThis.fetch = origFetch;
-    }
-  });
-
-  test("thinking payload is stripped before replying", async () => {
-    const origFetch = globalThis.fetch;
-    globalThis.fetch = (async () => new Response(JSON.stringify({ choices: [{ message: { content: "<think>隐藏推理</think>可见答案" } }] }), { status: 200 })) as typeof fetch;
-    try {
-      const { createRouter } = require("../src/router");
-      const replies: string[] = [];
-      const router = createRouter({
-        cfg: { VERBOSE: false, WORK_CHAT_API: "http://x/v1", WORK_CHAT_API_KEY: "k", WORK_CHAT_MODEL: "m", WORK_CHAT_TIMEOUT_MS: 1000, WORK_CHAT_MAX_HISTORY: 20, WORK_CHAT_MAX_CONTEXT: 50000, WORK_CHAT_NO_THINK: true },
+        cfg: { VERBOSE: false, WORK_CHAT_URL: "http://x", WORK_CHAT_TOKEN: "" },
         bindings: new BindingStore([{ groupId: 1, owner: "o", repo: "r" }], pinFile()),
         wakeList: new Set(["1"]),
         ework: { createIssue: async () => 9, addComment: async () => {} },
         store: { seenPost: () => false },
         reply: async (_g: number, x: string) => { replies.push(x); },
       });
-      await router.handleGroupMessage({ groupId: 1, userId: 1, nickname: "u", postId: "nt1", rawMessage: "[CQ:at,qq=2661222094] 问" });
-      expect(replies).toEqual(["可见答案"]);
+      await router.handleGroupMessage({ groupId: 1, userId: 1, nickname: "u", postId: "c3", rawMessage: "[CQ:at,qq=2661222094] 问点啥" });
+      expect(replies[0]).toContain("❌ 问答失败");
     } finally {
       globalThis.fetch = origFetch;
     }

package/src/chat.ts

@@ -1,129 +0,0 @@
-// Chat mode: instant Q&A directly against the OpenAI-compatible endpoint.
-// Issue mode (任务/#N/绑定) stays the path for long agent work — chat is for
-// the conversational 90% of group traffic that never needs a session.
-
-export interface ChatTurn {
-  role: "user" | "assistant";
-  name: string;
-  content: string;
-}
-
-export const CHAT_SYSTEM_PROMPT = [
-  "你是 QQ 群里的即时问答助手，背后是 ework 开发平台。",
-  "风格：简短直接，能用一两句话说清的就别铺开；技术问题给结论和关键理由，需要展开再展开。",
-  "群里成员通过 @你 提问。你看到的多轮对话里每条 user 消息前缀了提问者的昵称，注意区分不同人。",
-  "如果请求明显是需要长时间执行的开发任务（改代码、查仓库、提交 PR），不要假装去做——建议对方发「任务 <标题>」创建 issue，AI agent 会接单处理。",
-  "不知道就直说，不要编造。",
-].join("\n");
-
-// qwen-family heuristic: CJK ≈ 1 token/char, latin ≈ 1/4 — blended constant.
-export function estimateTokens(text: string): number {
-  return Math.ceil(text.length * 0.75);
-}
-
-// Single-message hard cap so one pasted log cannot eat the whole budget.
-export const CHAT_MSG_CHAR_CAP = 24000;
-
-// Evict in bulk, never slide. A sliding window re-derives "last N" on every
-// build, shifting the prompt prefix each request, missing the serving
-// engine's prefix cache (radix cache) and forcing a full re-prefill of the
-// whole context every turn once history is full. trimStored is instead
-// STICKY: it is a no-op while under the limits and, on crossing one, drops
-// the oldest turns down to TRIM_FLOOR_RATIO in a single shot. Between
-// evictions the stored history only ever grows append-only, so the prompt
-// prefix (system + history) stays byte-identical and cached prefixes keep
-// hitting; only the first request after each eviction re-fills.
-export const TRIM_FLOOR_RATIO = 0.7;
-
-export function capContent(text: string): string {
-  return text.length > CHAT_MSG_CHAR_CAP ? text.slice(0, CHAT_MSG_CHAR_CAP) + "…（已截断）" : text;
-}
-
-function historyTokens(history: ChatTurn[]): number {
-  return history.reduce((n, t) => n + estimateTokens(t.content), 0);
-}
-
-// Headroom for the fixed prompt furniture (system + one max-size question).
-function reserveTokens(): number {
-  return estimateTokens(CHAT_SYSTEM_PROMPT) + estimateTokens("x".repeat(CHAT_MSG_CHAR_CAP));
-}
-
-export function trimStored(history: ChatTurn[], maxHistory: number, maxContextTokens: number): ChatTurn[] {
-  const countLimit = Math.max(2, maxHistory * 2);
-  const tokenLimit = Math.max(0, maxContextTokens - reserveTokens());
-  if (history.length <= countLimit && historyTokens(history) <= tokenLimit) return history;
-  const countFloor = Math.max(2, Math.floor(countLimit * TRIM_FLOOR_RATIO));
-  const tokenFloor = Math.floor(tokenLimit * TRIM_FLOOR_RATIO);
-  let start = 0;
-  let total = historyTokens(history);
-  while (start < history.length && (history.length - start > countFloor || total > tokenFloor)) {
-    const turn = history[start];
-    if (!turn) break;
-    total -= estimateTokens(turn.content);
-    start++;
-  }
-  return history.slice(start);
-}
-
-export function buildChatMessages(
-  history: ChatTurn[],
-  question: ChatTurn,
-): { role: string; name?: string; content: string }[] {
-  return [
-    { role: "system", content: CHAT_SYSTEM_PROMPT },
-    ...history.map((t) => ({ role: t.role, name: t.name, content: capContent(t.content) })),
-    { role: question.role, name: question.name, content: capContent(question.content) },
-  ];
-}
-
-export function stripThink(text: string): string {
-  return text.replace(/<think>[\s\S]*?<\/think>/g, "").trim();
-}
-
-export async function chatComplete(
-  apiBase: string,
-  apiKey: string,
-  model: string,
-  messages: { role: string; name?: string; content: string }[],
-  timeoutMs: number,
-  noThink = true,
-): Promise<string> {
-  const ctrl = new AbortController();
-  const timer = setTimeout(() => ctrl.abort(), timeoutMs);
-  try {
-    const res = await fetch(`${apiBase.replace(/\/+$/, "")}/chat/completions`, {
-      method: "POST",
-      headers: { "Content-Type": "application/json", Authorization: `Bearer ${apiKey}` },
-      body: JSON.stringify(
-        noThink
-          ? { model, messages, max_tokens: 1024, temperature: 0.4, chat_template_kwargs: { enable_thinking: false } }
-          : { model, messages, max_tokens: 1024, temperature: 0.4 },
-      ),
-      signal: ctrl.signal,
-    });
-    if (!res.ok) {
-      throw new Error(`LLM ${res.status}: ${(await res.text()).slice(0, 120)}`);
-    }
-    const data = (await res.json()) as { choices?: { message?: { content?: string } }[] };
-    const text = stripThink(data.choices?.[0]?.message?.content ?? "");
-    if (!text) throw new Error("LLM returned empty content");
-    return text;
-  } finally {
-    clearTimeout(timer);
-  }
-}
-
-export function splitForQQ(text: string, maxLen = 1500): string[] {
-  if (text.length <= maxLen) return [text];
-  const parts: string[] = [];
-  let remaining = text;
-  while (remaining.length > maxLen) {
-    let cut = remaining.lastIndexOf("\n", maxLen);
-    if (cut < maxLen * 0.5) cut = remaining.lastIndexOf("。", maxLen);
-    if (cut < maxLen * 0.5) cut = maxLen;
-    parts.push(remaining.slice(0, cut + 1));
-    remaining = remaining.slice(cut + 1);
-  }
-  if (remaining) parts.push(remaining);
-  return parts;
-}