Private web UI

.claude/settings.json

@@ -15,7 +15,9 @@
       "Write(*.md)",
       "Write(*.ts)",
       "Edit(*.css)",
-      "Bash(ls*)"
+      "Bash(ls*)",
+      "Bash(ssh*)",
+      "Bash(curl*)"
     ],
     "deny": []
   },

.claude/skills/analyze-logs/SKILL.md

@@ -0,0 +1,75 @@
+---
+description: Analyze runtime logs, evolve logs, and agent logs to identify error patterns, repeated failures, and evolution opportunities. Use when asked to review logs, diagnose issues, or find improvement areas.
+user-invocable: true
+argument-hint: "[hours=24] [level=error|warn|all]"
+---
+
+# Analyze Logs Skill
+
+Scan all available log sources, identify patterns, and produce an actionable report for self-improvement.
+
+## Log Sources
+
+1. **Runtime Log** (`~/.claude/runtime.log`) — Captured console.error/warn/log from the running process (JSONL format)
+2. **Evolve Log** (`~/.claude/evolve-log.jsonl`) — Self-evolution history: what was changed, tsc pass/fail, restart count
+3. **Agent Logs** (`~/.claude/tasks/conversations/*.log`) — Background agent execution results
+4. **Session History** (`~/.claude/history.jsonl`) — User input history across sessions
+
+## Analysis Steps
+
+### Step 1: Read Runtime Logs
+Read `~/.claude/runtime.log` (the main log file). Parse JSONL entries. Focus on:
+- **error** level entries: categorize by module, extract patterns
+- **warn** level entries: identify recurring warnings
+- Count entries by module to find "noisy" components
+- Identify time clusters (many errors at once = incident)
+
+### Step 2: Read Evolve Log
+Read `~/.claude/evolve-log.jsonl`. Analyze:
+- Which modules were modified most frequently (indicates instability)
+- tsc failures (what caused them?)
+- Patterns in "reason" field — recurring themes
+
+### Step 3: Scan Agent Logs
+Grep across `~/.claude/tasks/conversations/*.log` for:
+- Errors other than "Task cancelled by user" and "Test error"
+- Stack traces
+- Timeout patterns
+- Empty log files (indicates logging bugs)
+
+### Step 4: Cross-Reference
+- Do runtime errors correlate with evolve changes? (regression detection)
+- Do agent failures cluster around certain time periods?
+- Are there modules that appear in both runtime errors AND evolve log? (unstable modules)
+
+## Output Format
+
+Produce a structured report:
+
+```
+## Log Analysis Report (last N hours)
+
+### Runtime Errors Summary
+- Total entries: X (errors: Y, warns: Z)
+- Top error modules: [Module]: count
+- Error patterns: [description]
+
+### Evolution Health
+- Total evolves: X (success: Y, tsc fail: Z)
+- Most modified modules: [module]: count
+- Unstable modules (evolved 3+ times): [list]
+
+### Agent Health
+- Total agent tasks: X (completed: Y, failed: Z, cancelled: W)
+- Real failures (excluding cancels): [list with details]
+
+### Recommendations
+1. [Specific actionable recommendation]
+2. [Another recommendation]
+```
+
+## Important Notes
+- Only do READ operations. Never modify any log files.
+- If runtime.log doesn't exist yet, note that the Logger was just installed and skip that section.
+- Use Grep with `head_limit` when scanning large directories to avoid timeout.
+- Write key findings to the project notebook so they persist across sessions.

docs/marketing/discord.md

@@ -0,0 +1,133 @@
+# Discord Promotion Posts
+
+## Target Communities
+
+1. Anthropic / Claude Community Discord
+2. AI Developer Communities (Cline, Roo Code, Cursor, Windsurf, etc.)
+3. Open Source / Coding Communities
+4. Your own Discord: https://discord.gg/bNyJKk6PVZ
+
+---
+
+## Post 1: General AI/Dev Communities (English)
+
+### Title / First Line
+**I built an open-source Claude Code with a full Web IDE and multi-agent system**
+
+### Body
+
+Hey everyone! I've been working on an open-source AI coding platform called **Claude Code Open** - it started as a reverse-engineering study of Anthropic's Claude Code CLI but evolved into something much bigger.
+
+**What makes it different:**
+
+**Web IDE** - Not just a terminal. It's a full browser-based IDE with Monaco editor, file tree, AI-enhanced code editing, terminal panel, and Git integration. Open `localhost:3456` and start coding.
+
+**Blueprint Multi-Agent System** - Break complex tasks into subtasks and dispatch them across multiple AI agents working in parallel. Smart Planner analyzes requirements, Lead Agent coordinates, Workers execute. Real-time Swarm Console shows everything happening.
+
+**37+ Built-in Tools** - File ops, ripgrep search, shell, Playwright browser automation, database client (PostgreSQL/MySQL/SQLite/Redis/MongoDB), DAP debugger, LSP, scheduled task daemon, and more.
+
+**Self-Evolution** - The AI can modify its own source code, run TypeScript compilation checks, and hot-reload. Full audit log for safety.
+
+**One-Click Install** - Scripts for Windows/macOS/Linux. Docker support too.
+
+Everything runs locally. MIT licensed. No telemetry. Supports Anthropic API / AWS Bedrock / Google Vertex AI.
+
+- GitHub: https://github.com/kill136/claude-code-open
+- Live Demo: http://voicegpt.site:3456/
+- Website: https://www.chatbi.site
+- Discord: https://discord.gg/bNyJKk6PVZ
+
+Would love your feedback!
+
+---
+
+## Post 2: Claude/Anthropic Specific Communities (English)
+
+### Title / First Line
+**Open-source Claude Code alternative with Web UI - 37+ tools, multi-agent, runs locally**
+
+### Body
+
+Built an open-source platform that extends what Claude Code can do:
+
+- **Browser-based IDE** instead of terminal-only - Monaco editor, file tree, inline AI code review
+- **Multi-agent orchestration** - Blueprint system breaks complex projects across parallel AI agents
+- **Database client** - Connect to PostgreSQL, MySQL, SQLite, Redis, MongoDB directly from conversation
+- **Scheduled automation** - "Every morning at 9am, review yesterday's commits and notify me"
+- **Self-evolution** - AI modifies its own source code with safety checks
+- **MCP protocol** support for external tool integration
+
+One-click install for Windows/macOS/Linux. Docker deployment. MIT licensed.
+
+Try the live demo: http://voicegpt.site:3456/
+GitHub: https://github.com/kill136/claude-code-open
+
+Star it if you find it useful!
+
+---
+
+## Post 3: Chinese Communities (中文)
+
+### Title / First Line
+**开源了一个 Claude Code 替代品：Web IDE + 多智能体 + 37+ 工具，一键安装**
+
+### Body
+
+大家好！分享一个我做的开源项目 **Claude Code Open**，一个完整的 AI 编程平台。
+
+**和官方 Claude Code 的区别：**
+
+1. **有 Web IDE** — 浏览器打开就能用，Monaco 编辑器 + 文件树 + AI 辅助编辑，不需要终端
+2. **多智能体协作** — Blueprint 系统把复杂任务拆分给多个 AI Agent 并行处理
+3. **37+ 内置工具** — 文件操作、数据库客户端、浏览器自动化、调试器、定时任务等
+4. **自我进化** — AI 可以修改自己的源码并热重载
+5. **一键安装** — Windows/macOS/Linux 一条命令搞定
+
+MIT 协议，完全开源，数据不出你的机器。
+
+在线体验：http://voicegpt.site:3456/
+GitHub：https://github.com/kill136/claude-code-open
+Discord：https://discord.gg/bNyJKk6PVZ
+
+欢迎 Star 和反馈！
+
+---
+
+## Post 4: Short Version (for channels with character limits)
+
+**Claude Code Open** - Open-source AI coding platform with Web IDE, multi-agent system, 37+ tools.
+
+Features: Browser IDE, Blueprint multi-agent, database client, scheduled tasks, self-evolution, MCP support.
+
+MIT licensed. One-click install. Runs locally.
+
+GitHub: https://github.com/kill136/claude-code-open
+Demo: http://voicegpt.site:3456/
+
+---
+
+## Discord Servers to Post In
+
+### Already posted (from project notebook):
+- Cline
+- Roo Code
+- WeMakeDevs
+- Cursor
+- Windsurf
+
+### New targets:
+1. **Anthropic Community** - discord.gg/anthropic (if exists)
+2. **Claude Users** - search for Claude-focused servers
+3. **AI Tools / AI Coding** - general AI developer communities
+4. **Open Source** - open source project showcase channels
+5. **TypeScript / Node.js** - language-specific communities
+6. **Vibe Coding** - communities for AI-assisted development
+7. **IndieHackers** - indie developer communities
+8. **AI Agent Builders** - communities focused on AI agents
+
+### Posting Tips:
+- Post in #showcase, #projects, #share-your-work, or #self-promotion channels
+- Don't spam - one post per server
+- Engage with comments/questions after posting
+- Include the live demo link - it's the best proof
+- Post during peak hours (US morning / evening)

src/auth/index.ts

@@ -97,6 +97,8 @@ export interface UserProfileResponse {
 const AUTH_DIR = path.join(os.homedir(), '.claude');
 const AUTH_FILE = path.join(AUTH_DIR, 'auth.json');
 const CREDENTIALS_FILE = path.join(AUTH_DIR, 'credentials.json');
+// 用户配置文件（Web UI 保存的 apiKey 等配置）
+const SETTINGS_FILE = path.join(AUTH_DIR, 'settings.json');
 // 官方 Claude Code 的配置文件（存储 primaryApiKey）
 const CONFIG_FILE = path.join(AUTH_DIR, 'config.json');
 // 官方 Claude Code 的 OAuth 凭据文件（存储 claudeAiOauth）
@@ -269,6 +271,24 @@ export function initAuth(): AuthConfig | null {
     return currentAuth;
   }
 
+  // 1c. 检查 settings.json 的 apiKey（Web UI 配置）
+  // 用户在 Web UI 设置页面保存的 API Key，优先级高于 OAuth token 和内置代理
+  if (fs.existsSync(SETTINGS_FILE)) {
+    try {
+      const settings = JSON.parse(fs.readFileSync(SETTINGS_FILE, 'utf-8'));
+      if (settings.apiKey) {
+        currentAuth = {
+          type: 'api_key',
+          accountType: 'api',
+          apiKey: settings.apiKey,
+        };
+        return currentAuth;
+      }
+    } catch (err) {
+      // 忽略解析错误
+    }
+  }
+
   // 2. 检查官方 Claude Code 的 .credentials.json（OAuth token）
   //
   // 重要发现（通过抓包和测试发现）：

src/browser/manager.ts

@@ -247,7 +247,7 @@ function killProc(proc: ChildProcessWithoutNullStreams): void {
     // Windows: use taskkill for reliable termination
     try {
       if (proc.pid) {
-        execSync(`taskkill /pid ${proc.pid} /T /F`, { stdio: 'ignore' });
+        execSync(`taskkill /pid ${proc.pid} /T /F`, { stdio: 'ignore', windowsHide: true });
       }
     } catch {
       proc.kill();

src/cli.ts

@@ -36,6 +36,10 @@ import {
   FAST_MODE_DISPLAY_NAME,
   forcePrefetchPenguinMode,
 } from './fast-mode/index.js';
+import { logger } from './utils/logger.js';
+
+// 初始化运行时日志系统（CLI 模式）
+logger.init({ interceptConsole: true, minLevel: 'info' });
 
 // 工作目录列表
 const additionalDirectories: string[] = [];