@zaofan/dsh-qqbot 1.4.5 → 1.4.7

package/README_EN.md

@@ -4,6 +4,8 @@ An **enhanced fork** of the QQ Bot IM plugin for [deepseek-harness](https://gith
 
 📦 Repo: [gcry13067381632-jpg/dsh-qqbot](https://github.com/gcry13067381632-jpg/dsh-qqbot) (forked from [tencent-connect/dsh-qqbot](https://github.com/tencent-connect/dsh-qqbot))
 
+> ⭐ **If it works well for you, please star the repo** — it is the only visible signal that someone is using it, and it keeps the project going.
+
 [中文文档](./README.md) | English
 
 ## 🐋 This fork (enhanced edition)
@@ -205,6 +207,66 @@ auto-cancelled when the agent is cancelled or dsh exits.
 > Idea source: QQ-approval design of wang-22-code/dsh-qqbot-bridge (host dsh `approval/request`
 > standard event — the same wiring used by official dsh-acp and the Web approval dialog).
 
+## 🧠 Smart reply (local small model, optional — saves tokens) (v1.4.5+)
+
+Let the plugin score incoming group messages **locally first** ("value scoring"): chatter that scores below the
+threshold **and is not @-mentioning the bot** will **not wake the model at all** → those tokens are simply not spent.
+Messages that @ the bot always pass; **messages with images always pass** (people usually send images *to* her).
+
+- **Model**: `bge-small-zh-v1.5` (Chinese-tuned, quantized ONNX **≈ 23MB**, millisecond-level on CPU, **fully offline**, zero tokens, nothing is uploaded)
+- **Location**: `{DSH_HOME | ~/.dsh}/models/bge-small-zh/` (user-level, shared across workspaces; a custom dir can be set in the config)
+- **Missing model is fine**: scoring silently disables itself and the plugin keeps working (you just don't get the savings)
+
+### ① One command (recommended)
+
+```bash
+node scripts/download-model.mjs                                   # from the source checkout
+node node_modules/@zaofan/dsh-qqbot/scripts/download-model.mjs    # when installed via npm
+```
+
+It prefers the China mirror `hf-mirror.com` and falls back to the official host, verifies file sizes, and prints the next steps.
+
+```bash
+node scripts/download-model.mjs --dir "D:\models\bge-small-zh"   # custom dir → put it in localModel.modelDir
+node scripts/download-model.mjs --source hf                       # force the official host
+```
+
+### ② Manual download (only three files)
+
+Replace `<base>` with `https://hf-mirror.com` or `https://huggingface.co`, and put the files under `~/.dsh/models/bge-small-zh/`:
+
+| Download | Put at | Size |
+|---|---|---|
+| `<base>/Xenova/bge-small-zh-v1.5/resolve/main/onnx/model_quantized.onnx` | `bge-small-zh/onnx/model_quantized.onnx` | ≈ 23MB |
+| `<base>/Xenova/bge-small-zh-v1.5/resolve/main/tokenizer.json` | `bge-small-zh/tokenizer.json` | ≈ 430KB |
+| `<base>/Xenova/bge-small-zh-v1.5/resolve/main/config.json` | `bge-small-zh/config.json` | < 1KB |
+
+### ③ Let your AI do it (paste this to the AI / to her)
+
+```text
+Please install the local small model used by the dsh qqbot plugin on this machine (offline; nothing is uploaded):
+1. Download three files from BAAI/bge-small-zh-v1.5 into `~/.dsh/models/bge-small-zh/`:
+   onnx/model_quantized.onnx (~23MB), tokenizer.json, config.json.
+   Prefer https://hf-mirror.com, fall back to https://huggingface.co.
+2. The layout must be: <model dir>/onnx/model_quantized.onnx, <model dir>/tokenizer.json, <model dir>/config.json
+3. Verify afterwards: model_quantized.onnx ≥ 20MB, tokenizer.json ≥ 300KB.
+   The plugin repo ships a one-shot script: `node scripts/download-model.mjs`
+4. Finally tell me which setting to fill in the plugin settings page, and what the three
+   scoring modes off / log / block actually do.
+```
+
+### Which mode to pick
+
+| Mode | Behaviour | Suggested |
+|---|---|---|
+| `off` | no scoring at all, everything wakes her | not interested |
+| `log` | **records scores but never blocks** | **observation period** — watch the panel for a few days |
+| `block` | below the threshold (default `0.5`) **does not wake** | production use, saves tokens |
+
+- **Per-session override**: the dock's "⚙ Single-session settings" can set mode / threshold / toggle **per group** (stored in `localModel.overrides` in `settings.yaml`), **applied live, no restart needed** (since v1.4.5).
+- **Boundaries**: @-mentions always pass; messages with images always pass; only a dispatch that was blocked **before waking her** (no tokens spent) rolls the group reply-cooldown back — the cooldown exists to save tokens, so already-spent tokens still count.
+- **Score log**: `{dataRoot}/.qqbot/value-scores.jsonl`, one JSON per line with `score / worth / gate / min / conf / mention / img / lib / agg / top`. The panel reads it directly, and the AI can read it to maintain the sample library (`{dataRoot}/.qqbot/value-samples.jsonl`, `{"m":"text","y":1|0}` — y=1 means "she would want to reply").
+
 ## Configuration
 
 | Config | Type | Default | Description |
@@ -348,6 +410,10 @@ export QQBOT_APPID="xxx" QQBOT_SECRET="xxx"
 npx @deepseek-ai/dsh web --patch /path/to/dsh-qqbot/cordis.dev.yml
 ```
 
+## Support
+
+If this plugin saved you some tokens — or made your QQ bot feel alive — a **⭐ Star** is the most concrete way to say thanks. Bugs and feature requests: [Issues](https://github.com/gcry13067381632-jpg/dsh-qqbot/issues).
+
 ## License
 
 [MIT](./LICENSE)

package/client/qqbot-settings.js

@@ -183,7 +183,10 @@ window.__ModuleLoader__.load({
         h('label', { style: labelStyle }, ' 按固定写法匹配文字(高手用,留空=不用): ',
           h('input', { className: 'qqs-inp', style: wideStyle, value: sv(r.conditions.contentRegex), placeholder: '如 ^早安', onChange: (e) => condUpd({ contentRegex: e.target.value }) })),
         h('label', { style: labelStyle }, ' 出现这些词就触发(逗号分隔): ',
-          h('input', { className: 'qqs-inp', style: wideStyle, value: (r.conditions.contentKeywords || []).join(','), onChange: (e) => condUpd({ contentKeywords: e.target.value.split(/[,，]/).map(function (s) { return s.trim() }).filter(Boolean) }) })),
+          // ⚠️ 2026-09-14 修（主人："我输入不了逗号"）：原来是**受控** input（value = 数组 join(',')），
+          //   输入"看不到,"时末尾空项被 filter(Boolean) 丢掉、文本又被重建成没逗号的样子
+          //   → 逗号永远打不进去。改成**非受控**（defaultValue）：输入过程自由，onChange 只把解析结果写进配置。
+          h('input', { className: 'qqs-inp', style: wideStyle, defaultValue: (r.conditions.contentKeywords || []).join(','), onChange: (e) => condUpd({ contentKeywords: e.target.value.split(/[,，]/).map(function (s) { return s.trim() }).filter(Boolean) }) })),
         h('label', { style: labelStyle }, ' 多个条件怎么算: ',
           h('select', { style: inputStyle, value: r.conditions.matchScope || 'any', onChange: (e) => condUpd({ matchScope: e.target.value }) },
             h('option', { value: 'any' }, '满足一个就行'),
@@ -1993,7 +1996,7 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
         return ac ? (ac.dataDir || '') : ''
       }
       function outNsQ() { return state.ns ? 'ns=' + encodeURIComponent(state.ns) : '' }
-      // ⚙ 单会话设置 · 本地小模型(2026-09-13 主人定)
+      // ⚙ 单会话设置 · 智能回复(本地小模型; 2026-09-13 主人定, 15:2x 改名)
       // 📊 评分列表 HTML(单独抽出: 供"局部刷新"只替换内容, 不整块重绘)
       function scoreListHtml(vs) {
         if (vs === undefined || vs === null) return '<div class="dk-empty" style="padding:8px 0">读取中…</div>'
@@ -2005,11 +2008,32 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
           var col = it.worth ? '#2f9e44' : '#c23131'
           out += '<div style="padding:4px 8px;border-bottom:1px solid #f5f5f5;font-size:12px">'
             + '<span style="color:#999">' + hh + '</span> '
-            + '<b style="color:' + col + '">' + (typeof it.score === 'number' ? it.score.toFixed(2) : (it.img ? '📷' : '-')) + '</b>'
+            + '<b style="color:' + col + '">' + (typeof it.score === 'number' ? Number(typeof it.scoreAdj === 'number' ? it.scoreAdj : it.score).toFixed(2) : (it.img ? '📷' : '-')) + '</b>'
+            // 好感偏移标记（2026-09-14：**偏移加在分数上，门槛不动**）—— 文案给普通用户看，写成"基础评分＋好感偏移"
+            + (it.aggWeighted && Array.isArray(it.aggParts) && it.aggParts.length
+              // 聚合加权（2026-09-14 主人定）：判定分是**加权平均**，这里把每一票摊开给人看，
+              //   不然"1.06 是哪来的"没法核对。格式：昵称 价值分×权重(好感占比)
+              ? ' <span style="color:#e8590c;cursor:help" title="' + esc(
+                  '综合分 ' + Number(it.scoreAdj).toFixed(2) + ' ＝ 窗口里 ' + it.aggParts.length + ' 条消息的加权平均'
+                  + '（价值越高、好感越高，权重越大）\n'
+                  + it.aggParts.map(function (p, i) {
+                      return (i + 1) + '. ' + p.n + '：价值 ' + Number(p.s).toFixed(2)
+                        + ' × 权重 ' + Number(p.w).toFixed(2)
+                        + '（好感占比 ' + (Number(p.r) > 0 ? '+' : '') + Number(p.r).toFixed(2) + '）'
+                    }).join('\n')
+                  + '\n记在谁头上：权重最大的那位（当前是 ' + String(it.sender || '?') + '）'
+                ) + '">[合' + it.aggParts.length + '·加权]</span>'
+              : (it.attOff && typeof it.score === 'number'
+                ? ' <span style="color:#e8590c;cursor:help" title="' + esc('基础评分 ' + Number(it.score).toFixed(2) + ' ＋ 好感偏移 ' + (it.attOff > 0 ? '+' : '−') + Math.abs(Number(it.attOff)).toFixed(2) + ' ＝ 有效分 ' + Number(it.score + it.attOff).toFixed(2) + '（好感档位：' + (it.attTier || '?') + '）') + '">[' + (it.attOff > 0 ? '+' : '−') + Math.abs(Number(it.attOff)).toFixed(2) + ']</span>'
+                : ''))
+            // 门槛（用户设置的基础评分门槛，好感度不改它）
+            + (typeof it.min === 'number'
+              ? ' <span style="color:#888" title="' + esc('基础评分门槛 ' + Number(it.min).toFixed(2) + '（你设置的判定线）；好感度以「好感偏移」加减在消息分数上（亲近加分、冷淡减分），不改动门槛') + '">门槛' + Number(it.min).toFixed(2) + '</span>'
+              : '')
             + (it.img ? ' <span style="color:#999" title="图片消息: ' + (it.lib ? '已在库 → 借它的标签当文字评分' : '无文字, 不可评分(默认不拦)') + '">' + (it.lib ? '[库内]' : '[无文字]') + '</span>' : '')
-            + (it.mention ? ' <span style="color:#1c7ed6" title="被@, 必回">[@]</span>' : '')
+            + (it.mention ? ' <span style="color:#1c7ed6" title="' + esc(it.mentionForced && !it.worth ? '被@ → 必回（分数没到门槛也回）' : '被@, 必回') + '">[@]</span>' : '')
             + (it.conf !== undefined && it.conf < 0.5 ? ' <span style="color:#999" title="低置信: 最近邻居相似度只有 ' + it.conf + ', 地图上没这类样本 —— 仅作补样本提示, 不再影响拦截">[?]</span>' : '')
-            + (it.agg ? ' <span style="color:#e8590c" title="聚合了 ' + it.agg + ' 条消息, 取其中最高分">[合' + it.agg + ']</span>' : '')
+            + (it.agg ? ' <span style="color:#e8590c" title="' + esc(it.aggWeighted ? ('聚合了 ' + it.agg + ' 条消息，**加权平均**综合判断（明细见左边的[合·加权]标记）') : ('聚合了 ' + it.agg + ' 条消息, 取其中最高分')) + '">[合' + it.agg + ']</span>' : '')
             + ' <span style="color:#555">' + esc(String(it.sender || '?')) + ':</span> '
             + esc(String(it.text || '').slice(0, 56))
             + '<div style="color:#bbb;font-size:11px;margin-left:14px">近邻: ' + esc((it.top || []).join(' · ')) + '</div>'
@@ -2020,6 +2044,89 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
         return out
       }
       // 局部刷新评分列表(不调 paintBody → 不闪、不打断滚动、不丢输入框焦点)
+      // 熟识度 + 好感度(2026-09-14 主人"直接推进"后分两列展示, 两个维度别混)
+      // ── 分列 + 图标(2026-09-14 主人:"好感度的文字太多了, 改成有分列, 图标显示") ──
+      //   列: # ｜ 群友 ｜ 熟识度 ｜ 好感度 ｜ 这次 ｜ 活跃
+      //   原来一行塞着"(内心亲近 +0.5 / 两好相凑 ×1.2)(来过2天/消息65/点名35/接话3 · 3分钟前)"两串中文,
+      //   现在压成图标 + 数字,**全文退进 title 悬浮提示** —— 想细看还看得到, 只是不再堵眼睛。
+      var AFF_TIER_ICON = { '陌生人': '👤', '眼熟': '👋', '熟人': '🤝' };
+      var AFF_TIER_COLOR = { '陌生人': '#868e96', '眼熟': '#1c7ed6', '熟人': '#2f9e44' };
+      var ATT_TIER_ICON = { '很亲近': '💖', '亲近': '💗', '中立': '😐', '冷淡': '🧊', '疏远': '❄️' };
+      var ATT_TIER_COLOR = { '很亲近': '#e64980', '亲近': '#f06595', '中立': '#868e96', '冷淡': '#4c6ef5', '疏远': '#364fc7' };
+      var DELTA_ICON = {
+        'near': ['💗', '内心亲近 +0.5'],
+        'refuse': ['🚫', '内心拒绝 −0.5'],
+        'yield': ['🕊️', '让步(心里不肯仍照顾) +0.8'],
+        'harsh': ['🥶', '又烦又冷(重罚) −0.5'],
+        'amp:cold-praise': ['❗×1.5', '对方夸她还掉好感(不领情) ×1.5'],
+        'amp:flatter': ['🎁×1.5', '对方冷她还涨好感(讨好) ×1.5'],
+        'amp:both-warm': ['➕×1.2', '两好相凑 ×1.2'],
+        'amp:both-cold': ['❄️×1.2', '两冷相叠 ×1.2']
+      };
+      var AFF_COLS = { n: 20, aff: 76, att: 92, why: 104, act: 150 };
+      function affinityHeadHtml() {
+        var s = function (w, t) { return '<span style="flex:0 0 ' + w + 'px">' + t + '</span>' }
+        return '<div style="display:flex;gap:8px;align-items:center;font-size:11px;color:#adb5bd;padding-bottom:3px;border-bottom:1px solid #e9ecef">'
+          + '<span style="flex:0 0 ' + AFF_COLS.n + 'px;text-align:right">#</span>'
+          + '<span style="flex:1 1 auto;min-width:0">群友</span>'
+          + s(AFF_COLS.aff, '熟识度') + s(AFF_COLS.att, '好感度') + s(AFF_COLS.why, '这次') + s(AFF_COLS.act, '活跃')
+          + '</div>'
+      }
+      function affinityRowHtml(i, x, a) {
+        var age = x.lastAt ? Math.round((Date.now() - x.lastAt) / 60000) : 0
+        var tier = x.tier || '陌生人'
+        var color = AFF_TIER_COLOR[tier] || '#868e96'
+        var affTip = '熟识度 ' + (x.score != null ? x.score : '-') + '/100 · 档位: ' + tier
+          + '(来过 ' + (x.reviews || 0) + ' 天 / 消息 ' + x.msgs + ' / 被点名 ' + x.mentions + ' / 接话 ' + x.replies + ')'
+        var affCell = '<span style="flex:0 0 ' + AFF_COLS.aff + 'px" title="' + esc(affTip) + '">'
+          + '<b style="color:' + color + '">' + (x.score != null ? x.score : '-') + '</b> <span>' + (AFF_TIER_ICON[tier] || '👤') + '</span></span>'
+        var at = (a && a.tier) || ''
+        var aNum = a ? (a.a >= 0 ? '+' : '') + Number(a.a).toFixed(2) : '—'
+        var aColor = !a ? '#adb5bd' : (a.a > 0.0001 ? '#2f9e44' : (a.a < -0.0001 ? '#e03131' : '#868e96'))
+        var aTip = a
+          ? ('好感度 ' + a.a + '(占范围 ' + (a.ratio != null ? (a.ratio * 100).toFixed(1) + '%' : '?') + ') · 档位: ' + (at || '中立')
+            + (a.why ? '；最近一次: ' + a.why : ''))
+          : '还没有好感度记录(她还没对他形成态度)'
+        var aCell = '<span style="flex:0 0 ' + AFF_COLS.att + 'px" title="' + esc(aTip) + '">'
+          + '<b style="color:' + aColor + '">' + aNum + '</b> <span style="color:' + (ATT_TIER_COLOR[at] || '#adb5bd') + '">' + (ATT_TIER_ICON[at] || '⚪') + '</span></span>'
+        var codes = (a && a.codes && a.codes.length) ? a.codes : []
+        var marks = codes.map(function (c) {
+          var d = DELTA_ICON[c]
+          return d ? '<span title="' + esc(d[1]) + '">' + d[0] + '</span>' : ''
+        }).filter(Boolean)
+        var whyCell = '<span style="flex:0 0 ' + AFF_COLS.why + 'px" title="' + esc((a && a.why) || '暂无拆解') + '">'
+          + (marks.length ? marks.join(' ') : '<span style="color:#ced4da">—</span>') + '</span>'
+        var actTip = '来过 ' + (x.reviews || 0) + ' 天 · 消息 ' + x.msgs + ' · 被点名 ' + x.mentions + ' 次 · 接她的话 ' + x.replies + ' 次'
+          + (age < 120 ? ' · 最近 ' + age + ' 分钟前' : '')
+        var actCell = '<span style="flex:0 0 ' + AFF_COLS.act + 'px;color:#adb5bd;font-size:11px" title="' + esc(actTip) + '">'
+          + '📅' + (x.reviews || 0) + ' 💬' + x.msgs + ' 📣' + x.mentions + ' ↩' + x.replies
+          + (age < 120 ? ' <span style="color:#868e96">🕐' + age + '分</span>' : '') + '</span>'
+        return '<div style="display:flex;gap:8px;align-items:center;font-size:12px;color:#495057;padding:2px 0;border-bottom:1px dashed #f1f3f5">'
+          + '<span style="flex:0 0 ' + AFF_COLS.n + 'px;text-align:right;color:#adb5bd">' + (i + 1) + '</span>'
+          + '<span style="flex:1 1 auto;min-width:0;overflow:hidden;text-overflow:ellipsis;white-space:nowrap;color:#343a40" title="' + esc(x.key) + '">' + esc(x.name || x.key) + '</span>'
+          + affCell + aCell + whyCell + actCell + '</div>'
+      }
+      function loadAffinity() {
+        if (state.tab !== 'session') return
+        var box = panel ? panel.querySelector('#dk-aff-box') : null
+        var q = state.ns ? '?' + outNsQ() : ''
+        Promise.all([
+          fetch('/api/qqbot-settings/affinity' + q).then(function (r) { return r.json() }).catch(function () { return {} }),
+          fetch('/api/qqbot-settings/attitude' + q).then(function (r) { return r.json() }).catch(function () { return {} })
+        ]).then(function (rs) {
+          var aff = (rs[0] && rs[0].items) || []
+          var att = (rs[1] && rs[1].items) || []
+          var attBy = {}
+          att.forEach(function (x) { attBy[x.key] = x })
+          if (aff.length === 0) { state.affHtml = '还没有数据 —— 群里聊几句就出来了'; }
+          else {
+            state.affHtml = affinityHeadHtml() + aff.map(function (x, i) {
+              return affinityRowHtml(i, x, attBy[x.key])
+            }).join('')
+          }
+          if (box) box.innerHTML = state.affHtml
+        }).catch(function () { if (box) box.textContent = '熟识度/好感度读取失败' })
+      }
       function refreshScoreList() {
         if (state.tab !== 'session') return
         fetch('/api/qqbot-settings/value-scores' + scoreQuery())
@@ -2106,6 +2213,7 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
           .then(function (r) { return r.json() })
           .then(function (d) { state.valueScores = d && d.ok ? d : { items: [], total: 0 }; paintBody() })
           .catch(function () { state.valueScores = { items: [], total: 0 }; paintBody() })
+          .then(function () { loadAffinity() })
           .then(function () { state.lmBusy = false })
         // 每 10 秒局部刷新一次评分列表(只换 #dk-vs-box 内容, 不整块重绘 → 不闪)
         if (state.lmTimer) clearInterval(state.lmTimer)
@@ -2222,6 +2330,35 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
           vsToggle.textContent = state.vsOpen ? '收起' : '展开编辑'
           if (state.vsOpen) loadValueSamplesEditor()
         }
+        // 📊 最近评分 折叠（2026-09-14 主人要求：这两块默认收起，别占地方）
+        var vsViewToggle = panel.querySelector('#dk-vs-view-toggle')
+        if (vsViewToggle) vsViewToggle.onclick = function () {
+          state.vsViewOpen = !state.vsViewOpen
+          var b = panel.querySelector('#dk-vs-box')
+          if (b) b.style.display = state.vsViewOpen ? 'block' : 'none'
+          vsViewToggle.textContent = state.vsViewOpen ? '收起' : '展开'
+        }
+        // 🧠 熟识度 / 💗 好感度 折叠
+        var affToggle = panel.querySelector('#dk-aff-toggle')
+        if (affToggle) affToggle.onclick = function () {
+          state.affOpen = !state.affOpen
+          var b = panel.querySelector('#dk-aff-box')
+          if (b) b.style.display = state.affOpen ? 'block' : 'none'
+          affToggle.textContent = state.affOpen ? '收起' : '展开'
+        }
+        // 📊 导出 Excel（2026-09-14 主人要"一键导出分享"）—— 直接走 GET，让浏览器下载
+        var affExport = panel.querySelector('#dk-aff-export')
+        if (affExport) affExport.onclick = function () {
+          var url = '/api/qqbot-settings/export.xlsx' + (state.ns ? '?' + outNsQ() : '')
+          var a = document.createElement('a')
+          a.href = url
+          a.download = ''
+          document.body.appendChild(a)
+          a.click()
+          document.body.removeChild(a)
+          affExport.textContent = '📊 已导出'
+          setTimeout(function () { affExport.textContent = '📊 导出 Excel' }, 2500)
+        }
         var vsSaveBtn = panel.querySelector('#dk-vs-save')
         if (vsSaveBtn) vsSaveBtn.onclick = function () { saveValueSamplesEditor() }
         // 🤖 让ai写(2026-09-13 主人要的): 模拟一条用户消息唤醒她 → 她自己读聊天记录写样例
@@ -2242,7 +2379,7 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
             if (h) h.textContent = d && d.ok ? ('✓ ' + d.msg) : ('失败: ' + ((d && d.error) || (d && d.msg) || '未知'))
           }).catch(function (e) { if (h) h.textContent = '异常: ' + e.message })
         }
-        // ⚙ 单会话设置: 次级标签切换(🧠 本地小模型 / 📄 空白样板)
+        // ⚙ 单会话设置: 次级标签切换(🧠 智能回复 / 📄 空白样板)
         panel.querySelectorAll('[data-sttab]').forEach(function (sb) {
           sb.onclick = function () {
             var v = sb.getAttribute('data-sttab')
@@ -3094,11 +3231,11 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
             }).join('')
           + '</div>'
       }
-      // ⚙ 单会话设置 · 次级标签栏(2026-09-13 主人要求: 本地小模型独立成子标签, 另留一个空标签当样板)
+      // ⚙ 单会话设置 · 次级标签栏(2026-09-13 主人要求: 智能回复独立成子标签, 另留一个空标签当样板)
       function sessionSubTabs() {
         var st = state.stTab || 'lm'
         return '<div class="dk-row" style="gap:4px;margin:4px 0 6px;flex-wrap:wrap;border-bottom:1px dashed #e2d9ff;padding-bottom:6px">'
-          + [['lm', '🧠 本地小模型'], ['blank', '📄 空白样板']]
+          + [['lm', '🧠 智能回复'], ['blank', '📄 空白样板']]
             .map(function (s) {
               return '<button class="dk-btn' + (st === s[0] ? ' on' : '') + '" data-sttab="' + s[0] + '">' + s[1] + '</button>'
             }).join('')
@@ -3352,7 +3489,7 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
           body += '<span class="dk-msg" id="dk-out-hint" style="color:#2f9e44;margin:4px 0"></span>'
         } else if (state.tab === 'session') {
           // ⚙ 单会话设置(2026-09-13 主人定): 放"每个会话单独生效"的设置。
-          // 次级标签: 🧠 本地小模型 / 📄 空白样板(留作以后加新设置的模板)。
+          // 次级标签: 🧠 智能回复 / 📄 空白样板(留作以后加新设置的模板)。
           body += sessionSubTabs()
           var st = state.stTab || 'lm'
           if (st === 'blank') {
@@ -3368,7 +3505,7 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
             body += '<div class="dk-empty">读取中…(若长时间不动, 点「🔄 重新检测」)</div>'
           } else {
             var lmMb = lm.bytes ? (Math.round((lm.bytes / 1048576) * 10) / 10) : 0
-            var lmState = lm.available ? ('✅ 已就绪' + (lmMb ? ' · ' + lmMb + ' MB' : '')) : ('⚠️ 未安装(缺 ' + (lm.missing || []).length + ' 个文件)')
+            var lmState = lm.available ? ('✅ 已就绪' + (lmMb ? ' · ' + lmMb + ' MB' : '')) : ('⚠️ 未安装(缺 ' + (lm.missing || []).length + ' 个文件) —— 点右边「⬇️ 下载模型」即可')
             // ⚠️ 2026-09-13 会话级: 评分模式/门槛/启用 **优先取当前会话的覆盖项**(overrides[group:xxx]),
             //    没有则继承账号级默认; 界面用"继承/已单独设置"标出来。
             var lmCfg = state.lmCfg || {}
@@ -3382,7 +3519,7 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
                 : (typeof lm.valueMinScore === 'number' ? lm.valueMinScore : 0.5))
             var onV = (typeof ov.enabled === 'boolean') ? ov.enabled
               : ((typeof lmCfg.enabled === 'boolean') ? lmCfg.enabled : (lm.enabled !== false))
-            body += '<div class="dk-row" style="font-weight:700;font-size:13px;margin:10px 0 2px">🧠 本地小模型 · bge-small-zh-v1.5</div>'
+            body += '<div class="dk-row" style="font-weight:700;font-size:13px;margin:10px 0 2px">🧠 智能回复（本地小模型） · bge-small-zh-v1.5</div>'
             // 本会话标识 + 继承状态(让"单会话"名副其实: 一眼看出这套值是谁的)
             body += '<div class="dk-row" style="gap:8px;align-items:center;flex-wrap:wrap;margin:0 0 4px">'
               + '<span class="dk-msg" style="font-size:12px;color:#7c6bd6">本会话: ' + esc(curHitLabel()) + '</span>'
@@ -3391,14 +3528,14 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
               + (hasOv ? '<button class="dk-btn" id="dk-lm-reset" style="font-size:11px;padding:1px 6px" title="删除本会话的单独设置, 恢复继承账号默认">↩ 恢复继承</button>' : '')
               + '</div>'
             body += '<div class="dk-row" style="gap:8px;flex-wrap:wrap;align-items:center">'
-              + '<label style="display:inline-flex;align-items:center;gap:4px;font-size:12px;cursor:pointer"><input type="checkbox" id="dk-lm-on"' + (onV ? ' checked' : '') + '> 启用(省 token)</label>'
+              + '<label style="display:inline-flex;align-items:center;gap:4px;font-size:12px;cursor:pointer"><input type="checkbox" id="dk-lm-on"' + (onV ? ' checked' : '') + '> 启用智能回复(省 token)</label>'
               + '<span class="dk-msg" style="font-size:12px;color:' + (lm.available ? '#2f9e44' : '#c23131') + '">' + esc(lmState) + '</span>'
               + '</div>'
             body += '<div class="dk-row" style="gap:6px;margin:4px 0;flex-wrap:wrap">'
               + '<input class="qqs-inp" id="dk-lm-dir" style="flex:1;min-width:200px" placeholder="模型目录(留空=默认 ~/.dsh/models/bge-small-zh)" value="' + esc(lm.modelDir || '') + '">'
               + '<button class="dk-btn" id="dk-lm-save">💾 保存</button>'
               + '<button class="dk-btn" id="dk-lm-check">🔄 重新检测</button>'
-              + (lm.available ? '' : '<button class="dk-btn ok" id="dk-lm-dl">⬇️ 下载模型</button>')
+              + (lm.available ? '' : '<button class="dk-btn ok" id="dk-lm-dl">⬇️ 下载模型(自动)</button>')
               + '</div>'
             // 评分模式 + 门槛(2026-09-13): off=不评分 / log=只记录(默认观察期) / block=低分不唤醒
             body += '<div class="dk-row" style="gap:8px;flex-wrap:wrap;align-items:center;margin:6px 0 2px">'
@@ -3415,11 +3552,21 @@ var QQS_CSS = ".qqs-btn{font:inherit;color:#333;background:linear-gradient(180de
             // (只换这个 div 的内容, 不整块 paintBody) → 既不闪, 又能自动更新。
             body += '<div class="dk-row" style="font-weight:700;font-size:13px;margin:10px 0 2px;gap:8px;align-items:center;flex-wrap:wrap">'
               + '<span>📊 最近评分(观察期)</span>'
+              + '<button class="dk-btn" id="dk-vs-view-toggle" style="font-size:12px;padding:1px 8px">' + (state.vsViewOpen ? '收起' : '展开') + '</button>'
               + '<span class="dk-msg" style="font-size:11px;color:#7c6bd6" id="dk-vs-scope">' + esc(curHitLabel()) + '</span>'
               + '<button class="dk-btn" id="dk-vs-reload" style="font-size:12px;padding:1px 8px">🔄 刷新</button>'
               + '<span class="dk-msg" style="font-size:11px;color:#999">每 10 秒自动更新</span>'
               + '</div>'
-            body += '<div id="dk-vs-box">' + scoreListHtml(state.valueScores) + '</div>'
+            body += '<div id="dk-vs-box" style="display:' + (state.vsViewOpen ? 'block' : 'none') + '">' + scoreListHtml(state.valueScores) + '</div>'
+            // 💗 熟识度(她记你多牢) + 好感度(她对你什么态度) —— 两个维度分两列, 别混
+            //   熟识度: {dataRoot}/.qqbot/affinity.json ｜ 好感度: {dataRoot}/.qqbot/attitude.json
+            body += '<div class="dk-row" style="font-weight:700;font-size:13px;margin:10px 0 2px;gap:8px;align-items:center;flex-wrap:wrap">'
+              + '<span>🧠 熟识度(记你多牢) ｜ 💗 好感度(对你什么态度)(观察期·只统计)</span>'
+              + '<button class="dk-btn" id="dk-aff-export" style="font-size:12px;padding:1px 8px" title="导成 Excel（三张表：熟识度 / 好感度 / 口径说明），可以直接转发给别人">📊 导出 Excel</button>'
+              + '<button class="dk-btn" id="dk-aff-toggle" style="font-size:12px;padding:1px 8px">' + (state.affOpen ? '收起' : '展开') + '</button>'
+              + '</div>'
+            // 分列 + 图标(2026-09-14): 注意这里**不能**再加 white-space:pre-wrap —— 现在是逐行的 flex 布局 HTML
+            body += '<div class="dk-msg" id="dk-aff-box" style="font-size:12px;color:#666;display:' + (state.affOpen ? 'block' : 'none') + '">' + (state.affHtml || '加载中…') + '</div>'
             // ✍️ 样例库编辑(2026-09-13 主人问"哪里写样本"): 直接改 jsonl, 保存后下一条消息生效
             body += '<div class="dk-row" style="font-weight:700;font-size:13px;margin:12px 0 2px;gap:8px;align-items:center;flex-wrap:wrap">'
               + '<span>✍️ 样例库(决定她的开口标准)</span>'

package/dist/channel-tools.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"channel-tools.d.ts","sourceRoot":"","sources":["../src/channel-tools.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,qBAAqB,CAAC;AAGnD,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,8BAA8B,CAAC;AACnE,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,gCAAgC,CAAC;AAyElE,6DAA6D;AAC7D,MAAM,WAAW,SAAS;IACxB,OAAO,EAAE,cAAc,CAAC;IACxB,MAAM,EAAE,WAAW,CAAC;CACrB;AAED,eAAO,MAAM,IAAI,wBAAwB,CAAC;AAC1C,eAAO,MAAM,MAAM,UAAY,CAAC;AAShC,wBAAgB,gBAAgB,CAAC,CAAC,EAAE,SAAS,GAAG,SAAS,GAAG,IAAI,CAI/D;AAkFD,mDAAmD;AACnD,wBAAsB,KAAK,CAAC,GAAG,EAAE,OAAO,GAAG,OAAO,CAAC,IAAI,CAAC,CAivDvD;AAED;;;;;;;;;;GAUG;AACH,wBAAsB,mBAAmB,CACvC,EAAE,EAAE,MAAM,EACV,KAAK,EAAE,OAAO,GAAG,KAAK,EACtB,MAAM,EAAE,MAAM,GACb,OAAO,CAAC;IAAE,EAAE,EAAE,OAAO,CAAC;IAAC,GAAG,EAAE,MAAM,CAAA;CAAE,CAAC,CAwCvC"}
+{"version":3,"file":"channel-tools.d.ts","sourceRoot":"","sources":["../src/channel-tools.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,qBAAqB,CAAC;AAGnD,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,8BAA8B,CAAC;AACnE,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,gCAAgC,CAAC;AA2ElE,6DAA6D;AAC7D,MAAM,WAAW,SAAS;IACxB,OAAO,EAAE,cAAc,CAAC;IACxB,MAAM,EAAE,WAAW,CAAC;CACrB;AAED,eAAO,MAAM,IAAI,wBAAwB,CAAC;AAC1C,eAAO,MAAM,MAAM,UAAY,CAAC;AAShC,wBAAgB,gBAAgB,CAAC,CAAC,EAAE,SAAS,GAAG,SAAS,GAAG,IAAI,CAI/D;AAkFD,mDAAmD;AACnD,wBAAsB,KAAK,CAAC,GAAG,EAAE,OAAO,GAAG,OAAO,CAAC,IAAI,CAAC,CAk1DvD;AAED;;;;;;;;;;GAUG;AACH,wBAAsB,mBAAmB,CACvC,EAAE,EAAE,MAAM,EACV,KAAK,EAAE,OAAO,GAAG,KAAK,EACtB,MAAM,EAAE,MAAM,GACb,OAAO,CAAC;IAAE,EAAE,EAAE,OAAO,CAAC;IAAC,GAAG,EAAE,MAAM,CAAA;CAAE,CAAC,CAwCvC"}

package/dist/channel-tools.js

@@ -22,6 +22,8 @@ import * as broadcastQueue from './features/broadcast.js';
 import { wakeSessionAgent, safeAppendUserMessage } from './features/group-hub.js';
 import { managersOf, findManagerByPeer, findManagerBySessionId } from './features/session-registry.js';
 import { handleInbound } from './transport/inbound.js';
+import { getQuote } from './features/quote-cache.js';
+import { appendMemoLine, deleteMemo, listMemos, readMemo } from './features/people-memo.js';
 /** 诊断日志路径: 默认关闭; 需要排查时设环境变量 QQBOT_DIAG_FILE 指向日志文件 */
 const DIAG_FILE = process.env.QQBOT_DIAG_FILE || '';
 /** 大文件异步发送阈值: 本地文件 >= 5MB 时走后台任务(SDK 分片上传耗时, 避免阻塞 LLM 回合) */
@@ -770,7 +772,8 @@ export async function apply(ctx) {
     // ⚠️ 描述只讲两个分支的效果, 不给场景规则——选哪个由模型按当前对话氛围自行权衡(写死规则会导致静默率过高)。
     const replyGateTool = defineTool({
         name: 'reply_gate',
-        description: '回复闸门: 开始回复前先调用它, 决定本回合开口还是静默。传 reply:true → 正常继续, 按当前对话决定说什么; 传 reply:false → 本回合立即静默终止, 你不会再输出任何内容(这条消息就像没发生过)。选 true 还是 false 由你根据当前对话自己判断: 觉得该回应就 true, 觉得安静待着更好就 false。',
+        description: '回复闸门: 开始回复前先调用它, 决定本回合开口还是静默。传 reply:true → 正常继续, 按当前对话决定说什么; 传 reply:false → 本回合立即静默终止, 你不会再输出任何内容(这条消息就像没发生过)。选 true 还是 false 由你根据当前对话自己判断: 觉得该回应就 true, 觉得安静待着更好就 false。' +
+            '\n⚠️ 传 reply:false 后**不要再调用任何工具**(包括记笔记、记喜好、发表情包): 本回合已经关闭, 那些调用会被直接中止 —— 想记录的话, 留到下一次开口的回合里做。',
         parameters: {
             reply: { type: 'boolean', required: true, description: 'true=正常开口回复; false=静默终止本回合(不再输出任何内容)' },
             reason: { type: 'string', required: true, description: '简短判定理由, 如"未@不插嘴" / "被@需回复" / "私聊"' },
@@ -2018,11 +2021,113 @@ export async function apply(ctx) {
             }
         },
     });
+    // quote_view: 查看引用内容(2026-09-13 主人定, 省 token) ——
+    //   引用原文超过 60 字时, 入站上下文只给**前 60 字**, 全文落在
+    //   `{dataRoot}/.qqbot/quote-cache.json`(**每会话最多 10 条**), 需要时用本工具取完整原文。
+    //   ⚠️ 工具描述**刻意写四个字**(主人要求): 描述进 system prompt, 越短越省。
+    const quoteViewTool = defineTool({
+        name: 'quote_view',
+        description: '查看引用内容',
+        parameters: {
+            idx: { type: 'integer', description: '引用编号(默认最新一条)' },
+        },
+        output: {
+            schema: {
+                type: 'object', additionalProperties: false,
+                properties: {
+                    ok: { type: 'boolean', required: true },
+                    msg: { type: 'string', required: true },
+                    text: { type: 'string' },
+                },
+            },
+            render: (_a, v) => [
+                { type: 'text', text: v.ok ? (v.text || '') : `查询失败: ${v.msg}` },
+            ],
+        },
+        async execute(args, exec) {
+            try {
+                const s = findSessionRec(channelOf(exec), exec);
+                const mgr = s?.ch?.manager;
+                const rec = s?.rec;
+                const dataRoot = String(mgr?.dataRoot || '');
+                if (!dataRoot || !rec?.scope || !rec?.peerId)
+                    return { ok: false, msg: '未找到当前 QQ 会话' };
+                const rawIdx = args.idx;
+                const idx = typeof rawIdx === 'number' && Number.isFinite(rawIdx) ? rawIdx : undefined;
+                const e = getQuote(dataRoot, `${rec.scope}:${rec.peerId}`, idx);
+                if (!e)
+                    return { ok: false, msg: '没有缓存的引用(短引用不缓存; 或这条引用是更早的、已被 10 条上限挤掉)' };
+                return { ok: true, msg: `引用#${e.id}`, text: `引用#${e.id}${e.sender ? `（${e.sender}）` : ''}: ${e.text}` };
+            }
+            catch (err) {
+                return { ok: false, msg: err instanceof Error ? err.message : String(err) };
+            }
+        },
+    });
+    // people_memo: 群友小传（2026-09-13 主人定）—— 文字版好感度的本体。
+    //   记一行(同一人一天最多一条) / 查看 / 删除（群里一句"别记人家"就删）。描述刻意写短。
+    const peopleMemoTool = defineTool({
+        name: 'people_memo',
+        description: '群友小传: 记一行/查看/删除。当群友说出自己的喜好/经历/身份等"旁人能复述的事实"时, 顺手记一条(同一人一天最多 3 条); 想归到八栏就带前缀, 如"喜好：最近在玩XX"(不带前缀则进记事区); 只写事实, 不写评价与猜测',
+        parameters: {
+            who: { type: 'string', required: true, description: '群友昵称或 openid' },
+            line: { type: 'string', description: '要记的一行; 不传=只看' },
+            del: { type: 'boolean', description: 'true=删掉这个人的小传' },
+        },
+        output: {
+            schema: {
+                type: 'object', additionalProperties: false,
+                properties: { ok: { type: 'boolean', required: true }, msg: { type: 'string', required: true } },
+            },
+            render: (_a, v) => [{ type: 'text', text: v.msg }],
+        },
+        async execute(args, exec) {
+            try {
+                const a = args;
+                const s = findSessionRec(channelOf(exec), exec);
+                const mgr = s?.ch?.manager;
+                const rec = s?.rec;
+                const dataRoot = String(mgr?.dataRoot || '');
+                if (!dataRoot)
+                    return { ok: false, msg: '未找到当前会话的数据根' };
+                // who → openid: 优先在群成员台账里按昵称找
+                let key = String(a.who || '').trim();
+                if (!/^[0-9A-Fa-f]{16,}$/.test(key) && rec?.scope === 'group' && rec.peerId) {
+                    try {
+                        const dataDir = join(dataRoot, '表情包');
+                        const mems = readGroupMembers(dataDir, rec.peerId) || [];
+                        const hit = mems.find((m) => m && m.name === key) || mems.find((m) => m && typeof m.name === 'string' && m.name.includes(key));
+                        if (hit && hit.mid)
+                            key = hit.mid;
+                    }
+                    catch { /* 用昵称兜底当 key */ }
+                }
+                if (a.del) {
+                    const r = deleteMemo(dataRoot, key);
+                    return { ok: r.ok, msg: r.ok ? '好, 已经把 ' + a.who + ' 那份小传删掉了, 以后也不记他' : r.msg };
+                }
+                if (a.line) {
+                    const r = appendMemoLine(dataRoot, key, a.line, { name: a.who, force: false });
+                    return { ok: r.ok, msg: r.ok ? '记下了: ' + a.line.slice(0, 60) : r.msg };
+                }
+                const cur = readMemo(dataRoot, key);
+                if (cur)
+                    return { ok: true, msg: cur.slice(0, 1200) };
+                const all = listMemos(dataRoot);
+                return { ok: true, msg: '还没有 ' + a.who + ' 的小传。现有 ' + all.length + ' 份: ' + all.slice(0, 8).map((x) => x.key).join(', ') };
+            }
+            catch (e) {
+                return { ok: false, msg: e instanceof Error ? e.message : String(e) };
+            }
+        },
+    });
     // 逐个注册并记录结果(便于线上定位是哪个工具失败)
     const toolDefs = [
         { name: 'send_media', tool: sendMediaTool },
         { name: 'recall_message', tool: recallTool },
         { name: 'text_break', tool: textBreakTool },
+        { name: 'quote_view', tool: quoteViewTool },
+        { name: 'people_memo', tool: peopleMemoTool },
         { name: 'tools_reload', tool: toolsReloadTool },
         { name: 'reply_gate', tool: replyGateTool },
         { name: 'outbound_mode', tool: outboundModeTool },