ppxans-harness 2.4.0 → 3.2.0

package/src/memory/memory-ticker.js

@@ -1,240 +1,239 @@
-// src/memory/memory-ticker.js - 记忆水位线 (参考 openhanako v4)
-// 架构: 会话事件日志 (SessionStore) 为唯一事实源; 今日视图由它派生
-// 不再独立维护 today.md 对话原文 (原 l0/session/today 三处重复, 已收敛到 session)
-// 今日视图 -> 滚动压缩 -> longterm.md (长期记忆)
-import fs from "node:fs";
-import path from "node:path";
-import { ensureDir, readText, writeText, logicalDay } from "../utils/store.js";
-
-const TURNS_PER_SUMMARY = 10;
-const COMPACT_THRESHOLD = 50;   // 今日事件超此条数触发滚动压缩
-const COMPACT_KEEP = 20;        // 压缩后保留的近期条数
-const COMPACT_MIN_INTERVAL_MS = 60000; // 压缩节流: 压缩后 60s 内不重复 (防每轮对话重复付 LLM 压缩成本, v1.0.7)
-
-// 派生今日视图行: 从 session 事件渲染 (role -> 中文说话人)
-function _renderLines(sessionStore, day) {
-  if (!sessionStore) return [];
-  return sessionStore.eventsByDay(day).map((r) => {
-    const who = r.role === "user" ? "用户" : "皮皮虾";
-    return `- [${new Date(r.timestamp).toISOString()}] ${who}: ${String(r.content).slice(0, 200)}`;
-  });
-}
-
-export class MemoryTicker {
-  constructor(dataDir, factStore, summarizer = null, sessionStore = null) {
-    this.summarizer = summarizer;
-    this.extractor = null; // P1#9: LLM 结构化提炼器 (agent 注入), 提取关键事实/偏好/待办
-    this.dir = path.join(dataDir, "memory");
-    ensureDir(this.dir);
-    ensureDir(path.join(this.dir, "daily"));
-    this.factStore = factStore;
-    this.sessionStore = sessionStore;   // 唯一事实源 (今日视图由它派生)
-    this.todayView = path.join(this.dir, "today.md"); // 兼容路径 (不再作为写入真相)
-    this.longtermMd = path.join(this.dir, "longterm.md");
-    this.stateFile = path.join(this.dir, "daily-state.json");
-    this.state = { day: null, turnCount: 0 };
-    this._lastCompactAt = 0; // 压缩节流时间戳 (v1.0.7)
-    this._loadState();
-    this._rollDay();
-  }
-
-  _loadState() {
-    try {
-      if (fs.existsSync(this.stateFile)) this.state = JSON.parse(fs.readFileSync(this.stateFile, "utf8"));
-    } catch { this.state = { day: null, turnCount: 0 }; }
-  }
-
-  _saveState() {
-    writeText(this.stateFile, JSON.stringify(this.state, null, 2));
-  }
-
-  _rollDay() {
-    const today = logicalDay();
-    if (this.state.day !== today) {
-      if (this.state.day) this._compileDaily();
-      this.state.day = today;
-      this._saveState();
-    }
-  }
-
-  // 跨天: 把上一日事件归档到 daily/ 并滚入 longterm (从 session 派生, 非 today.md)
-  _compileDaily() {
-    const day = this.state.day;
-    const lines = _renderLines(this.sessionStore, day);
-    const dailyFile = path.join(this.dir, "daily", `${day}.md`);
-    writeText(dailyFile, `# ${day}
-
-${lines.join("\n")}\n`);
-    if (lines.length) {
-      let longterm = readText(this.longtermMd) || "";
-      longterm += `\n## ${day}\n${lines.join("\n")}\n`;
-      writeText(this.longtermMd, longterm);
-    }
-  }
-
-  async recordTurn(user, assistant) {
-    this._rollDay();
-    this.state.turnCount += 1;
-    this._saveState();
-    // 对话原文已由 session 事件日志保存; 今日视图由 session 派生
-    if (this.state.turnCount % TURNS_PER_SUMMARY === 0) this._compileDaily_Rolling();
-    await this._compactIfNeeded();
-    // P1#9: 若配 LLM 提炼器且本次对话含高信号, 走结构化提炼; 否则退回启发式
-    if (this.extractor && _hasSignal(user, assistant)) {
-      try {
-        // 感知式提炼: 先检索与本次对话相关的已有记忆 (同主题 top 3), 喂给提炼器从源头去重
-        const related = String(user || "").trim()
-          ? this.factStore.query(user, { limit: 3 }).filter((f) => !(f.hits && f.hits > 5)) // 高命中已稳定, 不再重复提炼
-          : [];
-        const facts = await this.extractor(String(user || ""), String(assistant || ""), related);
-        if (facts && facts.length) {
-          // similarThreshold=0.6: LLM 提炼的字面变体 (同义不同词) 与已有事实语义相似时命中加分,
-          // 防「三件套」这类反复提炼的变体污染记忆库
-          for (const f of facts) this.factStore.add(f, { source: "extract", similarThreshold: 0.6 });
-          return;
-        }
-      } catch {}
-    }
-    if (user) this.factStore.addMemory(user);
-  }
-
-  // P1#9: 注入 LLM 结构化提炼器 (agent 调 setExtractor)
-  setExtractor(fn) { this.extractor = typeof fn === "function" ? fn : null; }
-
-  // 每 N 轮: 把今日新增事件滚动归档进 longterm (从 session 派生)
-  // v1.2.0 fix: 用游标(lastRolledDay/lastRolledSeq)只追加"本次滚动之后"的新事件,
-  // 不再每次把今日全文重复追加 —— 否则同一段对话会在 longterm 中反复出现, 加剧重复回话。
-  _compileDaily_Rolling() {
-    const today = logicalDay();
-    const st = this.state;
-    if (st.lastRolledDay !== today) { st.lastRolledDay = today; st.lastRolledSeq = 0; } // 跨天重置游标
-    const afterSeq = st.lastRolledSeq || 0;
-    const rows = [];
-    let maxSeq = afterSeq;
-    const dayEvents = this.sessionStore && typeof this.sessionStore.replayDay === "function"
-      ? this.sessionStore.replayDay(today)
-      : [];
-    for (const e of dayEvents) {
-      if (e.seq <= afterSeq) continue;
-      const who = e.type === "user/message" ? "用户" : "皮皮虾";
-      rows.push(`- [${new Date(e.ts).toISOString()}] ${who}: ${String(e.data?.content || "").slice(0, 200)}`);
-      if (e.seq > maxSeq) maxSeq = e.seq;
-    }
-    if (rows.length) {
-      let longterm = readText(this.longtermMd) || "";
-      longterm += `\n## ${today} (滚动)\n${rows.join("\n")}\n`;
-      writeText(this.longtermMd, longterm);
-    }
-    st.lastRolledSeq = maxSeq;
-    this._saveState();
-  }
-
-  // 滚动压缩: 今日事件超量时, 把最旧对话聚合压缩进 longterm, 只留近期
-  // v1.0.7 节流: 压缩后 COMPACT_MIN_INTERVAL_MS 内不重复, 防每轮对话都付 LLM 压缩成本
-  async _compactIfNeeded() {
-    const now = Date.now();
-    if (now - (this._lastCompactAt || 0) < COMPACT_MIN_INTERVAL_MS) return;
-    const lines = _renderLines(this.sessionStore, logicalDay());
-    if (lines.length < COMPACT_THRESHOLD) return;
-    const compactedLines = lines.slice(0, -COMPACT_KEEP);
-    const userMsgs = compactedLines
-      .filter((li) => li.indexOf("用户:") !== -1)
-      .map((li) => li.split("用户:")[1].trim())
-      .filter(Boolean);
-    let summary;
-    if (this.summarizer && compactedLines.length) {
-      try {
-        const raw = compactedLines.slice(0, 60).join("\n");
-        const s = await this.summarizer(raw);
-        summary = "[" + logicalDay() + " llm-summary] " + (s || "(空)");
-      } catch {
-        summary = "[" + logicalDay() + " thin] archived " + compactedLines.length + " lines (llm fail)";
-      }
-    } else if (userMsgs.length) {
-      summary = "[" + logicalDay() + " thin] " + userMsgs.length + " rounds archived: " + userMsgs.slice(0, 12).join(" | ") + (userMsgs.length > 12 ? " | ..." : "");
-    } else {
-      summary = "[" + logicalDay() + " thin] archived " + compactedLines.length + " lines";
-    }
-    let longterm = readText(this.longtermMd) || "";
-    longterm += "\n## " + logicalDay() + " (rollup)\n" + summary + "\n";
-    writeText(this.longtermMd, longterm);
-    this._lastCompactAt = now; // 节流: 压缩完成后记录, 60s 内不再压缩
-  }
-
-  context(userMsg) {
-    const todayCount = this.sessionStore ? _renderLines(this.sessionStore, logicalDay()).length : 0;
-    const rawLongterm = readText(this.longtermMd) || "";
-    const longterm = _longtermExcludingToday(rawLongterm, logicalDay()).slice(-3000);
-    const topFacts = this.factsTop(userMsg);
-    // v1.2.0 fix: 今日对话原文由会话历史(history)承载, 不再逐行重复注入到 system,
-    // 避免模型在 system 的"今日记忆"里看到与 history 相同的对话而重复回话。
-    const todayNote = todayCount
-      ? `今日已进行 ${todayCount} 轮对话（完整内容已随会话历史提供）。`
-      : "今日暂无对话。";
-    return `
-# 今日对话
-${todayNote}
-
-# 长期记忆 (最近)
-${longterm || "(暂无)"}
-
-# 关键事实
-${topFacts || "(暂无)"}
-`;
-  }
-
-  // 关键事实: 按当前问题语义检索优先, 不足 8 条用衰减分补齐 (去重)。
-  // 让每轮自动注入的记忆与当前问题相关, 而非纯衰减取 top (v0.8.1 语义注入)
-  factsTop(userMsg) {
-    try {
-      const q = String(userMsg || "").trim();
-      const semantic = q ? this.factStore.query(q, { limit: 8 }) : [];
-      const decay = this.factStore.query("", { limit: 8 });
-      const seen = new Set();
-      const merged = [];
-      for (const f of [...semantic, ...decay]) {
-        if (seen.has(f.id)) continue;
-        seen.add(f.id);
-        merged.push(f);
-        if (merged.length >= 8) break;
-      }
-      return merged.map((f) => `- [${f.score}] ${f.content}`).join("\n");
-    } catch { return ""; }
-  }
-
-  // 可观测: longterm 大小 + 今日事件数, 供 agent.stats() 聚合
-  stats() {
-    let longtermBytes = 0;
-    try { longtermBytes = fs.statSync(this.longtermMd).size; } catch {}
-    return {
-      longterm_bytes: longtermBytes,
-      events_today: _renderLines(this.sessionStore, logicalDay()).length,
-    };
-  }
-}
-
-// 读取长期记忆, 并剔除"今天"的段落 —— 当日对话已由会话历史(history)承载,
-// 若 longterm 里今天滚动块也注入, 会与 history 重复, 是"重复回话"的诱因。
-function _longtermExcludingToday(text, today) {
-  const out = [];
-  let inToday = false;
-  for (const l of String(text || "").split("\n")) {
-    const m = l.match(/^##\s+(\d{4}-\d{2}-\d{2})/);
-    if (m) { inToday = (m[1] === today); continue; }
-    if (!inToday) out.push(l);
-  }
-  return out.join("\n").trim();
-}
-
-// P1#9: 记忆信号预筛 - 命中关键词或长度信号才触发 LLM 提炼 (省成本)
-function _hasSignal(user, assistant) {
-  const u = String(user || "");
-  const a = String(assistant || "");
-  const text = (u + " " + a).slice(0, 500);
-  if (text.length < 8) return false;
-  const SIGNAL = /(我是|我喜欢|我讨厌|我记得|请记住|记住|偏好|习惯|目标是|我需要|我要|想买|喜欢|不喜欢|重要|关键|待办|计划|打算|希望|擅长|不擅长|生日|地址|电话|邮箱|账号|密码|股票|基金|仓位|止损|止盈|策略|工作|项目|公司|同事|老板|朋友|家人|对象|结婚|买房|买车|考试|学习)/;
-  const isGreeting = /^(你好|在吗|谢谢|好的|嗯|ok|hi|hello|再见|拜拜|哈喽)[!。？?]*$/i.test(u.trim());
-  // 收紧: 无信号关键词时, 仅对真正信息密集的长对话触发 LLM 提炼, 避免普通闲聊累积成本 [复审 P1#9]
-  return !isGreeting && (SIGNAL.test(text) || text.length > 200);
-}
-
+// src/memory/memory-ticker.js - 记忆水位线 (参考 openhanako v4)
+// 架构: 会话事件日志 (SessionStore) 为唯一事实源; 今日视图由它派生
+// 不再独立维护 today.md 对话原文 (原 l0/session/today 三处重复, 已收敛到 session)
+// 今日视图 -> 滚动压缩 -> longterm.md (长期记忆)
+import fs from "node:fs";
+import path from "node:path";
+import { ensureDir, readText, writeText, readJson, writeJson, logicalDay } from "../utils/store.js";
+import { debug } from "../utils/logger.js";
+
+const TURNS_PER_SUMMARY = 10;
+const COMPACT_THRESHOLD = 50;   // 今日事件超此条数触发滚动压缩
+const COMPACT_KEEP = 20;        // 压缩后保留的近期条数
+const COMPACT_MIN_INTERVAL_MS = 60000; // 压缩节流: 压缩后 60s 内不重复 (防每轮对话重复付 LLM 压缩成本, v1.0.7)
+
+// 派生今日视图行: 从 session 事件渲染 (role -> 中文说话人)
+function _renderLines(sessionStore, day) {
+  if (!sessionStore) return [];
+  return sessionStore.eventsByDay(day).map((r) => {
+    const who = r.role === "user" ? "用户" : "皮皮虾";
+    return `- [${new Date(r.timestamp).toISOString()}] ${who}: ${String(r.content).slice(0, 200)}`;
+  });
+}
+
+export class MemoryTicker {
+  constructor(dataDir, factStore, summarizer = null, sessionStore = null) {
+    this.summarizer = summarizer;
+    this.extractor = null; // P1#9: LLM 结构化提炼器 (agent 注入), 提取关键事实/偏好/待办
+    this.dir = path.join(dataDir, "memory");
+    ensureDir(this.dir);
+    ensureDir(path.join(this.dir, "daily"));
+    this.factStore = factStore;
+    this.sessionStore = sessionStore;   // 唯一事实源 (今日视图由它派生)
+    this.todayView = path.join(this.dir, "today.md"); // 兼容路径 (不再作为写入真相)
+    this.longtermMd = path.join(this.dir, "longterm.md");
+    this.stateFile = path.join(this.dir, "daily-state.json");
+    this.state = { day: null, turnCount: 0 };
+    this._lastCompactAt = 0; // 压缩节流时间戳 (v1.0.7)
+    this._loadState();
+    this._rollDay();
+  }
+
+  _loadState() {
+    this.state = readJson(this.stateFile, { day: null, turnCount: 0 });
+  }
+
+  _saveState() {
+    writeJson(this.stateFile, this.state);
+  }
+
+  _rollDay() {
+    const today = logicalDay();
+    if (this.state.day !== today) {
+      if (this.state.day) this._compileDaily();
+      this.state.day = today;
+      this._saveState();
+    }
+  }
+
+  // 跨天: 把上一日事件归档到 daily/ 并滚入 longterm (从 session 派生, 非 today.md)
+  _compileDaily() {
+    const day = this.state.day;
+    const lines = _renderLines(this.sessionStore, day);
+    const dailyFile = path.join(this.dir, "daily", `${day}.md`);
+    writeText(dailyFile, `# ${day}
+
+${lines.join("\n")}\n`);
+    if (lines.length) {
+      let longterm = readText(this.longtermMd) || "";
+      longterm += `\n## ${day}\n${lines.join("\n")}\n`;
+      writeText(this.longtermMd, longterm);
+    }
+  }
+
+  async recordTurn(user, assistant) {
+    this._rollDay();
+    this.state.turnCount += 1;
+    this._saveState();
+    // 对话原文已由 session 事件日志保存; 今日视图由 session 派生
+    if (this.state.turnCount % TURNS_PER_SUMMARY === 0) this._compileDaily_Rolling();
+    await this._compactIfNeeded();
+    // P1#9: 若配 LLM 提炼器且本次对话含高信号, 走结构化提炼; 否则退回启发式
+    if (this.extractor && _hasSignal(user, assistant)) {
+      try {
+        // 感知式提炼: 先检索与本次对话相关的已有记忆 (同主题 top 3), 喂给提炼器从源头去重
+        const related = String(user || "").trim()
+          ? this.factStore.query(user, { limit: 3 }).filter((f) => !(f.hits && f.hits > 5)) // 高命中已稳定, 不再重复提炼
+          : [];
+        const facts = await this.extractor(String(user || ""), String(assistant || ""), related);
+        if (facts && facts.length) {
+          // similarThreshold=0.6: LLM 提炼的字面变体 (同义不同词) 与已有事实语义相似时命中加分,
+          // 防「三件套」这类反复提炼的变体污染记忆库
+          for (const f of facts) this.factStore.add(f, { source: "extract", similarThreshold: 0.6 });
+          return;
+        }
+      } catch (e) { debug(`[memory/memory-ticker] 已忽略异常: ${e && e.message ? e.message : e}`); }
+    }
+    if (user) this.factStore.addMemory(user);
+  }
+
+  // P1#9: 注入 LLM 结构化提炼器 (agent 调 setExtractor)
+  setExtractor(fn) { this.extractor = typeof fn === "function" ? fn : null; }
+
+  // 每 N 轮: 把今日新增事件滚动归档进 longterm (从 session 派生)
+  // v1.2.0 fix: 用游标(lastRolledDay/lastRolledSeq)只追加"本次滚动之后"的新事件,
+  // 不再每次把今日全文重复追加 —— 否则同一段对话会在 longterm 中反复出现, 加剧重复回话。
+  _compileDaily_Rolling() {
+    const today = logicalDay();
+    const st = this.state;
+    if (st.lastRolledDay !== today) { st.lastRolledDay = today; st.lastRolledSeq = 0; } // 跨天重置游标
+    const afterSeq = st.lastRolledSeq || 0;
+    const rows = [];
+    let maxSeq = afterSeq;
+    const dayEvents = this.sessionStore && typeof this.sessionStore.replayDay === "function"
+      ? this.sessionStore.replayDay(today)
+      : [];
+    for (const e of dayEvents) {
+      if (e.seq <= afterSeq) continue;
+      const who = e.type === "user/message" ? "用户" : "皮皮虾";
+      rows.push(`- [${new Date(e.ts).toISOString()}] ${who}: ${String(e.data?.content || "").slice(0, 200)}`);
+      if (e.seq > maxSeq) maxSeq = e.seq;
+    }
+    if (rows.length) {
+      let longterm = readText(this.longtermMd) || "";
+      longterm += `\n## ${today} (滚动)\n${rows.join("\n")}\n`;
+      writeText(this.longtermMd, longterm);
+    }
+    st.lastRolledSeq = maxSeq;
+    this._saveState();
+  }
+
+  // 滚动压缩: 今日事件超量时, 把最旧对话聚合压缩进 longterm, 只留近期
+  // v1.0.7 节流: 压缩后 COMPACT_MIN_INTERVAL_MS 内不重复, 防每轮对话都付 LLM 压缩成本
+  async _compactIfNeeded() {
+    const now = Date.now();
+    if (now - (this._lastCompactAt || 0) < COMPACT_MIN_INTERVAL_MS) return;
+    const lines = _renderLines(this.sessionStore, logicalDay());
+    if (lines.length < COMPACT_THRESHOLD) return;
+    const compactedLines = lines.slice(0, -COMPACT_KEEP);
+    const userMsgs = compactedLines
+      .filter((li) => li.indexOf("用户:") !== -1)
+      .map((li) => li.split("用户:")[1].trim())
+      .filter(Boolean);
+    let summary;
+    if (this.summarizer && compactedLines.length) {
+      try {
+        const raw = compactedLines.slice(0, 60).join("\n");
+        const s = await this.summarizer(raw);
+        summary = "[" + logicalDay() + " llm-summary] " + (s || "(空)");
+      } catch {
+        summary = "[" + logicalDay() + " thin] archived " + compactedLines.length + " lines (llm fail)";
+      }
+    } else if (userMsgs.length) {
+      summary = "[" + logicalDay() + " thin] " + userMsgs.length + " rounds archived: " + userMsgs.slice(0, 12).join(" | ") + (userMsgs.length > 12 ? " | ..." : "");
+    } else {
+      summary = "[" + logicalDay() + " thin] archived " + compactedLines.length + " lines";
+    }
+    let longterm = readText(this.longtermMd) || "";
+    longterm += "\n## " + logicalDay() + " (rollup)\n" + summary + "\n";
+    writeText(this.longtermMd, longterm);
+    this._lastCompactAt = now; // 节流: 压缩完成后记录, 60s 内不再压缩
+  }
+
+  context(userMsg) {
+    const todayCount = this.sessionStore ? _renderLines(this.sessionStore, logicalDay()).length : 0;
+    const rawLongterm = readText(this.longtermMd) || "";
+    const longterm = _longtermExcludingToday(rawLongterm, logicalDay()).slice(-3000);
+    const topFacts = this.factsTop(userMsg);
+    // v1.2.0 fix: 今日对话原文由会话历史(history)承载, 不再逐行重复注入到 system,
+    // 避免模型在 system 的"今日记忆"里看到与 history 相同的对话而重复回话。
+    const todayNote = todayCount
+      ? `今日已进行 ${todayCount} 轮对话（完整内容已随会话历史提供）。`
+      : "今日暂无对话。";
+    return `
+# 今日对话
+${todayNote}
+
+# 长期记忆 (最近)
+${longterm || "(暂无)"}
+
+# 关键事实
+${topFacts || "(暂无)"}
+`;
+  }
+
+  // 关键事实: 按当前问题语义检索优先, 不足 8 条用衰减分补齐 (去重)。
+  // 让每轮自动注入的记忆与当前问题相关, 而非纯衰减取 top (v0.8.1 语义注入)
+  factsTop(userMsg) {
+    try {
+      const q = String(userMsg || "").trim();
+      const semantic = q ? this.factStore.query(q, { limit: 8 }) : [];
+      const decay = this.factStore.query("", { limit: 8 });
+      const seen = new Set();
+      const merged = [];
+      for (const f of [...semantic, ...decay]) {
+        if (seen.has(f.id)) continue;
+        seen.add(f.id);
+        merged.push(f);
+        if (merged.length >= 8) break;
+      }
+      return merged.map((f) => `- [${f.score}] ${f.content}`).join("\n");
+    } catch { return ""; }
+  }
+
+  // 可观测: longterm 大小 + 今日事件数, 供 agent.stats() 聚合
+  stats() {
+    let longtermBytes = 0;
+    try { longtermBytes = fs.statSync(this.longtermMd).size; } catch (e) { debug(`[memory/memory-ticker] 已忽略异常: ${e && e.message ? e.message : e}`); }
+    return {
+      longterm_bytes: longtermBytes,
+      events_today: _renderLines(this.sessionStore, logicalDay()).length,
+    };
+  }
+}
+
+// 读取长期记忆, 并剔除"今天"的段落 —— 当日对话已由会话历史(history)承载,
+// 若 longterm 里今天滚动块也注入, 会与 history 重复, 是"重复回话"的诱因。
+function _longtermExcludingToday(text, today) {
+  const out = [];
+  let inToday = false;
+  for (const l of String(text || "").split("\n")) {
+    const m = l.match(/^##\s+(\d{4}-\d{2}-\d{2})/);
+    if (m) { inToday = (m[1] === today); continue; }
+    if (!inToday) out.push(l);
+  }
+  return out.join("\n").trim();
+}
+
+// P1#9: 记忆信号预筛 - 命中关键词或长度信号才触发 LLM 提炼 (省成本)
+function _hasSignal(user, assistant) {
+  const u = String(user || "");
+  const a = String(assistant || "");
+  const text = (u + " " + a).slice(0, 500);
+  if (text.length < 8) return false;
+  const SIGNAL = /(我是|我喜欢|我讨厌|我记得|请记住|记住|偏好|习惯|目标是|我需要|我要|想买|喜欢|不喜欢|重要|关键|待办|计划|打算|希望|擅长|不擅长|生日|地址|电话|邮箱|账号|密码|股票|基金|仓位|止损|止盈|策略|工作|项目|公司|同事|老板|朋友|家人|对象|结婚|买房|买车|考试|学习)/;
+  const isGreeting = /^(你好|在吗|谢谢|好的|嗯|ok|hi|hello|再见|拜拜|哈喽)[!。？?]*$/i.test(u.trim());
+  // 收紧: 无信号关键词时, 仅对真正信息密集的长对话触发 LLM 提炼, 避免普通闲聊累积成本 [复审 P1#9]
+  return !isGreeting && (SIGNAL.test(text) || text.length > 200);
+}
+