了解复盘镜

复盘镜（WorkMirror）是一个 Windows 托盘常驻的本地回顾工具：自动收集你电脑上的工作“痕迹”（窗口、代码变更、浏览历史），整理成可追溯的工作片段，并生成日报/周报与技能变化。

你能得到什么

每天/每周的回顾：每条结论都能点开看到证据覆盖与明细。
技能树与能力变化：增长来源可追溯到贡献的工作片段。
离线也能用：不配 AI Key 也能生成规则版摘要与回顾。
遇到“空白页/没数据”也不需要找作者：状态页会告诉你缺口在哪一层，并给出下一步怎么修。

隐私与本地优先

数据默认写入本地 SQLite（默认 ./data/workmirror.db），不做云同步/账号体系。
本地 HTTP 仅监听 127.0.0.1，端口随机分配。
启用 AI 时仅发送脱敏后的最小必要文本，用于生成摘要与建议。

它以什么形式运行（v0.2）

workmirror.exe（Windows 10/11 托盘常驻）
内置本地 HTTP Server（随机端口，仅监听 127.0.0.1）
内置 Web UI（React/Vite SPA，优先加载同目录 frontend/dist/，否则回退内嵌资源）
端口发现：写入 ./data/http_base_url.txt

它怎么工作（简述）

采集 → 生成工作片段（切分/聚合/摘要）→ 汇总到技能与回顾 → 每条结论可点开证据链。

目前不包含什么

云同步/账号体系/团队协作
macOS/Linux（当前阶段）
复杂目标管理（TODO/OKR/习惯）

工作原理（高级）

下面两张图展示复盘镜在本机如何运行，以及“证据 → 工作片段（会话） → 回顾/技能”的口径链路。

运行时组件图

图：单进程本地架构（仅监听 127.0.0.1），UI 与 API 同源服务。

证据链流水线图（口径）

图：会话为权威输入；离线规则默认可用，AI 作为可选增强；失败路径通过状态与诊断动作显式暴露。

开发者：目录结构（Go / React）

cmd/workmirror-agent/：Windows Agent 入口（托盘、采集器、HTTP API/UI）
internal/collector/：采集器（窗口/Diff/浏览历史）
internal/server/：本地 HTTP Server（JSON API + SSE + 静态 UI）
internal/service/：Session/Skill/回顾等业务逻辑
internal/repository/：SQLite 数据访问与迁移门闸（schema_version）
frontend/：前端源码（开发时通过 VITE_API_TARGET 指向本地 Agent）

口径说明

上层的技能/回顾不直接从 raw 统计，而是统一以会话为输入，保证“同一套数字、同一条证据链”。

出了问题怎么自助排查

当数据为空/不可信时，优先从状态页定位缺口（采集/切分/分析/配置），并按提示执行：补配 watch paths、重建某天会话、导出诊断包等。

快速开始

复盘镜是一个 Windows 托盘常驻 Agent + 本地 Web UI。目标是让新用户在 10 分钟完成闭环：配置 watch paths → 看到当天会话 → 能点开证据。

1. 获取程序

推荐从 GitHub Releases 下载 Windows 版本；或从源码构建（需 Go 1.25.4，且仅支持 Windows 构建）。

go build -trimpath -ldflags "-H=windowsgui -s -w" -o .\workmirror.exe .\cmd\workmirror-agent\

说明：上述命令会构建为 GUI 子系统，运行时不会弹出前台控制台窗口。

2. 初始化

首次运行后，复盘镜会最小化到系统托盘。右键点击图标选择 "打开面板"。

默认数据库：./data/workmirror.db（相对路径以可执行文件目录为基准）
本地 HTTP：只监听 127.0.0.1，端口为随机分配
端口发现：启动后写入 ./data/http_base_url.txt

推荐“便携分发”：一个 workmirror.exe + 同目录 config/ 与 data/。

3. 配置监听目录

为了让 Diff Collector 生效，你需要在设置页添加你的 Git 项目目录（watch paths）：

C:\Users\Dev\Projects\Go-Backend

可选：启用浏览器采集（自动探测 Chrome/Edge History 路径），并保持隐私脱敏默认开启。

本地接口（高级）

本地 HTTP 仅监听 127.0.0.1，端口随机分配（见 ./data/http_base_url.txt）。

状态与可观测性（P0）

GET /api/status：采集/管道/AI/最近错误的汇总快照
GET /api/events：SSE（数据变更/配置变更等事件）

维护动作

POST /api/maintenance/sessions/rebuild：按日期重建会话
POST /api/maintenance/sessions/enrich：按日期补全语义（启用 AI 时）
GET /api/diagnostics/export：导出脱敏诊断包（zip）

常见问题

需要联网吗？

不需要。未配置 AI Key 时，复盘镜会以规则口径生成会话摘要与回顾；配置 AI 后，仅在需要语义化摘要/建议时调用外部提供商。

为什么页面为空/没有数据？

先打开 /status（或状态页入口）查看采集健康与最近错误。
Diff 采集需要配置 diff.watch_paths（或在设置/引导里选择项目目录）。
浏览器采集依赖本地 History 文件路径；可在设置中查看自动探测结果并手动指定。

支持 macOS/Linux/手机吗？

当前阶段只支持 Windows 10/11，且不提供云同步/账号体系。

如何导出诊断信息？

在状态页使用“导出诊断包”，或访问 /api/diagnostics/export 下载脱敏后的 zip。

隐私策略

最后更新：2025年12月

复盘镜遵循 Local-first & Privacy by default：默认本地存储、默认脱敏、默认可离线。

数据存储

采集数据（窗口事件、Diff、浏览历史等）会写入本地 SQLite 数据库（默认 ./data/workmirror.db）。本地 HTTP 仅监听 127.0.0.1，不会暴露到局域网/公网。

默认脱敏

默认开启脱敏：URL 会清空 query/fragment 并保留域名与“存在路径”的事实；窗口标题/浏览标题会按规则替换敏感片段。你可以在配置中调整 privacy.patterns。

AI 处理

不配置任何 Key 时，不会调用外部 AI。启用 AI 后，仅在需要生成语义摘要/建议时发送经脱敏的最小必要文本到你配置的提供商（例如 DeepSeek / SiliconFlow）。

设置与隐私

配置文件为 YAML：默认路径 ./config/config.yaml（首次启动会自动生成）。可以参考模板 ./config/config.yaml.example。

你大概率只需要改这几项

diff:
  enabled: true
  watch_paths:
    - "C:\\Users\\Dev\\Projects\\Go-Backend"

browser:
  enabled: true
  history_path: "" # 为空则自动探测

storage:
  db_path: "./data/workmirror.db"

privacy:
  enabled: true

AI Key 与环境变量

建议通过环境变量注入 Key（避免写入磁盘/提交仓库），例如 DEEPSEEK_API_KEY、SILICONFLOW_API_KEY。

服务条款

复盘镜（WorkMirror）是开源软件（MIT License），按"现状"提供，不对适用性或无缺陷作任何保证。你在本机使用时产生的任何数据与后果由你自行承担。

本项目不提供云服务、不提供账号体系；本地 HTTP 服务仅用于本机 UI 与本地集成。

About WorkMirror

WorkMirror is a Windows system tray resident local review tool: automatically collects your work "traces" (windows, code changes, browser history), organizes them into traceable work sessions, and generates daily/weekly reports and skill evolution.

What You Get

Daily/weekly reviews: Every conclusion can be clicked to see evidence coverage and details.
Skill tree and capability evolution: Growth sources can be traced back to contributing work sessions.
Works offline: Rule-based summaries and reviews work without AI Key.
No need to contact the author when you see "blank page/no data": The status page tells you where the gap is and what to do next.

Privacy & Local-first

Data is stored in local SQLite by default (./data/workmirror.db), no cloud sync or account system.
Local HTTP only listens on 127.0.0.1, port randomly assigned.
When AI is enabled, only the minimum necessary desensitized text is sent for generating summaries and suggestions.

How It Runs (v0.2)

workmirror.exe (Windows 10/11 system tray resident)
Built-in local HTTP Server (random port, only listens on 127.0.0.1)
Built-in Web UI (React/Vite SPA, prioritizes loading from frontend/dist/ in the same directory, otherwise falls back to embedded resources)
Port discovery: Written to ./data/http_base_url.txt

How It Works (Brief)

Collection → Generate work sessions (segmentation/aggregation/summary) → Aggregate to skills and reviews → Each conclusion can be clicked to view evidence chain.

What's Not Included

Cloud sync/account system/team collaboration
macOS/Linux (at this stage)
Complex goal management (TODO/OKR/habits)

How It Works (Advanced)

The diagrams below show how WorkMirror runs locally and the evidence → work sessions → reviews/skills pipeline.

Runtime Component Diagram

Fig: Single-process local architecture (only listens on 127.0.0.1), UI and API served from same origin.

Evidence Chain Pipeline Diagram

Fig: Sessions are the authoritative input; offline rules work by default, AI as optional enhancement; failure paths exposed via status and diagnostic actions.

Developers: Directory Structure (Go / React)

cmd/workmirror-agent/: Windows Agent entry (tray, collectors, HTTP API/UI)
internal/collector/: Collectors (window/Diff/browser history)
internal/server/: Local HTTP Server (JSON API + SSE + static UI)
internal/service/: Session/Skill/review business logic
internal/repository/: SQLite data access and migration gate (schema_version)
frontend/: Frontend source (during development, use VITE_API_TARGET to point to local Agent)

Pipeline Explanation

Upper-level skills/reviews don't count directly from raw data, but uniformly use sessions as input, ensuring "same numbers, same evidence chain".

Self-troubleshooting

When data is empty/unreliable, first locate the gap from the status page (collection/segmentation/analysis/config), and follow the prompts: add watch paths, rebuild sessions for a day, export diagnostic package, etc.

Quick Start

WorkMirror is a Windows system tray resident Agent + local Web UI. The goal is to let new users complete the loop in 10 minutes: configure watch paths → see today's sessions → click to view evidence.

1. Get the Program

Recommended to download from GitHub Releases; or build from source (requires Go 1.25.4, Windows only).

go build -trimpath -ldflags "-H=windowsgui -s -w" -o .\workmirror.exe .\cmd\workmirror-agent\

Note: The above command builds as GUI subsystem, no foreground console window will pop up when running.

2. Initialization

After first run, WorkMirror minimizes to system tray. Right-click the icon and select "Open Panel".

Default database: ./data/workmirror.db (relative to executable directory)
Local HTTP: Only listens on 127.0.0.1, randomly assigned port
Port discovery: Written to ./data/http_base_url.txt on startup

Recommended "portable distribution": One workmirror.exe + config/ and data/ in the same directory.

3. Configure Watch Directories

To enable Diff Collector, add your Git project directories (watch paths) in the settings page:

C:\Users\Dev\Projects\Go-Backend

Optional: Enable browser collection (auto-detect Chrome/Edge History path), and keep privacy desensitization enabled by default.

Local API (Advanced)

Local HTTP only listens on 127.0.0.1, port randomly assigned (see ./data/http_base_url.txt).

Status & Observability (P0)

GET /api/status: Summary snapshot of collection/pipeline/AI/recent errors
GET /api/events: SSE (data changes/config changes events)

Maintenance Actions

POST /api/maintenance/sessions/rebuild: Rebuild sessions by date
POST /api/maintenance/sessions/enrich: Enrich semantics by date (when AI enabled)
GET /api/diagnostics/export: Export desensitized diagnostic package (zip)

FAQ

Do I need internet?

No. Without AI Key configured, WorkMirror generates session summaries and reviews using rule-based approach; with AI configured, external providers are only called when semantic summaries/suggestions are needed.

Why is the page empty/no data?

First check /status (or status page entry) for collection health and recent errors.
Diff collection requires configuring diff.watch_paths (or select project directories in settings/onboarding).
Browser collection depends on local History file path; check auto-detection results in settings and specify manually if needed.

Does it support macOS/Linux/mobile?

Currently only Windows 10/11 is supported, with no cloud sync or account system.

How to export diagnostic info?

Use "Export Diagnostic Package" on the status page, or access /api/diagnostics/export to download desensitized zip.

Privacy Policy

Last updated: December 2025

WorkMirror follows Local-first & Privacy by default: local storage by default, desensitization by default, offline-capable by default.

Data Storage

Collected data (window events, Diff, browser history, etc.) is written to local SQLite database (default ./data/workmirror.db). Local HTTP only listens on 127.0.0.1, not exposed to LAN/public network.

Default Desensitization

Desensitization is enabled by default: URLs have query/fragment cleared, keeping only domain and "path exists" fact; window/browser titles have sensitive fragments replaced by rules. You can adjust privacy.patterns in config.

AI Processing

Without any Key configured, no external AI is called. With AI enabled, desensitized minimum necessary text is sent to your configured provider (e.g., DeepSeek / SiliconFlow) only when semantic summaries/suggestions are needed.

Settings & Privacy

Config file is YAML: default path ./config/config.yaml (auto-generated on first startup). Refer to template ./config/config.yaml.example.

You Probably Only Need to Change These

diff:
  enabled: true
  watch_paths:
    - "C:\\Users\\Dev\\Projects\\Go-Backend"

browser:
  enabled: true
  history_path: "" # empty for auto-detect

storage:
  db_path: "./data/workmirror.db"

privacy:
  enabled: true

AI Key & Environment Variables

Recommended to inject Key via environment variables (avoid writing to disk/committing to repo), e.g., DEEPSEEK_API_KEY, SILICONFLOW_API_KEY.

Terms of Service

WorkMirror is open-source software (MIT License), provided "as is" without any warranty of fitness or defect-free. Any data and consequences generated from your local use are your own responsibility.

This project does not provide cloud services or account system; local HTTP service is only for local UI and local integration.

每天自动复盘你的工作 一键生成日报

仪表盘

每日总结已生成

把 日报与复盘 写得有据可查

自动日报/周报与复盘

每条结论都能回溯

默认本地，离线也能用

为 Windows 工程师打造