imtoagent 0.3.6 → 0.3.8

package/README.md

@@ -4,12 +4,50 @@ Connect Feishu, Telegram, personal WeChat, and WeCom to AI coding agents like Cl
 
 One gateway, multiple IMs, multiple agents, unified port proxy.
 
+## 🚀 Quick Start (5 Minutes)
+
+### Step 1: Install (One Command)
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/imtoagent/imtoagent/main/install.sh | bash
+```
+
+This script detects your environment, installs bun if needed, installs imtoagent, and guides you through setup.
+
+### Step 2: Start
+
+```bash
+imtoagent setup
+```
+
+The interactive wizard guides you through:
+1. Select IM platform (Feishu/Telegram/WeChat/WeCom)
+2. Enter Bot credentials
+3. Choose Agent backend (Claude Code/Codex/OpenCode)
+4. Configure model providers (API keys)
+5. Generate soul files for personality injection
+
+### Step 2: Verify
+
+```bash
+imtoagent status    # check it's running
+```
+
+That's it! Send `/help` to your Bot in the IM to see available commands.
+
+---
+
+**Alternative install methods:** See [Installation Methods](#installation-methods) below for npm global install or source install.
+
+---
+
 ## Architecture
 
 ```
-飞书/Telegram/微信/企微 → IM Registry 工厂 → Bot 实例
-                                          → AgentRuntime SDK → Agent Adapter
-                                                             → 统一 Proxy (:18899) → 上游模型
+IM Platform (Feishu/Telegram/WeChat/WeCom)
+    → IM Registry Factory → Bot Instance
+        → AgentRuntime SDK → Agent Adapter
+            → Unified Proxy (:18899) → Upstream Models
 ```
 
 ### Supported IM Adapters
@@ -43,63 +81,115 @@ One gateway, multiple IMs, multiple agents, unified port proxy.
 | Codex | `npm install -g @openai/codex` |
 | OpenCode | `npm install -g opencode` |
 
-### Installation
+## Installation Methods
 
-#### Method 1: npm global install (recommended)
+### Method 1: One-Click Install (Recommended)
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/imtoagent/imtoagent/main/install.sh | bash
+```
+
+This script does everything automatically:
+- Detects your OS and environment
+- Installs bun if missing
+- Installs or upgrades imtoagent
+- Runs the setup wizard if no configuration exists
+- Starts the gateway and verifies it's running
+
+**Flags:**
+- `--non-interactive` — Skip all prompts, auto-install and auto-start
+- `--skip-bun` — Skip bun installation check
+- `--skip-start` — Don't start the gateway after install
+
+### Method 2: npm Global Install
 
 ```bash
 npm install -g imtoagent
 ```
 
-After installation, it automatically checks whether initial configuration is needed. An interactive terminal will guide you through the setup wizard.
+This is the simplest approach. After installation, the post-install script checks if you need initial configuration.
 
-#### Method 2: Source install
+### Method 3: Source Install
 
 ```bash
-git clone https://github.com/YOUR_USERNAME/imtoagent.git
+git clone https://github.com/imtoagent/imtoagent.git
 cd imtoagent
 bun install
 bun run bin/imtoagent setup
 ```
 
-### First-Time Configuration
+Use this for development or if you want to modify the source code.
+
+### Prerequisites
+
+- **Bun** runtime (≥1.0.0): `brew install oven-sh/bun/bun`
+- **macOS or Linux**
+- **At least one Agent backend** installed (see below)
+
+### Agent Backend Installation
+
+| Backend | Install Command |
+|---------|----------------|
+| Claude Code | `npm install -g @anthropic-ai/claude-agent-sdk` |
+| Codex | `npm install -g @openai/codex` |
+| OpenCode | `npm install -g opencode` |
+
+You can install backends before or after installing imtoagent.
+
+## Configuration
+
+### First-Time Setup
 
 ```bash
 imtoagent setup
 ```
 
-The interactive setup wizard guides you through:
+The interactive wizard will guide you through:
 
-1. **Configure Bot** — Select IM platform + Agent backend
+1. **Configure Bot** — Select IM platform + Agent backend combination
 2. **Configure Model Providers** — Add API credentials (DeepSeek, Dashscope, etc.)
-3. **Generate Soul Files** — Create rules.md / identity.md etc. for each Bot
+3. **Generate Soul Files** — Create personality files (rules.md, identity.md, etc.)
 4. **Write Config Files** — Auto-generate `~/.imtoagent/config.json`
 
-#### Feishu Bot Requirements
+### Platform-Specific Requirements
 
-- Feishu App ID (`cli_...`)
-- Feishu App Secret
-- Feishu app must enable: Bot, Event Subscription, Message Send/Receive permissions
+#### Feishu
+- **App ID** (`cli_...`) and **App Secret**
+- Enable in Feishu app console: Bot, Event Subscription, Message Send/Receive permissions
 
-#### Telegram Bot Requirements
-
-- Telegram Bot Token (obtain from @BotFather)
-- Optional: Proxy address (e.g., `http://127.0.0.1:7890`)
+#### Telegram
+- **Bot Token** from @BotFather
+- Optional: HTTP proxy (e.g., `http://127.0.0.1:7890`)
 
 #### Personal WeChat
+- QR code automatically appears on first `imtoagent start`
+- Scan with WeChat on your phone to complete binding
 
-- QR code automatically pops up on first run of `imtoagent start`
-- Scan with your phone's WeChat to complete binding
+#### WeCom (Enterprise WeChat)
+- Webhook callback URL configuration
+- REST API credentials
 
-### Start the Gateway
+## Running the Gateway
 
-```bash
-imtoagent start     # Start in background
-imtoagent status    # Check running status
-imtoagent stop      # Stop the gateway
-```
+### Basic Commands
+
+| Command | Description |
+|---------|-------------|
+| `imtoagent start` | Start gateway in background |
+| `imtoagent stop` | Stop the gateway |
+| `imtoagent status` | Check running status |
+| `imtoagent restore` | Hot reload recovery |
+| `imtoagent daemon` | Foreground daemon mode (crash auto-restart) |
+
+### Running Modes
 
-### Auto-Start on Boot (macOS launchd)
+- **`start`** — Background mode, terminal returns immediately
+- **`run`** — Foreground mode, real-time logs, Ctrl+C to stop
+- **`daemon`** — Foreground with auto-restart on crash
+
+### Auto-Start on Boot
+
+#### macOS (launchd)
 
 ```bash
 # Create launchd configuration
@@ -134,36 +224,54 @@ cat > ~/Library/LaunchAgents/com.imtoagent.plist << 'EOF'
 </plist>
 EOF
 
-# Load
+# Load the service
 launchctl load ~/Library/LaunchAgents/com.imtoagent.plist
 ```
 
-### Common Commands
+#### Linux (systemd)
 
-| Command | Description |
-|------|------|
-| `imtoagent setup` | Interactive setup wizard |
-| `imtoagent start` | Start gateway in background |
-| `imtoagent stop` | Stop the gateway |
-| `imtoagent status` | Check running status |
-| `imtoagent restore` | Hot reload recovery |
-| `imtoagent daemon` | Foreground daemon mode (suitable for launchd/systemd) |
+Create `/etc/systemd/system/imtoagent.service`:
 
-### Built-In Gateway Commands
+```ini
+[Unit]
+Description=IMtoAgent Gateway
+After=network.target
 
-Send to the Bot in IM chat:
+[Service]
+Type=simple
+User=youruser
+WorkingDirectory=/home/youruser/.imtoagent
+ExecStart=/usr/bin/bun run /usr/lib/node_modules/imtoagent/index.ts daemon
+Restart=on-failure
+RestartSec=5
+
+[Install]
+WantedBy=multi-user.target
+```
+
+```bash
+systemctl daemon-reload
+systemctl enable imtoagent
+systemctl start imtoagent
+```
+
+## Using the Gateway
+
+### Built-In Commands
+
+Send these to your Bot in the IM chat:
 
 | Command | Description |
-|------|------|
-| `/help` | Help information |
+|---------|-------------|
+| `/help` | Show available commands |
 | `/status` | Gateway status |
 | `/stats` | Usage statistics |
-| `/model` | Switch model |
-| `/providers` | View providers |
-| `/memory` | View memory |
+| `/model` | Switch AI model |
+| `/providers` | View model providers |
+| `/memory` | View memory status |
 | `/soul` | Soul management |
-| `/reload` | Reload |
-| `/clear` | Clear session |
+| `/reload` | Reload configuration |
+| `/clear` | Clear conversation session |
 | `/mode` | Switch mode (permission/auto/plan) |
 | `/dir` | Switch working directory |
 
@@ -206,27 +314,66 @@ imtoagent/
 
 ## Data Directory
 
-All runtime data is stored centrally in `~/.imtoagent/`:
+All runtime data is stored in `~/.imtoagent/`:
 
 ```
 ~/.imtoagent/
 ├── config.json          # Main config (Bot + providers + system)
-├── providers.json       # Model provider config
-├── opencode.json        # OpenCode config
-├── sessions/            # Session persistence
+├── providers.json       # Model provider configurations
+├── opencode.json        # OpenCode-specific config
+├── sessions/            # Conversation persistence
 ├── logs/                # Runtime logs
-└── soul/                # Soul files (one directory per Bot)
+└── soul/                # Bot personality files
     ├── ClaudeBot/
     ├── CodexBot/
-    └── ...
+    └── ...              # One directory per Bot
+```
+
+## Troubleshooting
+
+### Common Issues
+
+**Gateway won't start**
+- Check `imtoagent status` for details
+- Verify config with `cat ~/.imtoagent/config.json`
+- Check logs: `cat ~/.imtoagent/logs/*.log`
+
+**Setup wizard stuck**
+- Ensure terminal supports interactive input
+- Try running in a standard terminal (not IDE integrated)
+
+**Bot not responding in IM**
+- Verify credentials in config.json
+- Check IM platform permissions (Feishu events, Telegram webhook, etc.)
+- Ensure the gateway is running: `imtoagent status`
+
+**Port 18899 already in use**
+- Another service is using the proxy port
+- Kill the existing process or change port in config
+
+### Getting Help
+
+- Check logs: `~/.imtoagent/logs/`
+- Run `imtoagent status` for runtime information
+- Open an issue on GitHub with logs attached
 ```
 
 ## Development
 
 ```bash
+# Clone and setup development environment
+git clone https://github.com/imtoagent/imtoagent.git
+cd imtoagent
 bun install
-bun run index.ts          # Run directly
-bun run bin/imtoagent setup  # Run setup wizard
+
+# Run directly
+bun run index.ts
+
+# Run setup wizard
+bun run bin/imtoagent setup
+
+# Run CLI from source
+bun run bin/imtoagent status
 ```
 
 ## License

package/index.ts

@@ -72,8 +72,7 @@ registerIM('wechat', {
   },
 });
 import { startAnthropicProxy, stopAnthropicProxy } from './modules/proxy/anthropic-proxy';
-import { getProxyUsage, resetProxyUsage, initCodexProxyConfig } from './modules/proxy/codex-proxy';
-import { initOpenCodeConfig } from './modules/agent/opencode';
+import { initCodexProxyConfig } from './modules/proxy/codex-proxy';
 import { checkRateLimit, setRateLimitConfig } from './modules/rate-limiter';
 import { setCurrentBot } from './modules/bot-context';
 import { getDataDir, getSessionsDir, getSoulDir, getBotKey, getRestoreMarkerPath } from './modules/utils/paths';
@@ -904,11 +903,6 @@ async function main() {
         upstream: codexCfg.upstream || 'https://api.deepseek.com/v1/chat/completions',
         apiKey,
       });
-      const ocCfg = config.opencode || {};
-      initOpenCodeConfig({
-        serverUrl: ocCfg.serverUrl || 'http://localhost:4096',
-        defaultModel: ocCfg.defaultModel || { providerID: 'anthropic', modelID: 'claude-sonnet-4-5' },
-      });
       const rlCfg = config.rateLimit || {};
       if (rlCfg.enabled !== false) {
         setRateLimitConfig({

package/modules/agent/codex-exec-server.ts

@@ -22,9 +22,6 @@ let _config: ExecServerConfig = {
   maxToolCallsPerTurn: 80,
 };
 
-export function setExecServerConfig(cfg: Partial<ExecServerConfig>) {
-  _config = { ..._config, ...cfg };
-}
 
 // ================================================================
 // 事件类型
@@ -510,4 +507,3 @@ export async function shutdownAppServer(): Promise<void> {
 export const shutdownExecServer = shutdownAppServer;
 
 // 向后兼容的类型别名
-export type CodexExecServerClient = CodexAppServerClient;

package/modules/proxy/codex-proxy.ts

@@ -525,19 +525,10 @@ async function streamResponse(upstreamRes: Response, resWriter: WritableStreamDe
 }
 
 // ================================================================
-// ================================================================
-// usage 累加器 — 供网关读取 Codex 的 Token/成本统计
+// usage 累加器 — 内部统计，供 accumulateProxyUsage 记录
 // ================================================================
 let _proxyUsage = { inputTokens: 0, outputTokens: 0 };
 
-export function getProxyUsage() {
-  return { ..._proxyUsage };
-}
-
-export function resetProxyUsage() {
-  _proxyUsage = { inputTokens: 0, outputTokens: 0 };
-}
-
 export function accumulateProxyUsage(inputTokens: number, outputTokens: number) {
   _proxyUsage.inputTokens += inputTokens;
   _proxyUsage.outputTokens += outputTokens;
@@ -646,12 +637,3 @@ export async function handleCodexRequest(
     res.writeHead(500, { 'Content-Type': 'application/json' }); res.end(JSON.stringify({ error: 'internal error' })); return;
   }
 }
-
-// 兼容旧引用（不再启动独立服务器）
-export function startCodexProxy(_port?: number): Promise<number> {
-  console.log('[Codex Proxy] Merged into port 18899');
-  return Promise.resolve(18899);
-}
-export function stopCodexProxy(): Promise<void> {
-  return Promise.resolve();
-}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "imtoagent",
-  "version": "0.3.6",
+  "version": "0.3.8",
   "description": "IM ↔ Agent 统一网关 — 飞书/Telegram/微信/企业微信对接 Claude Code/Codex/OpenCode",
   "type": "module",
   "bin": {
@@ -32,7 +32,8 @@
   },
   "repository": {
     "type": "git",
-    "url": "git+https://github.com/YOUR_USERNAME/imtoagent.git"
+    "url": "https://github.com/imtoagent/imtoagent.git",
+    "directory": ""
   },
   "keywords": [
     "im",
@@ -50,7 +51,7 @@
   "author": "Keyi",
   "license": "MIT",
   "bugs": {
-    "url": "https://github.com/YOUR_USERNAME/imtoagent/issues"
+    "url": "https://github.com/imtoagent/imtoagent/issues"
   },
-  "homepage": "https://github.com/YOUR_USERNAME/imtoagent#readme"
+  "homepage": "https://github.com/imtoagent/imtoagent#readme"
 }

package/modules/agent/claude.ts

@@ -1,160 +0,0 @@
-// Claude Agent 模块
-// 对接 Claude Agent SDK，通过 :18899 Proxy 调用 Provider
-
-import { query } from '@anthropic-ai/claude-agent-sdk';
-import type { AgentContext } from '../types';
-import { parseToBlocks, type UnifiedBlock } from '../capabilities';
-import { buildSystemPrompt, resolveCapabilities } from '../prompt-builder';
-import type { SDKMessage } from '@anthropic-ai/claude-agent-sdk';
-
-// ================================================================
-// 工具函数
-// ================================================================
-function resolveAlias(modelSpec: string): string {
-  const i = modelSpec.indexOf('/');
-  return i >= 0 ? modelSpec.slice(i + 1) : modelSpec;
-}
-
-function extractText(msg: SDKMessage): string | null {
-  if (msg.type !== 'assistant') return null;
-  const content = (msg as any).message?.content;
-  if (!Array.isArray(content)) return null;
-  return content.filter((b: any) => b.type === 'text').map((b: any) => b.text).join('') || null;
-}
-
-function extractToolInfo(msg: SDKMessage): { name: string; summary: string } | null {
-  if (msg.type !== 'assistant') return null;
-  const content = (msg as any).message?.content;
-  if (!Array.isArray(content)) return null;
-  const tool = content.find((b: any) => b.type === 'tool_use');
-  if (!tool?.name) return null;
-  const input = tool.input || {};
-  let summary = '';
-  if (['Read','Edit','Write'].includes(tool.name) && input.file_path) {
-    const p = String(input.file_path);
-    summary = p.includes('/') ? p.split('/').pop()! : p;
-  } else if (tool.name === 'Bash' && input.command) {
-    summary = String(input.command).trim().slice(0, 60);
-  }
-  return { name: tool.name, summary };
-}
-
-// ================================================================
-// Claude 模块类
-// ================================================================
-
-/**
- * 注意：此模块目前依赖宿主 Bot 实例的方法（reply/sendProgress/addToolLog 等）。
- * Phase 2 第一步先原样提取，后续逐步解耦为干净接口。
- */
-export class ClaudeAgentModule {
-  private ctx: AgentContext;
-
-  constructor(ctx: AgentContext) {
-    this.ctx = ctx;
-  }
-
-  async handleMessage(chatId: string, text: string, session: any) {
-    session.generator.push({
-      type: 'user', message: { role: 'user', content: [{ type: 'text', text }] },
-    });
-    if (!session.running) this._startLoop(chatId);
-  }
-
-  private async _startLoop(chatId: string) {
-    const ctx = this.ctx;
-    const session = ctx.sessions.get(chatId);
-    if (!session || session.running) return;
-    session.running = true;
-    console.log(`[${ctx.name}] Claude loop started chat=${chatId.slice(-8)}`);
-
-    try {
-      const modelSpec = ctx.activeModel;
-      const modelName = modelSpec.slice(modelSpec.indexOf('/') + 1);
-      const aliases = ctx.modelAliases;
-      const customEnv: Record<string, string> = {
-        ...process.env,
-        ANTHROPIC_BASE_URL: 'http://localhost:18899',
-        ANTHROPIC_API_KEY: '', ANTHROPIC_AUTH_TOKEN: '', ANTHROPIC_MODEL: '',
-        ANTHROPIC_DEFAULT_SONNET_MODEL: resolveAlias(aliases.sonnet),
-        ANTHROPIC_DEFAULT_OPUS_MODEL: resolveAlias(aliases.opus),
-        ANTHROPIC_DEFAULT_HAIKU_MODEL: resolveAlias(aliases.haiku),
-      };
-
-      const queryOptions: any = {
-        cwd: session.cwd || ctx.defaultCwd,
-        maxTurns: 50, model: modelName,
-        permissionMode: session.permissionMode || 'bypassPermissions',
-        persistSession: true,
-      };
-      if (session.sdkSessionId) {
-        queryOptions.resume = session.sdkSessionId;
-      } else {
-        queryOptions.sessionId = crypto.randomUUID();
-      }
-
-      const botName = ctx.name;
-      const systemPrompt = buildSystemPrompt({
-        imModule: ctx.imModule || null,
-        botName,
-      });
-      console.log(`[Claude] 📝 system prompt built (${systemPrompt.length} chars, bot=${botName})`);
-      queryOptions.systemPrompt = systemPrompt;
-
-      const q = query({
-        prompt: session.generator.generate(),
-        options: queryOptions, env: customEnv,
-      });
-
-      let fullResponse = '', toolCalls = 0;
-      let callInput = 0, callOutput = 0, callCost = 0, callDur = 0;
-
-      for await (const msg of q) {
-        const msgAny = msg as any;
-        if (msgAny.session_id && !session.sdkSessionId) {
-          session.sdkSessionId = msgAny.session_id;
-          ctx.persistSession(chatId, session);
-        }
-        const text = extractText(msg);
-        if (text) fullResponse += text;
-        const toolInfo = extractToolInfo(msg);
-        if (toolInfo) { toolCalls++; ctx.addToolLog(chatId, toolInfo); }
-
-        if (msg.type === 'result') {
-          const result = msg as any;
-          callInput = result.usage?.input_tokens || 0;
-          callOutput = result.usage?.output_tokens || 0;
-          callCost = result.total_cost_usd || 0;
-          callDur = result.duration_ms || 0;
-
-          ctx.accumulateStats(session, {
-            inputTokens: callInput, outputTokens: callOutput,
-            costUSD: callCost, durationMs: callDur,
-            numTurns: result.num_turns || 0,
-          });
-
-          if (result.subtype === 'error' || result.subtype === 'cancelled') {
-            await ctx.reply(chatId, `❌ ${result.error || result.result || 'Unknown error'}`);
-          } else if (fullResponse) {
-            await ctx.sendFormattedReply(chatId, fullResponse);
-          } else {
-            await ctx.reply(chatId, `✅ Completed (${toolCalls} steps)`);
-          }
-
-          ctx.flushToolLog(chatId);
-          const costStr = callCost > 0 ? `Cost $${callCost.toFixed(4)}\n` : '';
-          await ctx.sendProgress(chatId,
-            `✅ Completed (${toolCalls} steps)\nInput ${callInput.toLocaleString()} Token\nOutput ${callOutput.toLocaleString()} Token\n${costStr}Duration ${(callDur/1000).toFixed(1)}s`);
-          fullResponse = ''; toolCalls = 0;
-        }
-      }
-    } catch (e: any) {
-      console.error(`[${ctx.name}] Claude error: ${e.message}`);
-      await ctx.reply(chatId, `❌ ${e.message}`);
-    } finally {
-      session.running = false;
-      session.generator.close();
-      ctx.persistSession(chatId, session);
-    }
-  }
-}

package/modules/agent/codex.ts

@@ -1,275 +0,0 @@
-// Codex Agent 模块
-// 对接 Codex CLI，通过 :18899 Proxy 调用 Provider
-
-import { getProxyUsage, resetProxyUsage } from '../proxy/codex-proxy';
-import { calculateCost } from '../proxy/anthropic-proxy';
-import type { AgentContext } from '../types';
-import { getAppServerManager, type AgentEvent } from './codex-exec-server';
-// ================================================================
-// 类型
-// ================================================================
-interface CodexJsonEvent {
-  type: string;
-  thread_id?: string;
-  item?: { type: string; text?: string; name?: string; arguments?: string; output?: string };
-  text?: string;
-  delta?: string;
-  message?: { content?: { type: string; text?: string }[] };
-  error?: string;
-}
-
-const TOOL_NAMES: Record<string, string> = {
-  Bash: 'Execute command', Read: 'Read file', Edit: 'Edit file', Write: 'Write file',
-  Glob: 'Search files', Grep: 'Search content', WebSearch: 'Web search', WebFetch: 'Fetch webpage',
-  NotebookEdit: 'Edit Notebook',
-  // Codex tool names
-  command_execution: 'Execute command', exec_command: 'Execute command', write_stdin: 'Write to stdin', update_plan: 'Update plan',
-  request_user_input: 'Request input', apply_patch: 'Apply patch', view_image: 'View image',
-  spawn_agent: 'Spawn agent', send_input: 'Send input', resume_agent: 'Resume agent',
-  wait_agent: 'Wait agent', close_agent: 'Close agent',
-};
-
-// ================================================================
-// Codex CLI 调用
-// ================================================================
-function processCodexStream(stdout: string, onTool?: (name: string, args: Record<string, any>) => void): { threadId: string; response: string } {
-  let threadId = '', response = '';
-  for (const line of stdout.split('\n')) {
-    if (!line.trim()) continue;
-    try {
-      const evt: CodexJsonEvent = JSON.parse(line);
-      if (evt.type === 'thread.started' && evt.thread_id) {
-        threadId = evt.thread_id;
-      } else if (evt.type === 'item.completed') {
-        if (evt.item?.type === 'agent_message') {
-          response = (response ? response + '\n' : '') + (evt.item.text || '');
-        } else if (TOOL_NAMES[evt.item?.type || ''] && onTool) {
-          onTool(evt.item.type || 'unknown', { command: (evt.item as any).command || '' });
-        }
-      } else if (evt.type === 'error' || evt.type === 'thread.error') {
-        console.error(`[Codex] event error: ${(evt as any).message || (evt as any).error || JSON.stringify(evt)}`);
-      }
-    } catch {}
-  }
-  return { threadId, response };
-}
-
-async function spawnCodexExec(
-  cwd: string, prompt: string,
-  onTool?: (name: string, args: Record<string, any>) => void
-): Promise<{ threadId: string; response: string }> {
-  console.error(`[Codex] spawnExec cwd=${cwd} prompt_len=${prompt.length}`);
-  const child = Bun.spawn(['codex', 'exec', '-p', 'imtoagent', '-s', 'danger-full-access',
-    '--skip-git-repo-check', '--json', prompt], {
-    cwd, stdout: 'pipe', stderr: 'pipe',
-  });
-
-  // Safe read stdout/stderr: catch subprocess kill exceptions, ensure reject carries Error object
-  let stdout = '', stderr = '';
-  try {
-    [stdout, stderr] = await Promise.all([
-      new Response(child.stdout).text().catch((e: any) => { throw new Error(`stdout read failed: ${e?.message || e}`); }),
-      new Response(child.stderr).text().catch((e: any) => { throw new Error(`stderr read failed: ${e?.message || e}`); }),
-    ]);
-  } catch (ioErr: any) {
-    // Subprocess may have been killed, try to get exit code
-    try { child.kill('SIGKILL'); } catch {}
-    throw new Error(`codex exec I/O error: ${ioErr.message}`);
-  }
-
-  const code = await child.exited.catch(() => -1);
-  console.error(`[Codex] exec exit=${code} stdout_len=${stdout.length} stderr_len=${stderr.length}`);
-  const { threadId, response } = processCodexStream(stdout, onTool);
-  if (code !== 0 || !threadId) throw new Error(`codex exec exit ${code}: ${stderr.slice(0, 300)}`);
-  return { threadId, response };
-}
-
-async function spawnCodexResume(
-  cwd: string, threadId: string, prompt: string,
-  onTool?: (name: string, args: Record<string, any>) => void
-): Promise<{ response: string }> {
-  console.error(`[Codex] spawnResume cwd=${cwd} threadId=${threadId.slice(-8)} prompt_len=${prompt.length}`);
-  const child = Bun.spawn(['codex', 'exec', 'resume', threadId,
-    '--dangerously-bypass-approvals-and-sandbox', '-c', 'model_provider=imtoagent', '-c', 'model=gpt-5.5', '--json', '--skip-git-repo-check', prompt], {
-    cwd, stdout: 'pipe', stderr: 'pipe',
-  });
-
-  let stdout = '', stderr = '';
-  try {
-    [stdout, stderr] = await Promise.all([
-      new Response(child.stdout).text().catch((e: any) => { throw new Error(`stdout read failed: ${e?.message || e}`); }),
-      new Response(child.stderr).text().catch((e: any) => { throw new Error(`stderr read failed: ${e?.message || e}`); }),
-    ]);
-  } catch (ioErr: any) {
-    try { child.kill('SIGKILL'); } catch {}
-    throw new Error(`codex exec resume I/O error: ${ioErr.message}`);
-  }
-
-  const code = await child.exited.catch(() => -1);
-  console.error(`[Codex] resume exit=${code} stdout_len=${stdout.length} stderr_len=${stderr.length}`);
-  if (code !== 0) throw new Error(`codex exec resume exit ${code}: ${stderr.slice(0, 300)}`);
-  return { response: processCodexStream(stdout, onTool).response };
-}
-
-// ================================================================
-// App-Server 路径（优先使用——流式输出 + 长记忆 + 崩溃不丢上下文）
-// ================================================================
-async function runViaAppServer(
-  cwd: string, prompt: string, chatId: string, session: any,
-  onTool: (name: string, args: Record<string, any>) => void,
-  isFresh: boolean
-): Promise<{ threadId: string; response: string; usage: { inputTokens: number; outputTokens: number } }> {
-  const manager = getAppServerManager();
-  const client = await manager.getClient(chatId);
-
-  // app-server 同进程内线程存活
-  // 但进程重启后旧 thread 过期，需要判断代际
-  const currentGen = manager.generation;
-  const threadExpired = session._appServerGen !== currentGen;
-  if (isFresh || !session.codexThreadId || threadExpired) {
-    session.codexThreadId = await client.startThread(cwd);
-    session._appServerGen = currentGen;
-    console.log(`[Codex] app-server new thread=${session.codexThreadId.slice(-8)}${threadExpired ? ' (process restarted)' : ''}`);
-  }
-  // 后续消息直接 turn/start（同线程延续上下文）
-
-  await client.sendPrompt(session.codexThreadId, prompt, cwd);
-
-  let response = '';
-  let totalUsage = { inputTokens: 0, outputTokens: 0 };
-  const startTime = Date.now();
-  const MAX_DURATION = 600_000; // 10 分钟总超时
-
-  for await (const event of client.receiveEvents()) {
-    // 超时保护
-    if (Date.now() - startTime > MAX_DURATION) {
-      console.error('[Codex] app-server task timed out (10min)');
-      break;
-    }
-
-    switch (event.type) {
-      case 'text_delta':
-        response += event.textDelta || '';
-        break;
-      case 'tool_call':
-        onTool(event.toolName || 'unknown', event.toolInput || {});
-        break;
-      case 'turn_result':
-        // 累加多轮 token（非终端 turn_result 来自每轮的 turn/completed）
-        totalUsage.inputTokens += event.usage?.inputTokens || 0;
-        totalUsage.outputTokens += event.usage?.outputTokens || 0;
-        break;
-      case 'error':
-        throw new Error(`app-server error: ${event.error}`);
-    }
-  }
-
-  return { threadId: session.codexThreadId, response, usage: totalUsage };
-}
-
-// ================================================================
-// Codex 模块类
-// ================================================================
-export class CodexAgentModule {
-  private ctx: AgentContext;
-
-  constructor(ctx: AgentContext) {
-    this.ctx = ctx;
-  }
-
-  async handleMessage(chatId: string, text: string, session: any) {
-    const ctx = this.ctx;
-    const cwd = session.cwd || ctx.defaultCwd;
-    console.log(`[${ctx.name}] Codex chat=${chatId.slice(-8)} startFresh=${session.startFresh || false}`);
-
-    const onTool = (name: string, args: Record<string, any>) => {
-      const cmd = args.cmd || args.command || '';
-      const justification = args.justification || args.description || '';
-      const summary = cmd ? cmd.slice(0, 80) : justification.slice(0, 80);
-      ctx.addToolLog(chatId, { name, summary });
-    };
-
-    resetProxyUsage();
-    try {
-      let effectiveText = text;
-      if (session.codexMode === 'plan') {
-        effectiveText = `[Mode: Plan then execute] Please create a clear plan first, wait for my confirmation before executing. User request: ${text}`;
-      }
-
-      const isFresh = session.startFresh || !session.codexThreadId;
-      let response: string;
-      let execServerUsage: { inputTokens: number; outputTokens: number } | null = null;
-
-      session.startFresh = false;
-      await ctx.sendProgress(chatId, '💭 Thinking...');
-
-      // 优先尝试 app-server
-      console.error(`[${ctx.name}] DEBUG entering app-server branch, isFresh=${isFresh}, threadId=${session.codexThreadId?.slice(-8)}`);
-      let useExecFallback = false;
-      try {
-        const r = await runViaAppServer(cwd, effectiveText, chatId, session, onTool, isFresh);
-        response = r.response;
-        execServerUsage = r.usage;
-      } catch (appErr: any) {
-        const errMsg = appErr.message || '';
-        console.error(`[${ctx.name}] app-server failed: ${errMsg}`);
-
-        // thread not found → app-server 进程内线程丢了，尝试重新创建
-        if (errMsg.includes('thread not found') || errMsg.includes('Thread not found')) {
-          try {
-            session.codexThreadId = undefined;
-            const r2 = await runViaAppServer(cwd, effectiveText, chatId, session, onTool, true);
-            response = r2.response;
-            execServerUsage = r2.usage;
-            console.error(`[${ctx.name}] app-server thread rebuilt successfully`);
-          } catch {
-            useExecFallback = true;
-          }
-        } else {
-          useExecFallback = true;
-        }
-      }
-
-      if (useExecFallback) {
-        getAppServerManager().removeClient(chatId);
-        if (isFresh || !session.codexThreadId) {
-          const r = await spawnCodexExec(cwd, effectiveText, onTool);
-          session.codexThreadId = r.threadId;
-          response = r.response;
-          console.log(`[${ctx.name}] Fresh session thread=${r.threadId.slice(-8)}`);
-        } else {
-          const r = await spawnCodexResume(cwd, session.codexThreadId, effectiveText, onTool);
-          response = r.response;
-        }
-      }
-      ctx.flushToolLog(chatId);
-
-      // 优先使用 app-server 返回的 usage，否则从 proxy 获取
-      const usage = execServerUsage || getProxyUsage();
-      if (usage.inputTokens > 0 || usage.outputTokens > 0) {
-        const cost = calculateCost(ctx.activeModel, usage.inputTokens, usage.outputTokens);
-        ctx.accumulateStats(session, { ...usage, costUSD: cost });
-        await ctx.sendProgress(chatId,
-          `Input ${usage.inputTokens.toLocaleString()} Token\nOutput ${usage.outputTokens.toLocaleString()} Token\nCost $${cost.toFixed(4)}`);
-      }
-
-      if (response) {
-        await ctx.sendFormattedReply(chatId, response);
-      }
-      else await ctx.reply(chatId, '✅ Completed');
-      ctx.persistSession(chatId, session);
-    } catch (e: any) {
-      console.error(`[${ctx.name}] Codex error: ${e.message}`);
-      session.codexThreadId = undefined;
-      try {
-        const r = await spawnCodexExec(cwd, text, onTool);
-        session.codexThreadId = r.threadId;
-        if (r.response) {
-          await ctx.sendFormattedReply(chatId, r.response);
-        }
-      } catch (e2: any) {
-        await ctx.reply(chatId, `❌ Processing failed: ${e2.message}`);
-      }
-    }
-  }
-}

package/modules/agent/opencode.ts

@@ -1,247 +0,0 @@
-// OpenCode Agent 模块
-// 对接 opencode serve HTTP API，通过 :18899 Anthropic Proxy 调 Provider
-//
-// 设计：薄模块 — oc serve 管理 session/工具/provider，Gateway 只做 IM 翻译
-
-import type { AgentContext, SessionData } from '../types';
-import { parseToBlocks } from '../capabilities';
-import { resolveCapabilities, buildSystemPrompt } from '../prompt-builder';
-import { calculateCost } from '../proxy/anthropic-proxy';
-import * as path from 'path';
-import * as fs from 'fs';
-import { getDataDir } from '../utils/paths';
-
-// ================================================================
-// 配置（从 config.json 读取）
-// ================================================================
-
-interface OpenCodeConfig {
-  serverUrl: string;
-  defaultModel: { providerID: string; modelID: string };
-}
-
-let _ocConfig: OpenCodeConfig | null = null;
-
-export function initOpenCodeConfig(cfg: OpenCodeConfig) {
-  _ocConfig = cfg;
-}
-
-function getOcConfig(): OpenCodeConfig {
-  if (!_ocConfig) {
-    try {
-      
-      const raw = JSON.parse(fs.readFileSync(path.join(getDataDir(), 'config.json'), 'utf-8'));
-      const oc = raw.opencode || {};
-      _ocConfig = {
-        serverUrl: oc.serverUrl || 'http://localhost:4096',
-        defaultModel: oc.defaultModel || { providerID: 'anthropic', modelID: 'claude-sonnet-4-5' },
-      };
-    } catch {
-      _ocConfig = {
-        serverUrl: 'http://localhost:4096',
-        defaultModel: { providerID: 'anthropic', modelID: 'claude-sonnet-4-5' },
-      };
-    }
-  }
-  return _ocConfig;
-}
-
-const OC_SERVER_URL = () => getOcConfig().serverUrl;
-const OC_DEFAULT_MODEL = () => getOcConfig().defaultModel;
-
-// ================================================================
-// OpenCode Server HTTP 客户端
-// ================================================================
-
-interface OcMessagePart {
-  type: string;
-  text?: string;
-  tool_call?: { name: string; arguments: Record<string, any> };
-  tool_result?: { content: string };
-}
-
-interface OcMessage {
-  info: { id: string; role: string; model?: string };
-  parts: OcMessagePart[];
-}
-
-async function ocCreateSession(title: string): Promise<string> {
-  const ac = new AbortController();
-  const timer = setTimeout(() => ac.abort(), 300_000);
-  const res = await fetch(`${OC_SERVER_URL()}/session`, {
-    method: 'POST',
-    headers: { 'Content-Type': 'application/json' },
-    body: JSON.stringify({ title }),
-    signal: ac.signal,
-  }).finally(() => clearTimeout(timer));
-  if (!res.ok) throw new Error(`oc create session: ${res.status} ${await res.text()}`);
-  const data = await res.json();
-  return data.id;
-}
-
-async function ocSendPrompt(
-  sessionId: string,
-  initialText: string,
-  system?: string,
-  onTool?: (name: string, args: Record<string, any>) => void
-): Promise<{ response: string }> {
-  const MAX_TURNS = 50;
-  const TURN_TIMEOUT = 300_000;
-  const startTime = Date.now();
-  const MAX_DURATION = 600_000; // 10 分钟总超时
-
-  let promptText = initialText;
-  let accumulatedResponse = '';
-  let turn = 0;
-
-  while (turn < MAX_TURNS) {
-    if (Date.now() - startTime > MAX_DURATION) {
-      console.error('[OpenCode] Task timed out (10min)');
-      break;
-    }
-    turn++;
-
-    const body: any = {
-      model: OC_DEFAULT_MODEL(),
-      parts: [{ type: 'text', text: promptText }],
-    };
-    if (turn === 1 && system) body.system = system;
-
-    const ac = new AbortController();
-    const timer = setTimeout(() => ac.abort(), TURN_TIMEOUT);
-    const res = await fetch(`${OC_SERVER_URL()}/session/${sessionId}/message`, {
-      method: 'POST',
-      headers: { 'Content-Type': 'application/json' },
-      body: JSON.stringify(body),
-      signal: ac.signal,
-    }).finally(() => clearTimeout(timer));
-    if (!res.ok) throw new Error(`oc send prompt: ${res.status} ${await res.text()}`);
-
-    const data: OcMessage = await res.json();
-    let hasToolCall = false;
-
-    for (const part of data.parts || []) {
-      if (part.type === 'text' && part.text) {
-        accumulatedResponse += (accumulatedResponse ? '\n' : '') + part.text;
-      } else if (part.type === 'tool_call' && part.tool_call) {
-        hasToolCall = true;
-        if (onTool) onTool(part.tool_call.name, part.tool_call.arguments);
-      }
-    }
-
-    // 纯文本回复或没有 tool_call → 任务完成
-    if (!hasToolCall) break;
-
-    // 有 tool_call，继续推进（空 prompt）
-    promptText = 'Continue';
-  }
-
-  return { response: accumulatedResponse };
-}
-
-async function ocDeleteSession(sessionId: string): Promise<void> {
-  await fetch(`${OC_SERVER_URL()}/session/${sessionId}`, { method: 'DELETE' }).catch(() => {});
-}
-
-async function ocHealthCheck(): Promise<boolean> {
-  try {
-    const res = await fetch(`${OC_SERVER_URL()}/global/health`);
-    return res.ok;
-  } catch {
-    return false;
-  }
-}
-
-// ================================================================
-// Agent 模块类
-// ================================================================
-export class OpenCodeAgentModule {
-  private ctx: AgentContext;
-
-  constructor(ctx: AgentContext) {
-    this.ctx = ctx;
-  }
-
-  async handleMessage(chatId: string, text: string, session: SessionData) {
-    const ctx = this.ctx;
-    console.log(`[${ctx.name}] OpenCode chat=${chatId.slice(-8)}`);
-
-    // 工具回调
-    const onTool = (name: string, args: Record<string, any>) => {
-      const summary = args.command || args.cmd || args.query || JSON.stringify(args).slice(0, 80);
-      ctx.addToolLog(chatId, { name, summary });
-    };
-
-    try {
-      // ① Plan 模式处理
-      let effectiveText = text;
-      if (session.codexMode === 'plan') {
-        effectiveText = `[Mode: Plan then execute] Please create a clear plan first, wait for my confirmation before executing. User request: ${text}`;
-      }
-
-      // ② 清理标记
-      const shouldClear = session.startFresh;
-      session.startFresh = false;
-
-      // ③ 获取或创建 OpenCode session
-      if (shouldClear || !session.ocSessionId) {
-        if (session.ocSessionId) {
-          await ocDeleteSession(session.ocSessionId);
-          console.log(`[${ctx.name}] Cleared oc session=${session.ocSessionId.slice(-8)}`);
-        }
-        session.ocSessionId = await ocCreateSession(chatId);
-        console.log(`[${ctx.name}] Created oc session=${session.ocSessionId.slice(-8)}`);
-      }
-
-      // ④ 发送进度提示
-      await ctx.sendProgress(chatId, '💭 Thinking...');
-
-      // ④.⑤ 构建系统提示词
-      const systemPrompt = buildSystemPrompt({
-        imModule: ctx.imModule || null,
-        botName: ctx.name,
-      });
-      console.log(`[${ctx.name}] 📝 system prompt built (${systemPrompt.length} chars, bot=${ctx.name})`);
-
-      // ⑤ 发送 prompt（多轮循环）
-      const { response } = await ocSendPrompt(
-        session.ocSessionId,
-        effectiveText,
-        systemPrompt,
-        onTool,
-      );
-
-      // ⑥ 刷新工具日志
-      ctx.flushToolLog(chatId);
-
-      // ⑦ 输出
-      if (response) {
-        await ctx.sendFormattedReply(chatId, response);
-      } else {
-        await ctx.reply(chatId, '✅ Completed');
-      }
-
-      // ⑧ 统计
-      const { sharedState } = await import('../proxy/anthropic-proxy');
-      const lastUsage = (sharedState as any).lastCallUsage;
-      if (lastUsage && (lastUsage.inputTokens > 0 || lastUsage.outputTokens > 0)) {
-        const cost = calculateCost(ctx.activeModel, lastUsage.inputTokens, lastUsage.outputTokens);
-        ctx.accumulateStats(session, { ...lastUsage, costUSD: cost });
-        await ctx.sendProgress(chatId,
-          `Input ${lastUsage.inputTokens.toLocaleString()} Token\nOutput ${lastUsage.outputTokens.toLocaleString()} Token\nCost $${cost.toFixed(4)}`);
-      }
-
-      // ⑨ 持久化会话
-      ctx.persistSession(chatId, session);
-
-    } catch (err: any) {
-      console.error(`[${ctx.name}] OpenCode error: ${err.message}`);
-      await ctx.reply(chatId, `⚠️ OpenCode error: ${err.message}`);
-    }
-  }
-
-  /** 健康检查 */
-  async healthCheck(): Promise<boolean> {
-    return ocHealthCheck();
-  }
-}

package/scripts/postinstall.ts

@@ -1,70 +0,0 @@
-#!/usr/bin/env bun
-// ================================================================
-// postinstall.ts — npm 安装后引导脚本
-// ================================================================
-// package.json 中 "scripts": { "postinstall": "bun run scripts/postinstall.ts" }
-// 安装后自动运行，检测是否需要初始化配置
-// 如果是全新安装且终端交互，自动引导进入 setup
-// ================================================================
-
-import * as fs from 'fs';
-import * as path from 'path';
-import { execSync } from 'child_process';
-
-const HOME = process.env.HOME || process.env.USERPROFILE || '';
-const DATA_DIR = path.join(HOME, '.imtoagent');
-
-try {
-  const configExists = fs.existsSync(path.join(DATA_DIR, 'config.json'));
-
-  if (configExists) {
-    console.log(`
-✅ imtoagent upgraded successfully!
-   Data directory: ${DATA_DIR}
-   Configuration file kept as-is, no need to reconfigure.
-   Run "imtoagent start" to start the gateway.
-`);
-  } else {
-    console.log(`
-╔══════════════════════════════════════════════════════════╗
-║                                                          ║
-║   🎉  imtoagent installed successfully!                  ║
-║                                                          ║
-║   First-time use requires configuring IM credentials and a model provider   ║
-║                                                          ║
-╚══════════════════════════════════════════════════════════╝
-`);
-
-    // 检测是否为交互式终端，是则自动引导进入 setup
-    if (process.stdin.isTTY) {
-      const readline = await import('readline');
-      const rl = readline.createInterface({ input: process.stdin, output: process.stdout });
-      const answer = await new Promise<string>(resolve => {
-        rl.question('Launch the configuration wizard now? [Y/n]: ', resolve);
-      });
-      rl.close();
-
-      const yes = answer.trim().toLowerCase();
-      if (yes === '' || yes === 'y' || yes === 'yes') {
-        console.log('\n🚀 Launching configuration wizard...\n');
-        // 调用 setup 向导
-        const pkgDir = path.resolve(import.meta.dirname, '..');
-        execSync('bun run bin/imtoagent setup', {
-          cwd: pkgDir,
-          stdio: 'inherit',
-          env: { ...process.env },
-        });
-      } else {
-        console.log('\nRun "imtoagent setup" later to configure.');
-      }
-    } else {
-      console.log('   Run "imtoagent setup" to start configuring');
-      console.log('   Then run "imtoagent start" to start the gateway\n');
-    }
-  }
-} catch (e: any) {
-  // Silently fail, do not affect installation
-  if (e.message && !e.message.includes('readline')) {
-    console.error(`[postinstall] Failed to display message: ${e.message}`);
-  }
-}