feat: support opening local files

.github/workflows/ci.yml

@@ -0,0 +1,34 @@
+name: CI
+
+on:
+  push:
+  pull_request:
+
+jobs:
+  rust:
+    name: Rust (${{ matrix.os }})
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-latest, windows-latest]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: dtolnay/rust-toolchain@stable
+        with:
+          components: rustfmt, clippy
+      - run: cargo fmt --all -- --check
+      - run: cargo clippy --all-targets --all-features
+      - run: cargo test --all-targets --all-features
+
+  extension:
+    name: Chrome extension syntax
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 24
+      - run: node --check assets/tmwd_cdp_bridge/background.js
+      - run: node --check assets/tmwd_cdp_bridge/content.js
+      - run: node --check assets/tmwd_cdp_bridge/popup.js

Cargo.toml

@@ -12,13 +12,16 @@ path = "src/main.rs"
 anyhow = "1.0"
 axum = { version = "0.7", features = ["json", "ws"] }
 clap = { version = "4.5", features = ["derive"] }
+dirs = "6.0"
+dunce = "1.0"
 fs2 = "0.4"
 html5ever = "0.27"
 libc = "0.2"
 markup5ever_rcdom = "0.3"
+path-clean = "1.0"
 reqwest = { version = "0.12", default-features = false, features = ["blocking", "json", "rustls-tls"] }
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"
 tokio = { version = "1.48", features = ["macros", "process", "rt-multi-thread", "signal", "sync", "time"] }
-tower-http = { version = "0.6", features = ["cors"] }
+url = "2.5"
 uuid = { version = "1.18", features = ["v4"] }

assets/tmwd_cdp_bridge/background.js

@@ -20,6 +20,7 @@ const DEFAULT_WS_PORT = 18765;
 const CLI_API_PORT = 18767;
 let wsPort = DEFAULT_WS_PORT;
 const browserId = `browser-${crypto.randomUUID()}`;
+const extensionVersion = chrome.runtime.getManifest().version;
 let profileId = null;
 let profileLabel = null;
 
@@ -40,7 +41,7 @@ function withClientIdentity(payload) {
 }
 
 async function handleExtMessage(msg, sender) {
-  if (msg.cmd === 'status') return handleStatus();
+  if (msg.cmd === 'status') return await handleStatus();
   if (msg.cmd === 'setPort') return await handleSetPort(msg);
   if (msg.cmd === 'setProfileLabel') return await handleSetProfileLabel(msg);
   lastCommandAt = Date.now();
@@ -67,7 +68,7 @@ async function handleExtMessage(msg, sender) {
         if (msg.allowFocus === true && tab.windowId) await chrome.windows.update(tab.windowId, { focused: true });
         return { ok: true };
       } else {
-        const tabs = (await chrome.tabs.query({})).filter(t => isScriptable(t.url));
+        const tabs = await queryScriptableTabs();
         const data = tabs.map(t => ({ id: t.id, url: t.url, title: t.title, active: t.active, windowId: t.windowId }));
         return { ok: true, data };
       }
@@ -109,7 +110,7 @@ async function handleExtMessage(msg, sender) {
   return { ok: false, error: 'Unknown cmd: ' + msg.cmd };
 }
 
-function handleStatus() {
+async function handleStatus() {
   return {
     ok: true,
     data: {
@@ -119,7 +120,10 @@ function handleStatus() {
       lastCommandAt,
       browserId,
       profileId,
-      profileLabel
+      profileLabel,
+      extensionId: chrome.runtime.id,
+      extensionVersion,
+      fileSchemeAccess: await chrome.extension.isAllowedFileSchemeAccess()
     }
   };
 }
@@ -645,7 +649,7 @@ async function handleBatch(msg, sender) {
       if (c.cmd === 'cookies') {
         R.push(await handleCookies(c, sender));
       } else if (c.cmd === 'tabs') {
-        const tabs = (await chrome.tabs.query({})).filter(t => isScriptable(t.url));
+        const tabs = await queryScriptableTabs();
         R.push({ ok: true, data: tabs.map(t => ({ id: t.id, url: t.url, title: t.title, active: t.active, windowId: t.windowId })) });
       } else if (c.cmd === 'cdp') {
         const tabId = c.tabId || msg.tabId || sender.tab?.id;
@@ -687,11 +691,16 @@ async function handleCDP(msg, sender) {
     return { ok: false, error: e.message };
   }
 }
-// Filter out chrome:// and other internal tabs that can't be scripted
-const isScriptable = url => url && /^https?:/.test(url);
+// Filter out chrome:// and other internal tabs that can't be scripted.
+const isScriptable = (url, fileSchemeAccess = false) => !!url && (/^https?:/i.test(url) || (fileSchemeAccess && /^file:/i.test(url)));
+
+async function queryScriptableTabs() {
+  const fileSchemeAccess = await chrome.extension.isAllowedFileSchemeAccess();
+  return (await chrome.tabs.query({})).filter(tab => isScriptable(tab.url, fileSchemeAccess));
+}
 
 async function injectContentScriptsIntoExistingTabs() {
-  const tabs = (await chrome.tabs.query({})).filter(t => isScriptable(t.url));
+  const tabs = await queryScriptableTabs();
   for (const tab of tabs) {
     try {
       await chrome.scripting.executeScript({
@@ -1029,9 +1038,11 @@ async function connectWS() {
     console.log('[TMWD-WS] Connected!');
     scheduleKeepalive(); // Keep SW alive while connected
     await loadClientIdentity();
-    const tabs = (await chrome.tabs.query({})).filter(t => isScriptable(t.url));
+    const tabs = await queryScriptableTabs();
     ws.send(JSON.stringify(withClientIdentity({
       type: 'ext_ready',
+      extension_version: extensionVersion,
+      file_scheme_access: await chrome.extension.isAllowedFileSchemeAccess(),
       tabs: tabs.map(t => ({ id: t.id, url: t.url, title: t.title }))
     })));
     console.log('[TMWD-WS] Sent ext_ready with', tabs.length, 'tabs');
@@ -1103,10 +1114,12 @@ chrome.runtime.onInstalled.addListener(() => {
 // Sync tab list on changes
 async function sendTabsUpdate() {
   if (!ws || ws.readyState !== WebSocket.OPEN) return;
-  const tabs = (await chrome.tabs.query({})).filter(t => isScriptable(t.url) && !/streamlit/i.test(t.title));
+  const tabs = (await queryScriptableTabs()).filter(t => !/streamlit/i.test(t.title));
   await loadClientIdentity();
   ws.send(JSON.stringify(withClientIdentity({
     type: 'tabs_update',
+    extension_version: extensionVersion,
+    file_scheme_access: await chrome.extension.isAllowedFileSchemeAccess(),
     tabs: tabs.map(t => ({ id: t.id, url: t.url, title: t.title }))
   })));
 }

assets/tmwd_cdp_bridge/manifest.json

@@ -1,7 +1,7 @@
 {
   "manifest_version": 3,
   "name": "Agent Browser CLI Bridge",
-  "version": "2.0",
+  "version": "2.1",
   "description": "Browser control bridge for agent-browser-cli",
   "permissions": [
     "cookies",

assets/tmwd_cdp_bridge/popup.html

@@ -21,6 +21,7 @@
 <h3>🔌 Bridge</h3>
 <div class="panel">
   <div id="bridgeStatus" class="status">读取连接状态...</div>
+  <div id="fileAccessStatus" class="status">读取文件网址权限...</div>
   <div class="row">
     <label>端口 <input id="port" type="number" min="1" max="65535"></label>
     <button id="savePort">保存并重连</button>

assets/tmwd_cdp_bridge/popup.js

@@ -20,10 +20,16 @@ async function refreshBridgeStatus() {
     const data = resp.data || {};
     portInput.value = data.wsPort || 18765;
     status.textContent = `状态: ${data.wsConnected ? '已连接' : '未连接'} ${data.wsUrl || ''}`;
+    const fileAccessStatus = document.getElementById('fileAccessStatus');
+    fileAccessStatus.textContent = `文件网址访问: ${data.fileSchemeAccess ? '已启用' : '未启用'}`;
+    fileAccessStatus.className = data.fileSchemeAccess ? 'status' : 'error';
     renderProfileStatus(data);
   } catch (e) {
     status.textContent = '状态读取失败: ' + e.message;
     status.className = 'error';
+    const fileAccessStatus = document.getElementById('fileAccessStatus');
+    fileAccessStatus.textContent = '文件网址权限读取失败: ' + e.message;
+    fileAccessStatus.className = 'error';
     const profileStatus = document.getElementById('profileStatus');
     profileStatus.textContent = 'Profile 读取失败: ' + e.message;
     profileStatus.className = 'error';

skills/agent-browser-cli/SKILL.md

@@ -30,6 +30,30 @@ agent-browser-cli logs --tail 100
 
 补充一个容易误判的点：`status` / `doctor` 里看到 `daemon_not_running`，如果此时还没有执行 `tabs` / `open` / `exec` / `scan`，通常只是 daemon 按需常驻而未启动，不代表故障。只有目标命令已经失败，或者日志/输出明确提示端口、扩展、标签页异常时，才进入排障。
 
+## 打开本地文件
+
+`open` 支持 HTTP(S) URL、绝对 `file://` URL 和本机文件路径。相对路径以执行 CLI 时的当前目录为基准；支持 `~`、中文、空格和符号链接，符号链接会解析到真实文件。所有普通文件类型均可打开，目录会被拒绝。
+
+```bash
+agent-browser-cli open ./demo.html
+agent-browser-cli open ~/Documents/report.pdf
+agent-browser-cli open 'file:///Users/me/My%20Files/demo.html#intro'
+agent-browser-cli open ./demo.html --timeout 15
+```
+
+本地文件要求 Chrome 扩展版本至少为 2.1，并在 `chrome://extensions` → Agent Browser CLI Bridge → 详情中开启“允许访问文件网址”。权限关闭时 `open` 会在创建标签前失败；`status` / `doctor` 会按 Profile 报告 `file_scheme_access`，但该可选能力不影响普通 HTTP(S) 健康状态。
+
+输入分类规则：
+
+- `./`、`../`、`~/`、绝对路径和显式 `file://` 表示本地文件；文件必须存在且是普通文件。
+- 裸输入若对应当前目录中已存在的文件，文件优先；否则按 Web 地址处理。要强制打开网站，显式写 `https://`。
+- 普通路径中的 `?` 和 `#` 是文件名字符；仅显式 `file://` URL 使用 query/fragment 语义。
+- `localhost`、回环/私网 IP、单标签主机和 `.local` 的无协议地址默认使用 HTTP；其他域名默认 HTTPS；端口 80/443 分别强制 HTTP/HTTPS。
+- 只支持 `http:`、`https:` 和 `file:`；其他显式 scheme 会返回 `unsupported_scheme`。
+- 异平台路径会明确报错，例如 macOS/Linux 不会把 `C:\\Users\\...` 错当作网址。
+
+本地文件 `open` 默认最多等待 10 秒，直到新标签成为可控制会话；可用 `--timeout` 调整。超时后标签会保留，错误结果包含稳定的 `error_code` 和 `opened_tab_id`。活动 `open` 成功后新标签成为默认会话，`--background` 不切换默认会话。
+
 ## 常用命令优先级
 
 先区分三个入口：