lumencode 0.4.4 → 1.1.0

package/README.md

@@ -11,22 +11,29 @@
 </p>
 
 <p align="center">
-  支持 <b>Claude Code · Codex · OpenCode</b> 三大 AI 编码工具 · 600+ 模型定价 · AI 贡献度归因 · 一键飞书/钉钉周报
+  支持 <b>Claude Code · Codex · OpenCode</b> 三大 AI 编码工具 · 600+ 模型定价 · AI 贡献度归因 · 按项目独立汇报 · 自定义时间范围 · 一键飞书/钉钉周报
 </p>
 
 <p align="center">
   <a href="README_EN.md">English</a> · <a href="#命令用法">命令</a> · <a href="#常见问题">FAQ</a> · <a href="#更新日志">更新日志</a>
 </p>
+<div align="center">
+  <img src="doc/数据分析页面.png" alt="LumenCode Dashboard" width="800">
+</div>
+
 ---
 
+
 ## 它解决什么问题？
 
-> 「这周 AI 帮你写了多少代码？」「订阅这些工具值不值？」—— 与其每周手算，不如一条命令搞定。
+> 「AI 帮你写了多少代码？」「订阅这些工具值不值？」—— 与其手算，不如一条命令搞定。
 
 | 场景 | 用 lumencode 解决 |
 |------|----------------------|
 | **写周报** | 选周报 → 点「工作汇报 → 复制」→ 粘贴飞书/钉钉。**3 秒搞定。** |
-| **证明 AI ROI** | 「本周 67% 提交有 AI 参与，AI 辅助新增 4,200 行，费用 $12.5」**有数据，有底气。** |
+| **证明 AI ROI** | 「67% 提交有 AI 参与，AI 辅助新增 4,200 行，费用 $12.5」**有数据，有底气。** |
+| **按项目汇报** | 配置多项目后，选择单个项目生成独立工作汇报，方便向不同项目负责人对齐 |
+| **对齐 Sprint 周期** | 除日/周/月外，支持自定义起止日期，不再被固定周期限制 |
 | **理解使用习惯** | 哪个项目用得最多？哪个模型最费 Token？什么时段是编码高峰？**一目了然。** |
 | **追踪 AI 成本** | 内置 **600+ 模型定价**（含 GLM、Kimi、Qwen、DeepSeek 等），自动算出等效 API 花销 |
 
@@ -49,64 +56,62 @@ npx lumencode serve
 
 ## 产品亮点
 
+<div align="center">
+  <img src="doc/核心能力.png" alt="LumenCode 核心能力" width="720">
+</div>
+
+
 | 亮点 | 说明 |
 |------|------|
 | 🌐 **三工具统一** | Claude Code / Codex / OpenCode 数据全自动汇总，左侧标签一键切换 |
 | 🤖 **AI 贡献度量化** | 识别 `Co-Authored-By: Claude` 等签名，多层归因引擎量化 AI 在你代码中的实际占比 |
 | 📝 **自然语言工作汇报** | 详报/简报一键生成，支持标准 Markdown / 飞书 / 钉钉三种格式，每个板块附诊断解读 |
+| 📂 **按项目独立汇报** | 右侧面板选择项目，生成该项目的独立工作汇报（commits + AI 交互量 + 热点文件） |
+| 📅 **自定义时间范围** | 除日/周/月外，支持选择任意起止日期，方便对齐 Sprint 周期 |
 | 💰 **精确费用估算** | 600+ 模型本地定价（含 GLM/Kimi/Qwen/DeepSeek）+ Portkey API 兜底，未知模型不计费而非乱算 |
 | 📦 **零配置开箱即用** | 首次运行自动检测工具目录、推导项目路径 |
 | 🔍 **数据钻取** | 点击任意图表下钻明细，从汇总数据到具体会话/提交一气呵成 |
 | 📈 **趋势与洞察** | 周报/月报附峰值日识别、连续活跃分析、工具使用五类分布（编辑/阅读/执行/规划/研究） |
-| 🌙 **暗色模式** | 亮/暗主题一键切换，全图表自适配 |
+| 🌙 **亮/暗主题** | 亮色/暗色主题一键切换，全图表自适配 |
 
 ---
 
 ## 产品截图
 
-### 多工具汇总视图
+### 数据分析总览
 
-> 全工具数据汇总，AI 贡献占比、Token 消耗、费用、活跃时段、模型分布、场景拆分一屏掌握。
-
-![全量汇总](doc/全量Ai工具汇总报告.png)
-
-### 单工具报告
-
-> 切换左侧标签即可单独查看任一工具的数据。
+> 左侧数据源面板一键切换工具，主区域汇总 Token 消耗、费用、模型分布、AI 贡献度归因。
 
 <table>
   <tr>
-    <td><img src="doc/claude_code工具使用报告.png" alt="Claude Code" width="400"></td>
-    <td><img src="doc/codex使用报告.png" alt="Codex" width="400"></td>
-  </tr>
-  <tr>
-    <td align="center"><b>Claude Code</b></td>
-    <td align="center"><b>OpenAI Codex</b></td>
-  </tr>
-  <tr>
-    <td><img src="doc/opencode使用报告.png" alt="OpenCode" width="400"></td>
-    <td></td>
+    <td><img src="doc/数据分析页面.png" alt="汇总面板与趋势图" width="400"></td>
+    <td><img src="doc/数据分析页面2.png" alt="项目分布与时段分布" width="400"></td>
   </tr>
   <tr>
-    <td align="center"><b>OpenCode</b></td>
-    <td></td>
+    <td align="center">汇总指标 + Token 趋势</td>
+    <td align="center">项目分布 + 时段分布 + 会话列表</td>
   </tr>
 </table>
 
-### 场景分析 & 模型分布
+![AI 贡献度与提交分析](doc/数据分析页面3.png)
 
-> 按工作类型分类（编码 / 测试 / 调试 / 文档 / 审查 / 规划），还能下钻看每个模型的具体 Token 用量。
+### 多工具维度
 
-<table>
-  <tr>
-    <td><img src="doc/工作类型分布_匹配示例.png" alt="场景分析" width="400"></td>
-    <td><img src="doc/模型使用分布_具体用量.png" alt="模型分布" width="400"></td>
-  </tr>
-  <tr>
-    <td align="center">工作类型分布（含匹配关键词示例）</td>
-    <td align="center">模型使用分布（含具体 Token 用量）</td>
-  </tr>
-</table>
+> 切换到「全部工具」视图，查看跨工具的汇总数据与对比分析。
+
+![多工具维度](doc/多工具维度.png)
+
+### 项目分布 & 会话记录
+
+> 按项目统计 Token、费用、会话数，点击下钻查看单条会话明细。
+
+![项目分布与会话记录](doc/项目分布-会话记录.png)
+
+### 场景分析
+
+> 按工作类型分类（编码 / 测试 / 调试 / 文档 / 审查 / 规划），附匹配关键词示例。
+
+![场景分析](doc/工作类型分布_匹配示例.png)
 
 ### 工作汇报 · 一键生成可直接发布的周报
 
@@ -115,6 +120,7 @@ npx lumencode serve
 - **详报** —— 完整数据 + 洞察解读 + 板块编号，适合周报、月报
 - **简报** —— 3-5 句话核心摘要，适合日报或群消息
 - **多平台格式** —— 标准 Markdown / 飞书 / 钉钉，一键切换
+- **按项目生成** —— 右侧面板选择项目，生成该项目的独立汇报
 
 <table>
   <tr>
@@ -127,11 +133,13 @@ npx lumencode serve
   </tr>
 </table>
 
-### 暗色模式
+### 亮色 / 暗色主题
 
 > 全图表配色自适配，长时间阅读不伤眼。
 
-![暗色模式](doc/暗色模式.png)
+![亮色模式](doc/浅色模式.png)
+
+> 暗色模式为默认主题，上方截图均为暗色模式下的效果。
 
 ---
 
@@ -224,6 +232,14 @@ v0.4.0 起支持 Claude Code、Codex、OpenCode 三种工具，**首次运行自
 
 ## 更新日志
 
+### v1.0.0 (2026-05-24) — 项目级汇报 & 自定义时间
+
+- **按项目独立汇报** — 工作汇报右侧面板新增项目选择器，选定后自动筛选该项目数据，生成独立工作汇报（commits + AI 交互量 + 热点文件）
+- **自定义时间范围** — 新增「自定义」周期选项，支持选择任意起止日期，方便对齐 Sprint 周期
+- **智能日期导航** — 左右箭头根据当前周期自动调整步长：日报 ±1天、周报 ±7天、月报 ±1个月
+- **侧边栏重构** — 版本信息、主题切换、收起按钮统一移至底部，顶部仅保留标题与链接，布局更紧凑
+- **界面重构** — 全新视觉设计，布局更紧凑美观，数据呈现更直观
+
 ### v0.4.0 (2026-05-22) — 多工具统一平台
 
 从 Claude Code 单工具报告升级为 AI 编码全栈分析平台。

package/index.js

@@ -72,12 +72,17 @@ function loadCliConfig() {
   return { config, dateArg, effectiveIncludeProjects, configPath };
 }
 
-async function buildReportData(period, dateArg, config, effectiveIncludeProjects, tool = 'all') {
-  // 使用新的多工具解析入口
-  const { records, toolBreakdown } = await parseAllEnabledTools(config, {
-    excludeProjects: config.excludeProjects,
-    includeProjects: effectiveIncludeProjects,
-  });
+async function buildReportData(period, dateArg, config, effectiveIncludeProjects, tool = 'all', preParsed = null, options = {}) {
+  // 使用预解析结果或全量解析
+  let records, toolBreakdown;
+  if (preParsed) {
+    ({ records, toolBreakdown } = preParsed);
+  } else {
+    ({ records, toolBreakdown } = await parseAllEnabledTools(config, {
+      excludeProjects: config.excludeProjects,
+      includeProjects: effectiveIncludeProjects,
+    }));
+  }
 
   if (records.length === 0) {
     return null;
@@ -92,54 +97,60 @@ async function buildReportData(period, dateArg, config, effectiveIncludeProjects
     return null;
   }
 
-  // 其余逻辑保持不变
-  const { filtered, start, end } = filterRecordsByPeriod(toolRecords, period, dateArg);
-  const usageStats = computeUsageStats(filtered, config.scenarioKeywords, config.costMode);
-  const sessions = groupBySessions(filtered);
+  const { filtered, start, end } = filterRecordsByPeriod(toolRecords, period, dateArg, { customStart: options.customStart, customEnd: options.customEnd });
+  const reposConfigured = !!(config.repos && config.repos.length > 0);
 
-  let gitStats = null;
-  if (config.repos && config.repos.length > 0) {
-    // 按工具过滤后，只统计该工具覆盖的项目对应的 repos
+  // ── 第一层并发：三个独立的同步计算 ──
+  const [usageStats, sessions, billingBlocks] = [
+    computeUsageStats(filtered, config.scenarioKeywords, config.costMode),
+    groupBySessions(filtered),
+    identifyBillingBlocks(filtered, config.blockQuota ? 5 : 5, config.costMode),
+  ];
+
+  // ── 第二层并发：gitStats(async) + trendData + prevStats ──
+  const gitStatsPromise = (async () => {
+    if (!reposConfigured) return null;
     const coveredBases = new Set(filtered.map(r => {
       const p = r.project || '';
       return p.replace(/\\/g, '/').replace(/\/$/, '').split('/').pop();
     }).filter(Boolean));
-    const toolRepos = config.repos.filter(r => coveredBases.has(r.replace(/\\/g, '/').replace(/\/$/, '').split('/').pop()));
-    if (toolRepos.length > 0) {
-      // 扩展 git 查询窗口 +2 天，以匹配 session 跨天的延迟提交
-      const extendedEnd = new Date(end);
-      extendedEnd.setDate(extendedEnd.getDate() + 2);
-      const extendedEndStr = extendedEnd.toISOString().slice(0, 10) + 'T23:59:59';
-      gitStats = await getGitStatsForMultipleReposAsync(toolRepos, start, extendedEndStr);
-      gitStats = finalizeGitStats(gitStats, sessions);
-      // 归因已完成，过滤 commitList 到原始窗口，重算基础统计
-      if (gitStats.commitList) {
-        const windowStart = start;
-        const windowEnd = end + 'T23:59:59';
-        const inWindow = gitStats.commitList.filter(c => (c.date || '') >= windowStart && (c.date || '') <= windowEnd);
-        gitStats.commits = inWindow.length;
-        gitStats.linesAdded = inWindow.reduce((s, c) => s + (c.linesAdded || 0), 0);
-        gitStats.linesDeleted = inWindow.reduce((s, c) => s + (c.linesDeleted || 0), 0);
-        gitStats.filesChanged = new Set(inWindow.flatMap(c => (c.files || []).map(f => f.path))).size;
-        // commitList 保留全部（含跨天），前端 drill-down 需要
-        // 但提交类型和热点只基于窗口内
-        gitStats.commitTypes = computeCommitTypes(inWindow);
-        gitStats.fileHotspots = computeFileHotspots(inWindow, 10);
-      }
+    let toolRepos = config.repos.filter(r => coveredBases.has(r.replace(/\\/g, '/').replace(/\/$/, '').split('/').pop()));
+    if (toolRepos.length === 0) toolRepos = config.repos;
+    if (toolRepos.length === 0) return null;
+    const extendedEnd = new Date(end);
+    extendedEnd.setDate(extendedEnd.getDate() + 2);
+    const extendedEndStr = extendedEnd.toISOString().slice(0, 10) + 'T23:59:59';
+    let gs = await getGitStatsForMultipleReposAsync(toolRepos, start, extendedEndStr);
+    gs = finalizeGitStats(gs, sessions);
+    if (gs.commitList) {
+      const windowStart = start;
+      const windowEnd = end + 'T23:59:59';
+      const inWindow = gs.commitList.filter(c => (c.date || '') >= windowStart && (c.date || '') <= windowEnd);
+      gs.commits = inWindow.length;
+      gs.linesAdded = inWindow.reduce((s, c) => s + (c.linesAdded || 0), 0);
+      gs.linesDeleted = inWindow.reduce((s, c) => s + (c.linesDeleted || 0), 0);
+      gs.filesChanged = new Set(inWindow.flatMap(c => (c.files || []).map(f => f.path))).size;
+      gs.commitTypes = computeCommitTypes(inWindow);
+      gs.fileHotspots = computeFileHotspots(inWindow, 10);
     }
-  }
+    return gs;
+  })();
 
-  const trendData = computeTrendData(toolRecords, period, dateArg);
+  const trendDataPromise = Promise.resolve(computeTrendData(toolRecords, period, dateArg));
 
-  // Previous period stats
-  const prevRange = computePrevPeriodRange(period, dateArg);
-  const prevFiltered = toolRecords.filter(r => {
-    if (!r.timestamp) return false;
-    const date = r.timestamp.slice(0, 10);
-    return date >= prevRange.start && date <= prevRange.end;
-  });
-  const prevStats = prevFiltered.length > 0 ? computeUsageStats(prevFiltered, config.scenarioKeywords, config.costMode) : null;
+  const prevStatsPromise = (async () => {
+    const prevRange = computePrevPeriodRange(period, dateArg, { customStart: options.customStart, customEnd: options.customEnd });
+    const prevFiltered = toolRecords.filter(r => {
+      if (!r.timestamp) return false;
+      const date = r.timestamp.slice(0, 10);
+      return date >= prevRange.start && date <= prevRange.end;
+    });
+    return prevFiltered.length > 0 ? computeUsageStats(prevFiltered, config.scenarioKeywords, config.costMode) : null;
+  })();
+
+  const [gitStats, trendData, prevStats] = await Promise.all([gitStatsPromise, trendDataPromise, prevStatsPromise]);
 
+  // ── 第三层：依赖 usageStats 的同步派生 ──
   const slimSessions = sessions.map(s => ({
     id: s.id,
     project: s.project,
@@ -149,10 +160,79 @@ async function buildReportData(period, dateArg, config, effectiveIncludeProjects
     commits: s.commits || [],
   }));
 
-  const billingBlocks = identifyBillingBlocks(filtered, config.blockQuota ? 5 : 5, config.costMode);
+  const statsTB = usageStats.toolBreakdown || {};
+  const mergedBreakdown = {};
+  for (const [name, base] of Object.entries(toolBreakdown)) {
+    const s = statsTB[name] || {};
+    mergedBreakdown[name] = {
+      inputTokens: s.inputTokens || 0,
+      outputTokens: s.outputTokens || 0,
+      cacheRead: s.cacheRead || 0,
+      cacheCreate: s.cacheCreate || 0,
+      count: s.count || 0,
+      sessionCount: base.sessionCount || 0,
+    };
+  }
+  for (const [name, data] of Object.entries(statsTB)) {
+    if (!mergedBreakdown[name]) {
+      mergedBreakdown[name] = {
+        inputTokens: data.inputTokens || 0,
+        outputTokens: data.outputTokens || 0,
+        cacheRead: data.cacheRead || 0,
+        cacheCreate: data.cacheCreate || 0,
+        count: data.count || 0,
+        sessionCount: 0,
+      };
+    }
+  }
 
-  const reposConfigured = !!(config.repos && config.repos.length > 0);
-  return { usageStats, gitStats, reposConfigured, sessions: slimSessions, start, end, trendData, prevStats, billingBlocks, toolBreakdown };
+  // ── 第四层：projectDetails（从 commitList 按 repo 分组派生，无需再次 git 调用）──
+  const projectDetails = {};
+  const projEntries = Object.entries(usageStats.projects || {}).sort((a, b) => b[1].requests - a[1].requests);
+  if (reposConfigured && gitStats?.commitList?.length) {
+    const windowEnd = end + 'T23:59:59';
+    const inWindow = gitStats.commitList.filter(c => (c.date || '') >= start && (c.date || '') <= windowEnd);
+    const repoGroups = new Map();
+    for (const c of inWindow) {
+      const base = (c.repo || '').replace(/\\/g, '/').replace(/\/$/, '').split('/').pop();
+      if (!base) continue;
+      if (!repoGroups.has(base)) repoGroups.set(base, []);
+      repoGroups.get(base).push(c);
+    }
+    for (const [projName, projStats] of projEntries) {
+      const repoCommits = repoGroups.get(projName) || [];
+      if (repoCommits.length === 0) {
+        projectDetails[projName] = { usage: projStats, git: null, topCommits: [] };
+        continue;
+      }
+      const uniqueFiles = new Set();
+      let linesAdded = 0, linesDeleted = 0;
+      for (const c of repoCommits) {
+        linesAdded += c.linesAdded || 0;
+        linesDeleted += c.linesDeleted || 0;
+        for (const f of c.files || []) uniqueFiles.add(f.path);
+      }
+      const topCommits = repoCommits
+        .filter(c => c.type === 'feat' || c.type === 'fix')
+        .slice(0, 5)
+        .map(c => ({ type: c.type, subject: c.subject, scope: c.scope }));
+      projectDetails[projName] = {
+        usage: projStats,
+        git: {
+          commits: repoCommits.length, linesAdded, linesDeleted,
+          filesChanged: uniqueFiles.size,
+          fileHotspots: computeFileHotspots(repoCommits, 5),
+        },
+        topCommits,
+      };
+    }
+  } else {
+    for (const [projName, projStats] of projEntries) {
+      projectDetails[projName] = { usage: projStats, git: null, topCommits: [] };
+    }
+  }
+
+  return { usageStats, gitStats, reposConfigured, sessions: slimSessions, start, end, trendData, prevStats, billingBlocks, toolBreakdown: mergedBreakdown, projectDetails };
 }
 
 if (!command || command === 'help' || command === '--help') {

package/lib/aggregate.js

@@ -250,11 +250,12 @@ export function computeUsageStats(records, scenarioKeywords, costMode = 'auto')
 
       // Model
       if (model) {
-        if (!stats.models[model]) stats.models[model] = { count: 0, outputTokens: 0, inputTokens: 0, cacheRead: 0, cost: 0, costMode: 'unknown' };
+        if (!stats.models[model]) stats.models[model] = { count: 0, outputTokens: 0, inputTokens: 0, cacheRead: 0, cacheCreate: 0, cost: 0, costMode: 'unknown' };
         stats.models[model].count++;
         stats.models[model].outputTokens += outputTokens;
         stats.models[model].inputTokens += inputTokens;
         stats.models[model].cacheRead += cacheRead;
+        stats.models[model].cacheCreate += cacheCreate;
       }
     }
 
@@ -275,10 +276,23 @@ export function computeUsageStats(records, scenarioKeywords, costMode = 'auto')
     const projectName = getProjectBaseName(r.project);
     if (projectName) {
       if (!stats.projects[projectName]) {
-        stats.projects[projectName] = { sessions: new Set(), requests: 0 };
+        stats.projects[projectName] = { sessions: new Set(), requests: 0, inputTokens: 0, outputTokens: 0, cacheRead: 0, cacheCreate: 0, estimatedCost: 0, models: {} };
       }
       if (isAssistant) {
         stats.projects[projectName].requests++;
+        stats.projects[projectName].inputTokens += inputTokens;
+        stats.projects[projectName].outputTokens += outputTokens;
+        stats.projects[projectName].cacheRead += cacheRead;
+        stats.projects[projectName].cacheCreate += cacheCreate;
+        const pricing = model ? resolveModelPricing(model) : { unknown: true };
+        const recCost = (r.costUSD != null && r.costUSD > 0) ? r.costUSD : calculateRecordCost(r, pricing);
+        stats.projects[projectName].estimatedCost += recCost;
+        if (model) {
+          if (!stats.projects[projectName].models[model]) stats.projects[projectName].models[model] = { count: 0, inputTokens: 0, outputTokens: 0 };
+          stats.projects[projectName].models[model].count++;
+          stats.projects[projectName].models[model].inputTokens += inputTokens;
+          stats.projects[projectName].models[model].outputTokens += outputTokens;
+        }
       }
       if (r.sessionId) {
         stats.projects[projectName].sessions.add(r.sessionId);
@@ -373,16 +387,23 @@ export function computeUsageStats(records, scenarioKeywords, costMode = 'auto')
 
   // Convert project session Sets to sizes
   for (const proj of Object.keys(stats.projects)) {
+    const p = stats.projects[proj];
     stats.projects[proj] = {
-      sessions: stats.projects[proj].sessions.size,
-      requests: stats.projects[proj].requests,
+      sessions: p.sessions.size,
+      requests: p.requests,
+      inputTokens: p.inputTokens || 0,
+      outputTokens: p.outputTokens || 0,
+      cacheRead: p.cacheRead || 0,
+      cacheCreate: p.cacheCreate || 0,
+      estimatedCost: p.estimatedCost || 0,
+      models: p.models || {},
     };
   }
 
   return stats;
 }
 
-export function filterRecordsByPeriod(records, period, refDate) {
+export function filterRecordsByPeriod(records, period, refDate, options = {}) {
   const d = new Date(refDate);
   let start, end;
 
@@ -406,6 +427,10 @@ export function filterRecordsByPeriod(records, period, refDate) {
       start = `${d.getFullYear()}-${String(d.getMonth() + 1).padStart(2, '0')}-01`;
       end = `${d.getFullYear()}-${String(d.getMonth() + 1).padStart(2, '0')}-${String(new Date(d.getFullYear(), d.getMonth() + 1, 0).getDate()).padStart(2, '0')}`;
       break;
+    case 'custom':
+      start = options.customStart || formatDate(d);
+      end = options.customEnd || start;
+      break;
     default:
       start = formatDate(d);
       end = start;
@@ -554,7 +579,7 @@ export function groupBySessions(records) {
   }).sort((a, b) => b.endTime.localeCompare(a.endTime));
 }
 
-export function computePrevPeriodRange(period, refDate) {
+export function computePrevPeriodRange(period, refDate, options = {}) {
   const d = new Date(refDate);
   switch (period) {
     case 'daily':
@@ -574,6 +599,15 @@ export function computePrevPeriodRange(period, refDate) {
       const y = d.getFullYear(), m = d.getMonth();
       return { start: `${y}-${String(m + 1).padStart(2, '0')}-01`, end: `${y}-${String(m + 1).padStart(2, '0')}-${String(new Date(y, m + 1, 0).getDate()).padStart(2, '0')}` };
     }
+    case 'custom': {
+      const cs = options.customStart || formatDate(d);
+      const ce = options.customEnd || cs;
+      const spanMs = new Date(ce) - new Date(cs) + 86400000; // inclusive days
+      const gap = 86400000; // 1 day gap
+      const prevEnd = new Date(new Date(cs).getTime() - gap);
+      const prevStart = new Date(prevEnd.getTime() - spanMs + 86400000);
+      return { start: formatDate(prevStart), end: formatDate(prevEnd) };
+    }
     default:
       d.setDate(d.getDate() - 1);
       return { start: formatDate(d), end: formatDate(d) };

package/lib/cache.js

@@ -2,6 +2,7 @@ import { statSync } from 'fs';
 import { parseJsonlFile } from './parser.js';
 
 const fileCache = new Map();
+const CACHE_MAX_FILES = 200;
 
 export function getCachedFileRecords(filePath) {
   const { mtimeMs } = statSync(filePath);
@@ -10,6 +11,13 @@ export function getCachedFileRecords(filePath) {
 
   const records = parseJsonlFile(filePath);
   fileCache.set(filePath, { mtime: mtimeMs, records });
+
+  // LRU eviction
+  while (fileCache.size > CACHE_MAX_FILES) {
+    const oldest = fileCache.keys().next().value;
+    fileCache.delete(oldest);
+  }
+
   return records;
 }
 

package/lib/git.js

@@ -157,6 +157,13 @@ const AI_CONFIDENCE = {
   HIGH: 'high',
 };
 
+const CONFIDENCE_WEIGHTS = {
+  [AI_CONFIDENCE.HIGH]: 1.0,
+  [AI_CONFIDENCE.MEDIUM]: 0.7,
+  [AI_CONFIDENCE.LOW]: 0.2,
+  [AI_CONFIDENCE.NONE]: 0,
+};
+
 function isCountedAIConfidence(confidence) {
   return confidence === AI_CONFIDENCE.HIGH || confidence === AI_CONFIDENCE.MEDIUM;
 }
@@ -285,6 +292,8 @@ export function detectAICommit(subject = '', author = '', body = '') {
 
 export function computeAIContribution(commits, toolFilter = null) {
   let aiCommits = 0, aiLinesAdded = 0, aiLinesDeleted = 0;
+  let possibleAICommits = 0, possibleAILinesAdded = 0, possibleAILinesDeleted = 0;
+  let weightedAILinesAdded = 0, weightedAILinesDeleted = 0;
   let aiCommitLinesAdded = 0, aiCommitLinesDeleted = 0;
   let aiFileLinesAdded = 0, aiFileLinesDeleted = 0;
   let highConfidenceCommits = 0, mediumConfidenceCommits = 0, lowConfidenceCommits = 0;
@@ -303,27 +312,39 @@ export function computeAIContribution(commits, toolFilter = null) {
     else if (confidence === AI_CONFIDENCE.MEDIUM) mediumConfidenceCommits++;
     else if (confidence === AI_CONFIDENCE.LOW) lowConfidenceCommits++;
 
+    // 计算文件级行数（用于 HIGH/MEDIUM/LOW 各自统计）
+    const matchedFiles = new Set((c.aiEvidenceDetails?.matchedFiles || []).map(normalizeCommitFilePath));
+    const useMatchedFiles = matchedFiles.size > 0;
+    let fileAdded = 0;
+    let fileDeleted = 0;
+    for (const f of c.files || []) {
+      const filePath = normalizeCommitFilePath(f.path);
+      if (useMatchedFiles && !matchedFiles.has(filePath)) continue;
+      fileAdded += f.added || 0;
+      fileDeleted += f.deleted || 0;
+    }
+    if (!useMatchedFiles && (c.attributionType === 'explicit' || c.attributionType?.startsWith('session_'))) {
+      fileAdded = c.linesAdded || 0;
+      fileDeleted = c.linesDeleted || 0;
+    }
+
     if (isCountedAIConfidence(confidence)) {
       aiCommits++;
       aiCommitLinesAdded += c.linesAdded || 0;
       aiCommitLinesDeleted += c.linesDeleted || 0;
-
-      const matchedFiles = new Set((c.aiEvidenceDetails?.matchedFiles || []).map(normalizeCommitFilePath));
-      const useMatchedFiles = matchedFiles.size > 0;
-      let fileAdded = 0;
-      let fileDeleted = 0;
-      for (const f of c.files || []) {
-        const filePath = normalizeCommitFilePath(f.path);
-        if (useMatchedFiles && !matchedFiles.has(filePath)) continue;
-        fileAdded += f.added || 0;
-        fileDeleted += f.deleted || 0;
-      }
-      if (!useMatchedFiles && (c.attributionType === 'explicit' || c.attributionType?.startsWith('session_'))) {
-        fileAdded = c.linesAdded || 0;
-        fileDeleted = c.linesDeleted || 0;
-      }
       aiFileLinesAdded += fileAdded;
       aiFileLinesDeleted += fileDeleted;
+    } else if (confidence === AI_CONFIDENCE.LOW) {
+      possibleAICommits++;
+      possibleAILinesAdded += fileAdded;
+      possibleAILinesDeleted += fileDeleted;
+    }
+
+    // 加权计算：所有归因的 commit 都参与（包括 LOW）
+    const weight = CONFIDENCE_WEIGHTS[confidence] || 0;
+    if (weight > 0) {
+      weightedAILinesAdded += fileAdded * weight;
+      weightedAILinesDeleted += fileDeleted * weight;
     }
   }
   aiLinesAdded = aiFileLinesAdded;
@@ -331,20 +352,32 @@ export function computeAIContribution(commits, toolFilter = null) {
   const total = allCommits.length;
   const totalLinesChanged = totalLinesAdded + totalLinesDeleted;
   const aiLinesChanged = aiLinesAdded + aiLinesDeleted;
+  const possibleAILinesChanged = possibleAILinesAdded + possibleAILinesDeleted;
+  const weightedAILinesChanged = Math.round(weightedAILinesAdded + weightedAILinesDeleted);
   return {
     aiCommits,
-    nonToolCommits: total - aiCommits,
-    humanCommits: total - aiCommits,
+    possibleAICommits,
+    nonToolCommits: total - aiCommits - possibleAICommits,
+    humanCommits: total - aiCommits - possibleAICommits,
     aiCommitRatio: total > 0 ? aiCommits / total : 0,
+    possibleAICommitRatio: total > 0 ? possibleAICommits / total : 0,
     aiRatio: totalLinesChanged > 0 ? aiLinesChanged / totalLinesChanged : 0,
+    aiLineRatio: totalLinesChanged > 0 ? aiLinesChanged / totalLinesChanged : 0,
+    possibleAILineRatio: totalLinesChanged > 0 ? possibleAILinesChanged / totalLinesChanged : 0,
+    weightedAILineRatio: totalLinesChanged > 0 ? weightedAILinesChanged / totalLinesChanged : 0,
     toolFilter: toolFilter || null,
     aiLinesAdded,
     aiLinesDeleted,
     aiLinesChanged,
+    possibleAILinesAdded,
+    possibleAILinesDeleted,
+    possibleAILinesChanged,
+    weightedAILinesAdded: Math.round(weightedAILinesAdded),
+    weightedAILinesDeleted: Math.round(weightedAILinesDeleted),
+    weightedAILinesChanged,
     totalLinesAdded,
     totalLinesDeleted,
     totalLinesChanged,
-    aiLineRatio: totalLinesChanged > 0 ? aiLinesChanged / totalLinesChanged : 0,
     aiCommitLinesAdded,
     aiCommitLinesDeleted,
     aiFileLinesAdded,
@@ -501,12 +534,19 @@ export function parseGitLogOutput(output, repo = '') {
   return result;
 }
 
+function sanitizeArg(s) {
+  // 移除 shell 特殊字符，防止命令注入
+  return String(s || '').replace(/[`$"\\|;&<>!\n\r]/g, '');
+}
+
 function buildGitArgs(since, until, author) {
   const sinceFull = since.includes('T') ? since : since + 'T00:00:00';
-  const authorArg = author ? ` --author="${author}"` : '';
+  const safeSince = sanitizeArg(sinceFull);
+  const safeUntil = sanitizeArg(until);
+  const authorArg = author ? ` --author="${sanitizeArg(author)}"` : '';
   // 格式：哨兵行(subject) → body 行(可多行) → ENDBODY 行 → numstat 行
   const pretty = `--pretty=format:"${COMMIT_SENTINEL}%H|%ad|%ae|%s%n%B${BODY_END}"`;
-  return `--all --no-renames ${pretty} --date=iso-strict --numstat --since="${sinceFull}" --until="${until}"${authorArg}`;
+  return `--all --no-renames ${pretty} --date=iso-strict --numstat --since="${safeSince}" --until="${safeUntil}"${authorArg}`;
 }
 
 function mergeGitStats(target, source) {
@@ -576,6 +616,21 @@ export async function getGitStatsForMultipleReposAsync(repos, since, until) {
   return merged;
 }
 
+// Per-repo git stats (unmerged), returns Map<repoPath, stats>
+export async function getPerRepoGitStats(repos, since, until) {
+  const results = await Promise.all(
+    repos.map(async repo => {
+      try {
+        const stats = await getGitStatsAsync(repo, since, until, getGitAuthor(repo));
+        return [repo, stats];
+      } catch {
+        return [repo, emptyResult()];
+      }
+    })
+  );
+  return new Map(results);
+}
+
 export function invalidateGitCache() {
   gitCache.clear();
 }