venombrowser 4.1.0

package/.env.example

@@ -0,0 +1,26 @@
+# VenomBrowser Bridge Server — Kimi-style configuration
+# v4.0.0
+
+# ─── Server Config ──────────────────────────────────────────────────────────
+
+# Single port for the daemon (HTTP + WebSocket)
+BRIDGE_PORT=10086
+
+# WebSocket port for extension (BRIDGE_PORT + 1 by default)
+WS_PORT=10087
+
+# ─── LLM Provider Config ─────────────────────────────────────────────────────
+# Used by bin/venombrowser-agent.js and bin/venombrowser-chat.js
+
+# Which provider to use: "openrouter" or "ollama"
+LLM_PROVIDER=openrouter
+
+# OpenRouter (Default)
+# Get a free API key at https://openrouter.ai/keys
+OPENROUTER_API_KEY=your-sk-or-v1-key
+OPENROUTER_MODEL=openrouter/free
+
+# Ollama (Local)
+# Requires Ollama running locally (ollama serve)
+OLLAMA_HOST=http://localhost:11434
+OLLAMA_MODEL=qwen2.5:7b

package/bin/venombrowser-agent.js

@@ -0,0 +1,182 @@
+#!/usr/bin/env node
+/**
+ * VenomBrowser Agent — CLI entry point (Kimi-style)
+ *
+ * Standalone browser automation agent powered by any configured LLM provider.
+ * Connects to the VenomBrowser daemon and executes browser tasks.
+ *
+ * v4.0.0 — Kimi-style architecture:
+ *   - Uses new bridge-client with Kimi-style protocol
+ *   - Screenshot returns file path, NOT base64 (fixes cost leak)
+ *   - Uses snapshot() with @e refs for element targeting
+ *
+ * Usage:
+ *   node bin/venombrowser-agent.js "search google for weather in tokyo"
+ */
+
+require('dotenv').config({ path: require('path').join(__dirname, '..', '.env') });
+
+const { BrowserBridge } = require('../src/bridge-client');
+const { createProvider } = require('../src/providers');
+const { TOOL_DEFINITIONS, executeTool } = require('../src/tools/browser-tools');
+
+const MAX_STEPS = parseInt(process.env.MAX_STEPS || '25', 10);
+const session = `agent-${Date.now()}`;
+
+// ─── Logging ─────────────────────────────────────────────────────────────────
+function log(level, message, data = null) {
+  const timestamp = new Date().toISOString().slice(11, 19);
+  const emoji = {
+    info: 'ℹ️', success: '✅', warning: '⚠️', error: '❌',
+    tool: '🔧', agent: '🤖', result: '📊'
+  }[level] || 'ℹ️';
+
+  console.log(`[${timestamp}] ${emoji} ${message}`);
+  if (data && typeof data === 'object') {
+    console.log(`         ${JSON.stringify(data, null, 2).split('\n').join('\n         ')}`);
+  }
+}
+
+// ─── Main ────────────────────────────────────────────────────────────────────
+async function main() {
+  let provider;
+  try {
+    provider = createProvider();
+  } catch (err) {
+    console.error(`❌ Provider error: ${err.message}`);
+    process.exit(1);
+  }
+
+  const bridge = new BrowserBridge({ session });
+
+  console.log('🚀 VenomBrowser Agent v4.0.0 (Kimi-style)');
+  console.log('═'.repeat(55));
+  log('info', `Provider: ${provider}`);
+  log('info', `Session: ${session}`);
+  log('info', `Max steps: ${MAX_STEPS}`);
+  console.log('═'.repeat(55));
+
+  // System checks
+  log('info', 'Running system checks...');
+
+  try {
+    await provider.healthCheck();
+    log('success', `LLM provider ready: ${provider}`);
+  } catch (err) {
+    log('error', `LLM check failed: ${err.message}`);
+    process.exit(1);
+  }
+
+  try {
+    const connected = await bridge.isConnected();
+    if (!connected) throw new Error('VenomBrowser daemon not connected');
+    log('success', 'VenomBrowser daemon connected');
+  } catch (err) {
+    log('error', `Daemon error: ${err.message}`);
+    log('info', 'Make sure daemon is running: npm start');
+    process.exit(1);
+  }
+
+  console.log();
+
+  const task = process.argv.slice(2).join(' ');
+  if (!task) {
+    console.error('❌ Usage: node venombrowser-agent.js "<task description>"');
+    process.exit(1);
+  }
+
+  // ─── Agent loop ──────────────────────────────────────────────────────
+  const sessionStart = Date.now();
+  let totalToolCalls = 0;
+
+  const messages = [
+    {
+      role: 'system',
+      content: `You are VenomBrowser, a browser automation agent controlling a real Chrome browser.
+
+CRITICAL RULES:
+- ALWAYS call snapshot() after navigating to understand the page before clicking anything.
+- Use @e refs from snapshot to click/fill elements (e.g. click("@e3"), fill("@e5", "text")).
+- NEVER guess CSS selectors. Use @e refs from snapshot instead.
+
+NAVIGATION STRATEGY:
+1. navigate() to the target URL
+2. snapshot() to see all interactive elements with @e refs
+3. click("@eN") or fill("@eN", "text") to interact
+
+FORM FILLING: Call snapshot() to see all form fields with their @e refs. Use fill() for inputs, click() for submit.
+
+ERROR RECOVERY: If an action fails, re-run snapshot() to get fresh @e refs. Never repeat the same failed action.
+
+When done, respond with plain text summarizing what you found or accomplished.`
+    },
+    { role: 'user', content: task }
+  ];
+
+  for (let step = 1; step <= MAX_STEPS; step++) {
+    log('info', `═══ Step ${step}/${MAX_STEPS} ═══`);
+
+    try {
+      const reply = await provider.chat(messages, TOOL_DEFINITIONS);
+
+      if (reply.content) {
+        log('agent', reply.content);
+      }
+
+      if (!reply.tool_calls || reply.tool_calls.length === 0) {
+        const duration = ((Date.now() - sessionStart) / 1000).toFixed(1);
+        log('success', 'Agent completed the task!');
+        log('info', `Duration: ${duration}s | Tools: ${totalToolCalls} | Steps: ${step}`);
+        return;
+      }
+
+      const assistantMsg = { role: 'assistant', content: reply.content || '' };
+      if (reply.tool_calls) assistantMsg.tool_calls = reply.tool_calls;
+      messages.push(assistantMsg);
+
+      for (const call of reply.tool_calls) {
+        const name = call.function.name;
+        const args = call.function.arguments || {};
+
+        log('tool', `→ ${name}(${JSON.stringify(args)})`);
+        totalToolCalls++;
+
+        let resultPayload;
+        try {
+          const result = await executeTool(bridge, name, args);
+          resultPayload = { success: true, result };
+
+          const preview = JSON.stringify(result);
+          log('result', preview.length > 200 ? preview.slice(0, 200) + '...' : preview);
+        } catch (error) {
+          resultPayload = { success: false, error: error.message };
+          log('error', `Tool error: ${error.message}`);
+        }
+
+        messages.push({
+          role: 'tool',
+          tool_call_id: call.id,
+          name,
+          content: JSON.stringify(resultPayload)
+        });
+      }
+    } catch (error) {
+      log('error', `Step ${step} failed: ${error.message}`);
+      messages.push({
+        role: 'system',
+        content: `Error occurred: ${error.message}. Try a different approach.`
+      });
+    }
+
+    console.log();
+  }
+
+  const duration = ((Date.now() - sessionStart) / 1000).toFixed(1);
+  log('warning', `Hit MAX_STEPS (${MAX_STEPS}) without completion.`);
+  log('info', `Duration: ${duration}s | Tools: ${totalToolCalls}`);
+}
+
+main().catch(err => {
+  log('error', `Agent failed: ${err.message}`);
+  process.exit(1);
+});

package/bin/venombrowser-chat.js

@@ -0,0 +1,300 @@
+#!/usr/bin/env node
+/**
+ * VenomBrowser Interactive Chat Agent — CLI entry point (Kimi-style)
+ *
+ * Live chat interface with the configured LLM provider + VenomBrowser daemon.
+ * Talk to the agent naturally and give it browser automation tasks in real-time.
+ *
+ * v4.0.0 — Kimi-style architecture:
+ *   - Uses new bridge-client with Kimi-style protocol
+ *   - Screenshot returns file path, NOT base64 (fixes cost leak)
+ *   - Uses snapshot() with @e refs for element targeting
+ *
+ * Usage:
+ *   node bin/venombrowser-chat.js
+ */
+
+require('dotenv').config({ path: require('path').join(__dirname, '..', '.env') });
+const readline = require('readline');
+const { BrowserBridge } = require('../src/bridge-client');
+const { createProvider } = require('../src/providers');
+const { TOOL_DEFINITIONS, executeTool } = require('../src/tools/browser-tools');
+
+const MAX_STEPS = parseInt(process.env.MAX_STEPS || '25', 10);
+const session = `chat-${Date.now()}`;
+
+// ─── Logging ─────────────────────────────────────────────────────────────────
+function log(level, message, skipTimestamp = false) {
+  const timestamp = new Date().toISOString().slice(11, 19);
+  const emoji = {
+    info: 'ℹ️', success: '✅', warning: '⚠️', error: '❌',
+    tool: '🔧', agent: '🤖', user: '👤', system: '🔔'
+  }[level] || 'ℹ️';
+
+  if (skipTimestamp) {
+    console.log(`${emoji} ${message}`);
+  } else {
+    console.log(`[${timestamp}] ${emoji} ${message}`);
+  }
+}
+
+// ─── System Prompt (Kimi-style) ─────────────────────────────────────────────
+const SYSTEM_PROMPT = `You are VenomBrowser, a browser automation assistant controlling a real Chrome browser.
+
+## CRITICAL RULES — ALWAYS FOLLOW THESE
+1. **ALWAYS use snapshot() after navigating** to understand what's on the page. This gives you an accessibility tree with @e refs for all interactive elements.
+2. **Use click("@eN") and fill("@eN", "text")** to interact with elements. The @e refs from snapshot are MUCH more reliable than CSS selectors.
+3. **NEVER guess CSS selectors.** The @e refs replace selector guessing entirely.
+
+## WORKFLOW FOR ANY TASK
+1. navigate() to the target URL
+2. snapshot() to see all interactive elements with @e refs
+3. click("@eN") or fill("@eN", "text") to interact
+4. Repeat steps 2-3 as the page changes
+
+## NAVIGATION STRATEGY
+1. **Direct URL first**: For searches, use URL params: google.com/search?q=...
+2. **@e ref click second**: Use click("@eN") after calling snapshot()
+3. **Text-based click third**: Use click(text: "visible text") as fallback
+
+## FILLING FORMS
+1. Call snapshot() to see all form fields with their @e refs
+2. Use fill("@eN", text) for each field
+3. Use click("@eN") to submit
+
+## EXTRACTING DATA
+- Use evaluate() for structured data extraction
+- Use snapshot() to understand page structure
+
+## SCROLLING
+- Use evaluate("window.scrollBy(0, 800)") to scroll
+- Then snapshot() again to discover newly visible elements
+
+## TAB MANAGEMENT
+- Use list_tabs() to see all open tabs
+- Use close_tab() / navigate(url, { newTab: true })
+
+## ERROR RECOVERY
+- If an element is stale, re-run snapshot() to get fresh @e refs
+- If a click doesn't work, try a different @e ref
+- Never repeat the same failed action
+
+## RESPONSE STYLE
+- Be concise. Summarize what you found/accomplished.
+- When done, give the final answer directly.`;
+
+// ─── Main Chat Interface ─────────────────────────────────────────────────────
+class ChatAgent {
+  constructor() {
+    try {
+      this.provider = createProvider();
+    } catch (err) {
+      console.error(`❌ Provider error: ${err.message}`);
+      process.exit(1);
+    }
+
+    this.bridge = new BrowserBridge({ session });
+
+    this.conversationHistory = [
+      { role: 'system', content: SYSTEM_PROMPT }
+    ];
+
+    this.rl = readline.createInterface({
+      input: process.stdin,
+      output: process.stdout,
+      prompt: '\n👤 You: '
+    });
+  }
+
+  async start() {
+    console.log('🚀 VenomBrowser Interactive Chat Agent v4.0.0 (Kimi-style)');
+    console.log('═'.repeat(60));
+    log('info', `Provider: ${this.provider}`, true);
+    log('info', `Session: ${session}`, true);
+    log('info', `Max steps per task: ${MAX_STEPS}`, true);
+    console.log('═'.repeat(60));
+
+    try {
+      await this._systemCheck();
+      console.log();
+
+      log('agent', 'Hello! I can help you automate browser tasks.', true);
+      log('agent', 'Just tell me what you want me to do with the browser!', true);
+      console.log();
+      console.log('Examples:');
+      console.log('   • "go to google and search for python tutorials"');
+      console.log('   • "take a screenshot of this page"');
+      console.log('   • "find the top news stories on hacker news"');
+      console.log();
+      console.log('Commands:');
+      console.log('   • "clear"  - clear conversation history');
+      console.log('   • "status" - show system status');
+      console.log('   • "quit"   - exit the chat');
+
+      this.rl.prompt();
+
+      this.rl.on('line', async (input) => {
+        const message = input.trim();
+
+        if (message === '') {
+          this.rl.prompt();
+          return;
+        }
+
+        const cmd = message.toLowerCase();
+        if (cmd === 'quit' || cmd === 'exit') {
+          log('system', 'Goodbye!', true);
+          this.rl.close();
+          process.exit(0);
+        }
+
+        if (cmd === 'clear') {
+          this.conversationHistory = [this.conversationHistory[0]];
+          log('system', 'Conversation history cleared', true);
+          this.rl.prompt();
+          return;
+        }
+
+        if (cmd === 'status') {
+          try {
+            const status = await this.bridge.getStatus();
+            log('info', `Bridge connected: ${status.connected}`, true);
+            log('info', `Commands executed: ${status.commands_executed}`, true);
+            log('info', `Uptime: ${status.uptime_seconds}s`, true);
+            log('info', `Provider: ${this.provider}`, true);
+          } catch (error) {
+            log('error', `Status check failed: ${error.message}`, true);
+          }
+          this.rl.prompt();
+          return;
+        }
+
+        console.log();
+        log('agent', 'Working on it...', true);
+
+        try {
+          const result = await this._processUserMessage(message);
+
+          if (!result.printedInline) {
+            console.log();
+            log('agent', result.response, true);
+          }
+
+          if (result.toolCalls > 0) {
+            log('info', `Done — ${result.steps} step(s), ${result.toolCalls} tool call(s).`, true);
+          }
+        } catch (error) {
+          console.log();
+          log('error', `Sorry, I encountered an error: ${error.message}`, true);
+        }
+
+        this.rl.prompt();
+      });
+
+      this.rl.on('close', () => {
+        log('system', 'Goodbye!', true);
+        process.exit(0);
+      });
+
+    } catch (err) {
+      log('error', `Startup failed: ${err.message}`);
+      process.exit(1);
+    }
+  }
+
+  async _systemCheck() {
+    log('system', 'Checking system status...');
+
+    await this.provider.healthCheck();
+    log('success', `LLM provider ready: ${this.provider}`);
+
+    const connected = await this.bridge.isConnected();
+    if (!connected) throw new Error('VenomBrowser daemon not connected. Is the daemon running?');
+
+    log('success', 'VenomBrowser daemon connected');
+  }
+
+  async _processUserMessage(userInput) {
+    this.conversationHistory.push({ role: 'user', content: userInput });
+
+    let stepCount = 0;
+    let toolCallCount = 0;
+    let lastContent = null;
+    let printedInline = false;
+
+    while (stepCount < MAX_STEPS) {
+      stepCount++;
+
+      const reply = await this.provider.chat(this.conversationHistory, TOOL_DEFINITIONS);
+
+      const assistantMsg = { role: 'assistant', content: reply.content || '' };
+      if (reply.tool_calls) assistantMsg.tool_calls = reply.tool_calls;
+      this.conversationHistory.push(assistantMsg);
+
+      lastContent = reply.content;
+
+      if (!reply.tool_calls || reply.tool_calls.length === 0) {
+        if (reply.content) {
+          console.log();
+          log('agent', reply.content, true);
+          printedInline = true;
+        }
+        return {
+          response: reply.content || 'Task completed.',
+          completed: true,
+          steps: stepCount,
+          toolCalls: toolCallCount,
+          printedInline
+        };
+      }
+
+      if (reply.content) {
+        log('agent', reply.content, true);
+        printedInline = true;
+      }
+
+      for (const call of reply.tool_calls) {
+        toolCallCount++;
+        const name = call.function.name;
+        const args = call.function.arguments || {};
+
+        log('tool', `→ ${name}(${JSON.stringify(args)})`, true);
+
+        let resultPayload;
+        try {
+          const result = await executeTool(this.bridge, name, args);
+          resultPayload = { success: true, result };
+
+          if (name === 'navigate') log('success', `✓ Loaded: ${result.title || args.url}`, true);
+          else if (name === 'snapshot') log('success', `✓ Got ${result.tree?.children?.length || 0} elements`, true);
+          else if (name === 'click') log('success', '✓ Clicked successfully', true);
+          else if (name === 'fill') log('success', '✓ Filled successfully', true);
+          else if (name === 'screenshot') log('success', `✓ Screenshot saved: ${result.path}`, true);
+          else log('success', `✓ ${name} completed`, true);
+
+        } catch (error) {
+          resultPayload = { success: false, error: error.message };
+          log('error', `Failed: ${error.message}`, true);
+        }
+
+        this.conversationHistory.push({
+          role: 'tool',
+          tool_call_id: call.id,
+          name,
+          content: JSON.stringify(resultPayload)
+        });
+      }
+    }
+
+    return {
+      response: `Hit the step limit (${MAX_STEPS}). The task might need to be broken into smaller parts.`,
+      completed: false,
+      steps: stepCount,
+      toolCalls: toolCallCount,
+      printedInline: false
+    };
+  }
+}
+
+const chatAgent = new ChatAgent();
+chatAgent.start().catch(console.error);

package/bin/venombrowser-cli.js

@@ -0,0 +1,194 @@
+#!/usr/bin/env node
+/**
+ * VenomBrowser CLI — Unified command dispatcher
+ *
+ * Usage:
+ *   venombrowser setup           — run the interactive setup wizard
+ *   venombrowser start           — start the bridge daemon
+ *   venombrowser status          — check daemon + extension status
+ *   venombrowser connect         — auto-register with known agents
+ *   venombrowser agent "<task>"  — run a one-shot browser task
+ *   venombrowser uninstall       — remove all MCP entries + stop daemon
+ *   venombrowser chat            — start interactive chat agent
+ *   venombrowser help            — show this help
+ */
+
+const path = require('path');
+const { fork, execSync } = require('child_process');
+const http = require('http');
+
+const BIN_DIR = __dirname;
+const ENV_PATH = path.join(BIN_DIR, '..', '.env');
+
+// Load .env for BRIDGE_PORT
+require('dotenv').config({ path: ENV_PATH });
+
+const BRIDGE_PORT = parseInt(process.env.BRIDGE_PORT || '10086', 10);
+
+// ─── Subcommand handlers ────────────────────────────────────────────────────
+
+function showHelp() {
+  console.log(`
+🕷️  VenomBrowser CLI — Browser Automation Bridge
+
+Usage: venombrowser <command> [options]
+
+Commands:
+  setup           Run the interactive setup wizard
+  start           Start the bridge daemon
+  status          Check daemon & extension connection status
+  connect         Auto-register MCP server with known agents
+  uninstall       Remove MCP entries, stop daemon (does NOT npm uninstall)
+  agent "<task>"  Run a one-shot browser automation task
+  chat            Start the interactive chat agent
+  help            Show this help message
+
+Examples:
+  venombrowser setup
+  venombrowser start
+  venombrowser status
+  venombrowser agent "search google for weather in tokyo"
+  venombrowser connect
+`);
+}
+
+function runSetup() {
+  const child = fork(path.join(BIN_DIR, 'venombrowser-setup.js'), process.argv.slice(3), {
+    stdio: 'inherit'
+  });
+  child.on('exit', (code) => process.exit(code || 0));
+}
+
+function runStart() {
+  // Load dotenv and require bridge-server directly (same as venombrowser-server.js)
+  require('dotenv').config({ path: ENV_PATH });
+  require('../src/bridge-server');
+}
+
+function runStatus() {
+  const url = `http://127.0.0.1:${BRIDGE_PORT}/health`;
+
+  const req = http.get(url, { timeout: 3000 }, (res) => {
+    let body = '';
+    res.on('data', (chunk) => { body += chunk; });
+    res.on('end', () => {
+      try {
+        const data = JSON.parse(body);
+        const serverUp = data.server === 'ok';
+        const extConnected = data.extension === 'connected';
+
+        console.log('🕷️  VenomBrowser Status');
+        console.log('═'.repeat(40));
+        console.log(`   Server:    ${serverUp ? '✅ Running' : '❌ Not running'} (port ${data.port || BRIDGE_PORT})`);
+        console.log(`   Extension: ${extConnected ? '✅ Connected' : '❌ Disconnected'}`);
+        console.log('═'.repeat(40));
+
+        if (serverUp && extConnected) {
+          console.log('   Status: Connected');
+          process.exit(0);
+        } else if (serverUp) {
+          console.log('   Status: Disconnected (extension not connected)');
+          process.exit(1);
+        } else {
+          console.log('   Status: Disconnected');
+          process.exit(1);
+        }
+      } catch {
+        console.error('❌ Invalid response from daemon');
+        process.exit(1);
+      }
+    });
+  });
+
+  req.on('error', () => {
+    console.log('🕷️  VenomBrowser Status');
+    console.log('═'.repeat(40));
+    console.log(`   Server:    ❌ Not running (port ${BRIDGE_PORT})`);
+    console.log(`   Extension: ❌ Disconnected`);
+    console.log('═'.repeat(40));
+    console.log('   Status: Disconnected');
+    console.log('');
+    console.log('   Start the daemon with: venombrowser start');
+    process.exit(1);
+  });
+
+  req.on('timeout', () => {
+    req.destroy();
+    console.log('🕷️  VenomBrowser Status');
+    console.log('═'.repeat(40));
+    console.log(`   Server:    ❌ Not responding (port ${BRIDGE_PORT})`);
+    console.log(`   Extension: ❌ Disconnected`);
+    console.log('═'.repeat(40));
+    console.log('   Status: Disconnected');
+    process.exit(1);
+  });
+}
+
+function runConnect() {
+  const { connect } = require('../src/connect');
+  connect();
+}
+
+function runUninstall() {
+  const { uninstall } = require('../src/uninstall');
+  uninstall(process.argv.slice(3));
+}
+
+function runAgent() {
+  const args = process.argv.slice(3);
+  if (args.length === 0) {
+    console.error('❌ Usage: venombrowser agent "<task description>"');
+    process.exit(1);
+  }
+  const child = fork(path.join(BIN_DIR, 'venombrowser-agent.js'), args, {
+    stdio: 'inherit'
+  });
+  child.on('exit', (code) => process.exit(code || 0));
+}
+
+function runChat() {
+  const child = fork(path.join(BIN_DIR, 'venombrowser-chat.js'), process.argv.slice(3), {
+    stdio: 'inherit'
+  });
+  child.on('exit', (code) => process.exit(code || 0));
+}
+
+// ─── Dispatch ───────────────────────────────────────────────────────────────
+
+const command = process.argv[2];
+
+switch (command) {
+  case 'setup':
+    runSetup();
+    break;
+  case 'start':
+    runStart();
+    break;
+  case 'status':
+    runStatus();
+    break;
+  case 'connect':
+    runConnect();
+    break;
+  case 'uninstall':
+    runUninstall();
+    break;
+  case 'agent':
+    runAgent();
+    break;
+  case 'chat':
+    runChat();
+    break;
+  case 'help':
+  case '--help':
+  case '-h':
+    showHelp();
+    break;
+  case undefined:
+    showHelp();
+    break;
+  default:
+    console.error(`❌ Unknown command: ${command}`);
+    console.error('   Run "venombrowser help" for available commands.');
+    process.exit(1);
+}