ework-qq-bridge 0.5.1 → 0.5.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ework-qq-bridge",
-  "version": "0.5.1",
+  "version": "0.5.2",
   "description": "QQ group <-> ework issue bridge (OneBot 11 reverse WebSocket)",
   "type": "module",
   "main": "src/index.ts",

package/src/chat.ts

@@ -76,12 +76,17 @@ export function buildChatMessages(
   ];
 }
 
+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);
@@ -89,14 +94,18 @@ export async function chatComplete(
     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 }),
+      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 = data.choices?.[0]?.message?.content?.trim();
+    const text = stripThink(data.choices?.[0]?.message?.content ?? "");
     if (!text) throw new Error("LLM returned empty content");
     return text;
   } finally {

package/src/config.ts

@@ -35,6 +35,10 @@ const Schema = z.object({
   // 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"),
 
   // 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

@@ -56,7 +56,7 @@ export function createRouter(deps: RouterDeps) {
       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);
+      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)) {
         await deps.reply(ev.groupId, part);

package/tests/bridge.test.ts

@@ -361,3 +361,53 @@ describe("pure-API chat", () => {
     }
   });
 });
+
+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 },
+        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"]);
+    } finally {
+      globalThis.fetch = origFetch;
+    }
+  });
+});