mingdao-harness 0.3.0 → 0.3.2

package/src/web/index.html

@@ -31,20 +31,23 @@ body{margin:0;background:var(--bg);color:var(--text);font:15px/1.65 -apple-syste
 body.traj-open{padding-left:294px}
 body.sub-open{padding-right:320px}
 body.tasks-open{padding-right:280px}
+body.dash-open{padding-right:400px}
 @media (max-width:1200px){
   #trajPanel{width:210px}body.traj-open{padding-left:252px}
   #subPanel{width:270px}body.sub-open{padding-right:270px}
   #tasksPanel{width:230px}body.tasks-open{padding-right:230px}
+  #dashPanel{width:360px}body.dash-open{padding-right:360px}
 }
 @media (max-width:860px){
   #trajPanel{width:160px}body.traj-open{padding-left:202px}
   #subPanel{width:190px}body.sub-open{padding-right:190px}
   #tasksPanel{width:190px}body.tasks-open{padding-right:190px}
+  #dashPanel{width:min(340px,94vw)}body.dash-open{padding-right:min(340px,94vw)}
 }
 @media (max-width:560px){
-  body.traj-open,body.sub-open,body.tasks-open{padding:0}
+  body.traj-open,body.sub-open,body.tasks-open,body.dash-open{padding:0}
   #trajPanel{width:calc(100vw - 42px);left:42px}
-  #subPanel,#tasksPanel{width:100vw}
+  #subPanel,#tasksPanel,#dashPanel{width:100vw}
 }
 header{min-height:52px;display:flex;align-items:center;gap:10px;padding:8px 12px;border-bottom:1px solid var(--border);background:var(--bg2);flex:none;flex-wrap:nowrap;white-space:nowrap;overflow-x:auto}
 header .logo{font-weight:700;font-size:16px;color:var(--accent);letter-spacing:.5px;flex:none}
@@ -229,6 +232,36 @@ footer{flex:none;border-top:1px solid var(--border);background:var(--bg2);paddin
 @keyframes spin{to{transform:rotate(360deg)}}
 /* 自绘悬浮气泡 tooltip（替换原生 title）：暗色悬浮层 + 阴影，跟随锚点居中，越界自动回弹 */
 .mdtip{position:fixed;z-index:1000;max-width:320px;background:var(--bg3);color:var(--text);border:1px solid var(--border);border-radius:8px;padding:7px 11px;font-size:12px;line-height:1.6;box-shadow:0 8px 24px rgba(0,0,0,.55);pointer-events:none;white-space:pre-line;word-break:break-word;display:none}
+/* —— 省钱仪表盘（v0.3.1 顶部入口，真·仪表盘） —— */
+#dashPanel{position:fixed;right:0;top:52px;bottom:0;width:400px;max-width:94vw;background:var(--bg2);border-left:1px solid var(--border);z-index:45;display:flex;flex-direction:column;box-shadow:-8px 0 32px rgba(0,0,0,.4)}
+.dash-head{padding:12px 14px;font-size:14px;font-weight:700;color:var(--accent);border-bottom:1px solid var(--border);display:flex;align-items:center;gap:8px}
+.dash-head .dash-sub{font-size:11.5px;color:var(--faint);font-weight:400}
+.dash-head button{margin-left:auto;padding:2px 8px;font-size:12px}
+.dash-body{flex:1;overflow-y:auto;padding:14px}
+.dash-kpi{display:grid;grid-template-columns:1fr 1fr;gap:8px;margin-bottom:14px}
+.kpi{background:var(--bg3);border:1px solid var(--border);border-radius:10px;padding:10px 12px;display:flex;flex-direction:column;gap:2px}
+.kpi-l{font-size:11px;color:var(--faint)}
+.kpi-v{font-size:17px;font-weight:700;font-variant-numeric:tabular-nums}
+.dash-gauge-wrap{position:relative;width:120px;margin:4px auto 14px;text-align:center}
+.gauge-label{position:absolute;inset:0;display:flex;flex-direction:column;align-items:center;justify-content:center;gap:0}
+.gauge-label b{font-size:20px;color:var(--accent2)}
+.gauge-label span{font-size:10.5px;color:var(--faint)}
+.dash-sec{font-size:11.5px;color:var(--faint);margin:12px 0 6px;border-top:1px dashed var(--border);padding-top:10px}
+.dash-sec:first-of-type{border-top:none;padding-top:0}
+.dash-chart{background:var(--bg3);border:1px solid var(--border);border-radius:10px;padding:8px}
+.dash-bars{display:flex;flex-direction:column;gap:6px}
+.dbar{background:var(--bg3);border:1px solid var(--border);border-radius:8px;padding:6px 10px}
+.dbar .t{display:flex;justify-content:space-between;font-size:11.5px;margin-bottom:4px}
+.dbar .t .n{color:var(--text);overflow:hidden;text-overflow:ellipsis;white-space:nowrap;max-width:60%}
+.dbar .t .v{color:var(--dim);font-variant-numeric:tabular-nums}
+.dbar .track{height:5px;border-radius:3px;background:var(--bg2);overflow:hidden}
+.dbar .fill{height:100%;border-radius:3px;background:linear-gradient(90deg,var(--accent2),var(--accent))}
+.dash-recent{display:flex;flex-direction:column;gap:5px;font-size:11.5px}
+.dr{display:flex;justify-content:space-between;background:var(--bg3);border:1px solid var(--border);border-radius:8px;padding:5px 10px}
+.dr .m{color:var(--text);overflow:hidden;text-overflow:ellipsis;white-space:nowrap;max-width:50%}
+.dr .p{color:var(--accent2);font-variant-numeric:tabular-nums}
+.dr .c{color:var(--faint);font-variant-numeric:tabular-nums}
+.badge-btn:hover{border-color:var(--accent2);color:var(--accent2)}
 </style>
 </head>
 <body>
@@ -237,7 +270,7 @@ footer{flex:none;border-top:1px solid var(--border);background:var(--bg2);paddin
   <span id="cfgMsg" class="badge" style="display:none"></span>
   <button id="cfgBtn" title="设置">⚙</button>
   <select id="wsSel" title="工作空间（切换 / 新建，目录自动创建）"></select>
-  <span class="badge" id="envBadge"></span><span class="badge" id="costBadge" title="今日费用 / 缓存命中率（每 15s 刷新）"></span>
+  <span class="badge" id="envBadge"></span><span class="badge badge-btn" id="costBadge" title="省钱仪表盘（点击展开）" style="cursor:pointer">📊 —</span>
   <div class="spacer"></div>
   <input id="sessionSearch" placeholder="搜索会话…" title="关键词检索历史会话">
   <select id="sessions" title="历史会话"><option value="">历史会话</option></select>
@@ -260,6 +293,29 @@ footer{flex:none;border-top:1px solid var(--border);background:var(--bg2);paddin
   <div class="tp-head">任务面板 <span id="tpCount"></span><button id="tpClose" style="float:right">✕</button></div>
   <div id="tpList"></div>
 </aside>
+<aside id="dashPanel" style="display:none">
+  <div class="dash-head">📊 省钱仪表盘 <span class="dash-sub">费用 · 命中 · 节省</span><button id="dashClose">✕</button></div>
+  <div class="dash-body">
+    <div class="dash-kpi">
+      <div class="kpi"><span class="kpi-l">今日费用</span><span class="kpi-v" id="kpiToday">—</span></div>
+      <div class="kpi"><span class="kpi-l">缓存命中</span><span class="kpi-v" id="kpiRate">—</span></div>
+      <div class="kpi"><span class="kpi-l">已节省</span><span class="kpi-v" style="color:var(--accent)" id="kpiSaved">—</span></div>
+      <div class="kpi"><span class="kpi-l">轮次/步数</span><span class="kpi-v" id="kpiTurns">—</span></div>
+    </div>
+    <div class="dash-gauge-wrap">
+      <svg id="rateGauge" viewBox="0 0 120 120" width="120" height="120"></svg>
+      <div class="gauge-label"><b id="rateGaugeNum">—</b><span>命中率</span></div>
+    </div>
+    <div class="dash-sec">近 14 天费用趋势（¥）</div>
+    <div id="trendChart" class="dash-chart"></div>
+    <div class="dash-sec">模型费用 Top</div>
+    <div id="dashModels" class="dash-bars"></div>
+    <div class="dash-sec">工具耗时 Top</div>
+    <div id="dashTools" class="dash-bars"></div>
+    <div class="dash-sec">最近缓存命中</div>
+    <div id="dashRecent" class="dash-recent"></div>
+  </div>
+</aside>
 <main><div id="chat"></div></main>
 <footer>
   <div id="workStatus" style="display:none"></div>
@@ -304,7 +360,7 @@ footer{flex:none;border-top:1px solid var(--border);background:var(--bg2);paddin
       <button class="cfg-nav-btn" data-panel="schedule">⏰ 调度队列</button>
       <button class="cfg-nav-btn" data-panel="workspace">📁 工作空间</button>
       <button class="cfg-nav-btn" data-panel="memory">🧠 长期记忆</button>
-      <button class="cfg-nav-btn" data-panel="cache">📊 缓存仪表盘</button>
+
       <button class="cfg-nav-btn" data-panel="mcp">🔌 MCP 生态</button>
       <button class="cfg-nav-btn" data-panel="skills">🧩 技能库</button>
     </nav>
@@ -322,6 +378,11 @@ footer{flex:none;border-top:1px solid var(--border);background:var(--bg2);paddin
       <input id="cmUrl" placeholder="API 地址 https://…/v1" style="flex:1.4;min-width:170px;background:var(--bg3);color:var(--text);border:1px solid var(--border);border-radius:8px;padding:6px 10px;font-size:13px">
       <input id="cmKey" type="password" placeholder="API Key（可选）" style="flex:1;min-width:120px;background:var(--bg3);color:var(--text);border:1px solid var(--border);border-radius:8px;padding:6px 10px;font-size:13px">
     </div>
+    <div class="cfg-row" style="display:flex;gap:6px;flex-wrap:wrap">
+      <input id="cmCtx" type="number" placeholder="上下文窗口 tokens（本地模型建议填，如 131072）" style="flex:1.6;min-width:180px;background:var(--bg3);color:var(--text);border:1px solid var(--border);border-radius:8px;padding:6px 10px;font-size:13px">
+      <input id="cmMaxOut" type="number" placeholder="最大输出 tokens（可选，如 8192）" style="flex:1;min-width:120px;background:var(--bg3);color:var(--text);border:1px solid var(--border);border-radius:8px;padding:6px 10px;font-size:13px">
+    </div>
+    <div class="cfg-hint">本地部署模型（127.0.0.1/localhost）：填「上下文窗口」后预算自动收紧到窗口 75% 以内并留输出余量，长任务不再撑爆窗口。</div>
     <div class="row" style="justify-content:flex-start"><button id="cmAdd">＋ 添加自定义模型</button></div>
     <div class="cfg-row" style="display:flex;gap:6px"><input id="baseUrlOverride" placeholder="API 地址覆盖（可选，覆盖当前服务商地址）" style="flex:1;background:var(--bg3);color:var(--text);border:1px solid var(--border);border-radius:8px;padding:6px 10px;font-size:13px"><button id="baseUrlSave">保存</button></div>
       </section>
@@ -362,6 +423,12 @@ footer{flex:none;border-top:1px solid var(--border);background:var(--bg2);paddin
     </div>
     <div class="cfg-row"><label><input type="checkbox" id="routeChk"> 自动模型路由（规划→pro / 执行→flash）</label></div>
     <div class="cfg-row"><label>上下文预算 tokens <input type="number" id="budgetInput" min="1000" step="1000"></label></div>
+    <div class="cfg-row" style="display:flex;gap:6px;flex-wrap:wrap;align-items:flex-end">
+      <label style="font-size:12px">首 token 等待（秒）<input type="number" id="toFirstToken" min="0" step="10" placeholder="自动"></label>
+      <label style="font-size:12px">流式空闲（秒）<input type="number" id="toStreamIdle" min="0" step="10" placeholder="自动"></label>
+      <label style="font-size:12px">总量（秒）<input type="number" id="toTotal" min="0" step="10" placeholder="自动"></label>
+    </div>
+    <div class="cfg-hint">留空 = 自适应：本地模型自动给更长首 token 等待（慢 prefill 不被误杀）。仅在默认值仍不够/过松时手填。</div>
     <div class="cfg-row"><label><input type="checkbox" id="autoStartChk"> 开机自启（登录后自动启动服务器）</label></div>
     <div class="cfg-row"><label><input type="checkbox" id="notifyChk"> 后台任务完成/失败时桌面通知</label></div>
       </section>
@@ -402,24 +469,7 @@ footer{flex:none;border-top:1px solid var(--border);background:var(--bg2);paddin
     <textarea id="memArea" rows="5" placeholder="例如：所有文档用中文；提交信息用约定式提交…" style="width:100%;background:var(--bg3);color:var(--text);border:1px solid var(--border);border-radius:8px;padding:6px 10px;font-size:12.5px;resize:vertical"></textarea>
     <div class="row" style="justify-content:flex-start"><button id="memSave">保存记忆</button><button id="memDedupe">去重合并</button><span id="memMsg" style="font-size:12px;color:var(--faint)"></span></div>
       </section>
-    <section class="cfg-panel" data-panel="cache">
-    <hr style="border:none;border-top:1px solid var(--border);margin:14px 0 8px">
-    <h4 style="margin:4px 0;color:var(--accent2)">缓存命中率仪表盘（最近 10 次）</h4>
-    <div id="cacheSummary" style="font-size:12.5px;color:var(--dim);padding:2px"></div>
-    <div id="cacheRecent" style="max-height:130px;overflow-y:auto"></div>
-    <h4 style="margin:10px 0 4px;color:var(--accent2)">费用二级分账（模型 / 工具 Top5 · 近 14 天）</h4>
-    <div style="display:flex;gap:10px;flex-wrap:wrap">
-      <div style="flex:1;min-width:170px">
-        <div style="font-size:11.5px;color:var(--faint)">模型 Top5（费用）</div>
-        <div id="costTopModels" style="font-size:12px;color:var(--dim)"></div>
-      </div>
-      <div style="flex:1;min-width:170px">
-        <div style="font-size:11.5px;color:var(--faint)">工具 Top5（耗时）</div>
-        <div id="costTopTools" style="font-size:12px;color:var(--dim)"></div>
-      </div>
-    </div>
-    <div id="costLine" style="margin-top:8px"></div>
-      </section>
+
     <section class="cfg-panel" data-panel="mcp">
     <hr style="border:none;border-top:1px solid var(--border);margin:14px 0 8px">
     <h4 style="margin:4px 0;color:var(--accent2)">MCP 生态预设（一键接入，重启生效）</h4>

package/src/web/routes/domains/config.js

@@ -50,6 +50,7 @@ export async function handle({ req, res, method, p, url }, deps, shared) {
       routing: cfg.routing?.enabled ? cfg.routing : null,
       reasoning,
       contextBudget: cfg.contextBudget || 128000,
+      timeout: cfg.timeout || null, // v0.3.2：分层超时（firstTokenMs/streamIdleMs/totalMs）
       pricingAsOf: PRICE_DATA_AS_OF,
       autostart: autostartStatus(),
       notify: cfg.notify !== false,
@@ -112,6 +113,17 @@ export async function handle({ req, res, method, p, url }, deps, shared) {
       next.contextBudget = n;
       cfg.contextBudget = n;
     }
+    // v0.3.2 超时分层：firstTokenMs（首 token 等待）/ streamIdleMs（流式空闲）/ totalMs（总量）
+    if (body.timeout !== undefined && body.timeout && typeof body.timeout === 'object') {
+      cfg.timeout = cfg.timeout || {};
+      for (const k of ['firstTokenMs', 'streamIdleMs', 'totalMs']) {
+        const v = Number(body.timeout[k]);
+        if (Number.isFinite(v) && v > 0) cfg.timeout[k] = Math.round(v);
+        else if (body.timeout[k] === null || body.timeout[k] === '') delete cfg.timeout[k];
+      }
+      // 超时在 createProvider 时按 cfg 固化进实例，改后清缓存让新值即时生效（否则要重启）
+      providerCache.clear();
+    }
     if (body.reasoningEffort !== undefined) {
       const re = String(body.reasoningEffort);
       if (!['off', 'low', 'high', 'max'].includes(re)) {
@@ -165,6 +177,8 @@ export async function handle({ req, res, method, p, url }, deps, shared) {
         baseUrl: cm.baseUrl || '',
         envKey: cm.envKey || null,
         vision: Boolean(cm.vision),
+        contextWindow: Number(cm.contextWindow) > 0 ? Number(cm.contextWindow) : null,
+        maxOutputTokens: Number(cm.maxOutputTokens) > 0 ? Number(cm.maxOutputTokens) : null,
         keyState: stored ? 'stored' : 'none',
         keyMasked: stored ? maskKey(stored) : null,
       };
@@ -232,14 +246,23 @@ export async function handle({ req, res, method, p, url }, deps, shared) {
         return json(res, 400, { error: `自定义模型 ${name} 已存在（可修改）` });
       }
       cfg.customModels = cfg.customModels || {};
+      // 修改时保留未提交字段（tokenizer/contextWindow/maxOutputTokens/vision）——
+      // 前端「修改」只发 baseUrl+label，整体替换会丢这些声明，导致本地模型窗口回退兜底
+      const prev = action === 'updateCustom' ? (cfg.customModels[name] || {}) : {};
       cfg.customModels[name] = {
         label,
         baseUrl,
-        envKey: String(body.envKey || '').trim() || undefined,
-        vision: body.vision === true || body.vision === 'on' ? true : undefined,
+        envKey: String(body.envKey || '').trim() || prev.envKey || undefined,
+        vision: body.vision === true || body.vision === 'on' ? true : prev.vision === true ? true : undefined,
+        ...(prev.tokenizer ? { tokenizer: prev.tokenizer } : {}),
+        // v0.3.2 本地模型自适应：显式声明上下文窗口/最大输出，预算据此推导（不撑爆窗口）
+        ...(Number(body.contextWindow) > 0 ? { contextWindow: Math.round(Number(body.contextWindow)) } : (prev.contextWindow ? { contextWindow: prev.contextWindow } : {})),
+        ...(Number(body.maxOutputTokens) > 0 ? { maxOutputTokens: Math.round(Number(body.maxOutputTokens)) } : (prev.maxOutputTokens ? { maxOutputTokens: prev.maxOutputTokens } : {})),
       };
       if (String(body.key || '').trim()) setStoredKey(`custom:${name}`, String(body.key).trim());
       saveConfig(cfg);
+      // baseUrl 改动影响 isLocal 判定 → 超时档位变化；清缓存让新超时/新端点即时生效
+      providerCache.clear();
       return json(res, 200, { ok: true, name });
     }
     if (action === 'removeCustom') {
@@ -252,6 +275,7 @@ export async function handle({ req, res, method, p, url }, deps, shared) {
         cfg.model = state.modelName;
       }
       saveConfig(cfg);
+      providerCache.clear();
       return json(res, 200, { ok: true, name, model: state.modelName });
     }
     // 质检（自定义模型连通性）：发起一次最小对话验证 baseUrl+Key 可用

package/src/web/server.js

@@ -29,8 +29,8 @@ import { MAX_CONCURRENT } from './constants.js';
 import { createAgent } from '../agent.js';
 import { createPermission } from '../permissions.js';
 import { buildSystemPrompt } from '../prompts.js';
-import { loadProjectMemory, extractAndAppendProjectMemory } from '../memory.js';
-import { saveTaskState, clearTaskState, loadTaskState, resumePrompt } from '../task-state.js';
+import { loadProjectMemory, loadProjectMemoryEntries, retrieveRelevant, extractAndAppendProjectMemory } from '../memory.js';
+import { saveTaskStateMerge, clearTaskState, loadTaskState, resumePrompt } from '../task-state.js';
 import { createWebIO } from './web-io.js';
 import { startMcpServers } from '../mcp.js';
 import {
@@ -397,7 +397,11 @@ export async function runWebServer({ host = '127.0.0.1', port = 3820, authToken
     // 每轮结束提取的新条目只写文件、不回灌当前会话。
     let projectMemorySnapshot = sessionMemoryCache.get(sessionName);
     if (projectMemorySnapshot === undefined) {
-      projectMemorySnapshot = loadProjectMemory(taskDir);
+      // v0.3.1 语义检索：用本轮任务（新会话即首条消息）取「相关记忆条目」，而非全量截 4K——
+      // 换任务只注入相关记忆、省 token；快照保证同一会话前缀稳定。
+      const allEntries = loadProjectMemoryEntries(taskDir);
+      const query = String(built.persistText || '');
+      projectMemorySnapshot = query ? retrieveRelevant(allEntries, query, 6).join('\n') : loadProjectMemory(taskDir);
       sessionMemoryCache.set(sessionName, projectMemorySnapshot);
       if (sessionMemoryCache.size > 200) {
         const oldest = sessionMemoryCache.keys().next().value;
@@ -527,7 +531,7 @@ export async function runWebServer({ host = '127.0.0.1', port = 3820, authToken
       // v0.3.0 P0-2：任务检查点——跑满步数(capHit)或中断(aborted)时落盘供续跑，正常完成清除
       const finalSessionName = path.basename(session.file);
       if (r.capHit || r.aborted) {
-        saveTaskState(finalSessionName, {
+        saveTaskStateMerge(finalSessionName, {
           goal: built.persistText,
           progress: r.text || '',
           artifacts: io.stats().deliverables,
@@ -547,7 +551,9 @@ export async function runWebServer({ host = '127.0.0.1', port = 3820, authToken
         const { messageTokens } = await import('../context.js');
         const counter = makeTokenCounter(runModel);
         const used = messages.reduce((/** @type {any} */ sum, /** @type {any} */ m) => sum + messageTokens(m, counter), 0);
-        budgetInfo = { used, total: cfg.contextBudget || modelPreset(runModel)?.budgetTokens || 128000 };
+        // v0.3.2：展示用 total 与 Agent 实际预算同源（safeBudget 按窗口推导），避免「显示 128k 实际 98k」的误导
+        const { resolveModelCaps, safeBudget } = await import('../model-caps.js');
+        budgetInfo = { used, total: safeBudget(cfg, resolveModelCaps(cfg, runModel)) };
       } catch {}
       send({
         type: 'done',