ework-qq-bridge 0.4.0 → 0.5.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ework-qq-bridge",
-  "version": "0.4.0",
+  "version": "0.5.0",
   "description": "QQ group <-> ework issue bridge (OneBot 11 reverse WebSocket)",
   "type": "module",
   "main": "src/index.ts",

package/src/chat.ts

@@ -0,0 +1,108 @@
+// 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;
+
+function capContent(text: string): string {
+  return text.length > CHAT_MSG_CHAR_CAP ? text.slice(0, CHAT_MSG_CHAR_CAP) + "…（已截断）" : text;
+}
+
+// Keep the newest turns that fit the token budget (oldest dropped first).
+export function trimToContext(
+  history: ChatTurn[],
+  maxContextTokens: number,
+  reservedTokens: number,
+): ChatTurn[] {
+  const budget = Math.max(0, maxContextTokens - reservedTokens);
+  const kept: ChatTurn[] = [];
+  let used = 0;
+  for (let i = history.length - 1; i >= 0; i--) {
+    const turn = history[i];
+    if (!turn) continue;
+    const cost = estimateTokens(turn.content);
+    if (used + cost > budget) break;
+    used += cost;
+    kept.unshift(turn);
+  }
+  return kept;
+}
+
+export function buildChatMessages(
+  history: ChatTurn[],
+  question: ChatTurn,
+  maxHistory: number,
+  maxContextTokens: number,
+): { role: string; name?: string; content: string }[] {
+  const system = { role: "system", content: CHAT_SYSTEM_PROMPT };
+  const recent = history.slice(-maxHistory);
+  const kept = trimToContext(recent, maxContextTokens, estimateTokens(CHAT_SYSTEM_PROMPT) + estimateTokens(question.content));
+  return [
+    system,
+    ...kept.map((t) => ({ role: t.role, name: t.name, content: capContent(t.content) })),
+    { role: question.role, name: question.name, content: capContent(question.content) },
+  ];
+}
+
+export async function chatComplete(
+  apiBase: string,
+  apiKey: string,
+  model: string,
+  messages: { role: string; name?: string; content: string }[],
+  timeoutMs: number,
+): 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({ 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 = data.choices?.[0]?.message?.content?.trim();
+    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;
+}

package/src/config.ts

@@ -25,6 +25,17 @@ 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),
+
   // QQ user_ids allowed to dispatch AI work (comma-separated). Messages from
   // other members are logged and ignored — same trust model as the GitHub
   // side (WORK_WAKE_LOGINS): strangers never wake the AI.

package/src/router.ts

@@ -3,6 +3,7 @@ import type { EworkClient } from "./ework";
 import type { GroupMessageEvent } from "./onebot";
 import type { BridgeStore } from "./db";
 import type { BindingStore } from "./bindings";
+import { buildChatMessages, chatComplete, splitForQQ, type ChatTurn } from "./chat";
 
 const HELP_TEXT = [
   "用法：",
@@ -11,7 +12,8 @@ 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",
-  "  （绑定后：@我 和普通发言等效，都进入绑定的 issue，AI 回复自动回群）",
+  "  @我 <问题> —— 即时问答（纯 API，不留 issue，上下文满自动清理）",
+  "  （绑定后：普通发言进绑定的 issue，AI 回复自动回群）",
 ].join("\n");
 
 export interface RouterDeps {
@@ -46,6 +48,25 @@ 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: question };
+      const history = chatHistory.get(ev.groupId) ?? [];
+      const messages = buildChatMessages(history, turn, cfg.WORK_CHAT_MAX_HISTORY, cfg.WORK_CHAT_MAX_CONTEXT);
+      const answer = await chatComplete(cfg.WORK_CHAT_API, cfg.WORK_CHAT_API_KEY, cfg.WORK_CHAT_MODEL, messages, cfg.WORK_CHAT_TIMEOUT_MS);
+      history.push(turn, { role: "assistant", name: "bot", content: answer });
+      chatHistory.set(ev.groupId, history.slice(-cfg.WORK_CHAT_MAX_HISTORY * 2));
+      for (const part of splitForQQ(answer)) {
+        await deps.reply(ev.groupId, part);
+      }
+    } catch (e) {
+      const msg = e instanceof Error ? e.message : String(e);
+      console.error(`[qq-bridge] chat failed: ${msg}`);
+      await deps.reply(ev.groupId, `❌ 问答失败：${msg}`);
+    }
+  }
   async function handleGroupMessage(ev: GroupMessageEvent): Promise<void> {
     if (store.seenPost(ev.postId)) return;
     const binding = bindings.resolve(ev.groupId);
@@ -59,8 +80,12 @@ export function createRouter(deps: RouterDeps) {
     const atBot = ev.rawMessage.includes("[CQ:at,qq=") || /^\s*(任务|task|新任务|#|绑定|解绑|帮助|help|查询)/.test(ev.rawMessage);
     const cmd = parseCommand(ev.rawMessage);
     if (!cmd) {
+      const text = ev.rawMessage.replace(/\[CQ:[^\]]*\]/g, "").trim();
+      if (atBot && text && cfg.WORK_CHAT_API) {
+        await answerChat(ev, text);
+        return;
+      }
       if (binding.issue !== undefined) {
-        const text = ev.rawMessage.replace(/\[CQ:[^\]]*\]/g, "").trim();
         if (text) {
           await ework.addComment(binding.owner, binding.repo, binding.issue, `> 来自 QQ 群用户 **${ev.nickname}** (${ev.userId})\n\n${text}`);
         }

package/tests/bridge.test.ts

@@ -97,7 +97,7 @@ describe("@bot unified routing", () => {
   };
   const ev = (postId: string, raw: string) => ({ groupId: 1, userId: 1, nickname: "u", postId, rawMessage: raw });
 
-  test("pinned: @bot question goes to bound issue like plain messages", async () => {
+  test("pinned: @bot falls to bound issue when chat API unset", async () => {
     const { router, replies, comments } = mk(new BindingStore([{ groupId: 1, owner: "o", repo: "r", issue: 7 }], pinFile()));
     await router.handleGroupMessage(ev("p1", "[CQ:at,qq=2661222094] 这个报错啥意思"));
     expect(comments.length).toBe(1);
@@ -229,3 +229,91 @@ describe("issue pinning", () => {
     expect(bs.resolve(1)?.issue).toBe(12);
   });
 });
+
+describe("pure-API chat", () => {
+  const { estimateTokens, trimToContext, 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("trimToContext drops oldest first when over budget", () => {
+    const hist = [turn("a".repeat(1000)), turn("b".repeat(1000)), turn("c".repeat(1000)), turn("d".repeat(1000))];
+    const budget = estimateTokens("c".repeat(1000)) + estimateTokens("d".repeat(1000));
+    const kept = trimToContext(hist, budget, 0);
+    expect(kept.map((t: { content: string }) => t.content[0])).toEqual(["c", "d"]);
+  });
+
+  test("trimToContext keeps everything under budget", () => {
+    const hist = [turn("hi"), turn("yo")];
+    expect(trimToContext(hist, 100000, 0).length).toBe(2);
+  });
+
+  test("buildChatMessages truncates oversized single message", () => {
+    const msgs = buildChatMessages([], turn("x".repeat(CHAT_MSG_CHAR_CAP + 500)), 20, 50000);
+    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(hist, turn("q"), 20, 50000);
+    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 () => {
+    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 });
+    }) 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 },
+        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" });
+      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);
+    } finally {
+      globalThis.fetch = origFetch;
+    }
+  });
+
+  test("router: chat failure surfaces error reply", async () => {
+    const origFetch = globalThis.fetch;
+    globalThis.fetch = (async () => new Response("boom", { status: 500 })) 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 },
+        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;
+    }
+  });
+});