dsh-builtin-browser 0.1.14 → 0.1.16

package/README.en.md

@@ -18,7 +18,7 @@
 | --- | --- |
 | Why a shared real browser, and how it differs from headless approaches | [Why a shared real browser](docs/why-browser.md) |
 | Installation, configuration, day-to-day use | [User guide](docs/user-guide.md) |
-| All 20 tools: parameters, output, examples | [Tool reference](docs/tool-reference.md) |
+| All 33 tools: parameters, output, examples | [Tool reference](docs/tool-reference.md) |
 | How the seam / provider / tools layers and self-hosting work | [Architecture](docs/architecture.md) |
 | Documentation index and README split | [Docs index](docs/README.md) |
 
@@ -28,7 +28,7 @@
 
 - **A real view, not a relay**: the browser is a native `WebContentsView`; the human sees every step the agent takes and can grab control at any time;
 - **Install-and-use**: with a desktop shell the shell's embedded view is used; on plain `dsh web` the plugin **self-hosts** — it spawns its own Electron window with zero extra configuration;
-- **One plugin, one toolset**: after install the agent automatically gets 20 `browser_*` tools (open, inspect, interact, fill forms, screenshot, download, auth management…).
+- **One plugin, one toolset**: after install the agent automatically gets 33 `browser_*` tools (open, a11y tree, wait, semantic/coordinate interaction, scroll, back/forward, batch and single-control form filling, keys, structured scraping, screenshot, download, auth management…).
 
 In one sentence: **installing the plugin gives you a real browser that is shared with the user and drivable by the agent.**
 
@@ -131,13 +131,26 @@ See the full list in [Tool reference](#tool-reference).
 | Tool | Purpose | Guard |
 | --- | --- | --- |
 | `browser_open` | Open a URL (optionally in a new tab); returns a page snapshot | ✅ |
-| `browser_snapshot` | Numbered inventory of interactive elements (inputs/buttons/links) | – |
+| `browser_wait` | Wait for page load (optional expected URL / CSS selector), returns readiness | – |
+| `browser_snapshot` | Numbered inventory of interactive elements (inputs/buttons/links; pierces same-origin iframes and Shadow DOM) | – |
+| `browser_a11y` | Accessibility tree: semantic role/name/value/states + coordinates per interactive node (pierces same-origin iframes and Shadow DOM) | – |
 | `browser_execute` | Run JS in the page; args arrive as `arguments[0..n]` | ✅ |
 | `browser_content` | Fetch the page as html / markdown / txt / json (selector, maxChars, timeoutMs) | – |
-| `browser_click` | Click at viewport coordinates (for vision-located targets) | ✅ |
-| `browser_type` | Type text into the focused element (CDP `Input.insertText`) | ✅ |
+| `browser_click` | Click a semantic target (`target`: css/text/xpath, scrolls into view and clicks center) or viewport coordinates (vision-located) | ✅ |
+| `browser_type` | Type text (optionally focusing a `target` element first; CDP `Input.insertText`) | ✅ |
+| `browser_key` | Press a named key (Enter/Tab/arrows/Home/End…) | ✅ |
+| `browser_scroll` | Scroll the page (pixel deltas / selector / top-bottom) | ✅ |
+| `browser_back` | One step back in page history (no-op at the start) | ✅ |
+| `browser_forward` | One step forward in page history (no-op at the end) | ✅ |
+| `browser_refresh` | Reload the current page (like a browser refresh button) | ✅ |
 | `browser_fill` | Batch form fill (selector/name/label matching, controlled inputs, selects, checkbox/radio, optional submit) | ✅ |
-| `browser_screenshot` | PNG capture, optional `fullPage` and `savePath` | – |
+| `browser_set_value` | Set one control's value (`target`-located; native setter + input/change, React-controlled friendly) | ✅ |
+| `browser_check` | Check/uncheck a checkbox or radio (`target`-located) | ✅ |
+| `browser_select` | Select an option of a `<select>` by value/text/index (`target`-located) | ✅ |
+| `browser_clear` | Clear an input/textarea/contenteditable, or uncheck (`target`-located) | ✅ |
+| `browser_get_value` | Read an element's current value for verification (`target`-located) | – |
+| `browser_scrape` | Structured extraction: container selector + field map (`selector@attr`), static CSS only, CSP-safe | – |
+| `browser_screenshot` | Capture, optional `fullPage`, `savePath`, JPEG (`format`/`quality`) and scaling (`maxWidth`/`maxHeight`) | – |
 | `browser_list_tabs` | List the session's tabs | – |
 | `browser_switch_tab` | Switch to a tab by id (also switches the visible view when self-hosted) | ✅ |
 | `browser_close_tab` | Close a tab by id; closing the active tab activates the next | – |
@@ -146,13 +159,24 @@ See the full list in [Tool reference](#tool-reference).
 | `browser_reset_session` | Close and rebuild this task's browser session | ✅ |
 | `browser_history` | Operation log (newest last), with per-step success/error and result summary | – |
 | `browser_replay` | Replay one step by sequence number (navigate/execute/click/type) | ✅ |
-| `browser_download` | Download a URL with session cookies to a local file (256 MB cap) | ✅ |
+| `browser_download` | Download an HTTP(S) URL with session cookies to a local file (absolute `savePath`, 256 MB cap) | ✅ |
 | `browser_auth` | Export/restore cookies (login persistence, self-hosted) | ✅ |
 | `browser_challenge` | Detect a human-verification challenge (CAPTCHA / Cloudflare / reCAPTCHA / hCaptcha / Turnstile) | – |
-| `browser_restrict` | Restrict allowed browser actions (allow-list; empty list lifts it) | – |
+| `browser_restrict` | Restrict allowed browser actions (allow-list; empty list lifts it). **Soft guardrail** — the model can lift it itself; not a security boundary | – |
 
 > "Guard" column: ✅ actions are governed by the `browser_restrict` allow-list; read-only tools (snapshot/content/screenshot/list_tabs/session/challenge/history) are never blocked.
 
+### Waiting for the page
+
+- **After `browser_open`, before `browser_snapshot`, call `browser_wait` on slow sites**: pass `url` (what you opened) and an optional `selector`, and wait for `ready: true` — otherwise you snapshot the old document or a white screen.
+- **Content you cannot see may live in an iframe / Shadow DOM**: snapshots and the a11y tree pierce same-origin iframes and shadow roots and mark them `(iframe)`; coordinates are always top-document, so `browser_click` works directly. DOM selectors are frame-scoped — reach them via `iframe.contentDocument` in `browser_execute`.
+
+### Semantic targets and the a11y tree
+
+- **`browser_a11y` is the best way to understand a page**: every interactive node carries its semantic role (button/textbox/checkbox…), accessible name, current value, states (enabled/checked/expanded…) and coordinates — click/type them directly.
+- **`browser_click`/`browser_type` accept a `target`**: `{by: css|text|xpath, value, index?}` — `text` matches an element's own visible text (exact first, then contains, deepest preferred); clicks scroll the element to the viewport center first; typing focuses it first.
+- **Use the single-control tools for one field** (`browser_set_value`/`browser_check`/`browser_select`/`browser_clear`/`browser_get_value`), `browser_fill` for batches, and `browser_scrape` for structured list extraction.
+
 ### Operating discipline (click/fill)
 
 - **Prefer DOM semantics over coordinates**: submit forms with `form.requestSubmit()`; click with `element.click()`; coordinate clicks are the last resort.
@@ -171,6 +195,7 @@ The plugin mounts through `cordis.patch.yml` (three rows); per-row config:
 | `browser-electron` | `httpOnly` | boolean | `true` | Allow HTTP(S) navigation only; other protocols (e.g. `file:`/`data:`) rejected (`BROWSER_NAVIGATION_BLOCKED`) |
 | `browser-electron` | `snapshotMaxElements` | number | `60` | Max snapshot elements before truncation |
 | `browser-electron` | `contentMaxChars` | number | `100000` | Default content character cap |
+| `browser-electron` | `downloadDir` | string | `~/Downloads` | Confine `browser_download` save paths to this directory (stops a prompt-injected agent writing arbitrary paths); defaults to the OS Downloads folder, override for a sandbox dir |
 | `tool-browser` | `timeoutMs` | number | `60000` | Cooperative tool timeout (ms) |
 | `tool-browser` | `tabTools` | boolean | `true` | Register tab-management tools (`browser_list_tabs` etc.) |
 
@@ -186,9 +211,11 @@ agent (browser_* tools)
 
 - **Seam** (`browser` row): provides the `ctx.browser` service — provider registration, session lifecycle, error codes — decoupled from any implementation.
 - **Provider** (`browser-electron` row): operates views through the `ElectronBrowserViewHost` seam (create/destroy/show, `sendCommand`), implemented with real Electron objects by the shell.
-- **Tools** (`tool-browser` row): the 20 model-facing `browser_*` tools, maintaining one browser session per calling task (DSH session).
+- **Tools** (`tool-browser` row): the 33 model-facing `browser_*` tools, maintaining one browser session per calling task (DSH session).
+
+**Self-hosted mode**: without a desktop shell, the plugin spawns its own Electron child process (`host-main.js`) and drives it over loopback TCP JSON-RPC. The RPC is authenticated with a random per-spawn token delivered over **both stdin and an environment variable** — on Windows the Electron GUI process never receives piped stdin, so the env fallback keeps the handshake reliable. The child auto-restarts after a crash; screenshots prefer Electron's native `capturePage` (CDP capture can hang with multiple views in the window); the plugin automatically picks the **newest** Electron in the environment (33.x has a compositor defect; ≥ 40 recommended; 44+ downloads its binary on first use, needs network).
 
-**Self-hosted mode**: without a desktop shell, the plugin spawns its own Electron child process (`host-main.js`) and drives it over loopback TCP JSON-RPC (window title `dsh-browser`). The child auto-restarts after a crash; screenshots prefer Electron's native `capturePage` (CDP capture can hang with multiple views in the window); the plugin automatically picks the **newest** Electron in the environment (33.x has a compositor defect; ≥ 40 recommended).
+**The self-hosted browser IS a real browser**: every task (DSH session) gets its **own browser window** with a full toolbar — address bar, back/forward/reload buttons, and a tab strip (new/switch/close tabs). A human can use it exactly like Chrome: type a URL in the address bar (https:// is added automatically), click tabs, open new ones. Keyboard focus follows your clicks — **click the address bar to type, click the page to interact** (Windows focus routing; fixes the case where clicks did not move focus and the address bar could not receive typed URLs). Human and agent actions feed the **same session model** (same tabs, history, and navigation); the window title always shows the task label plus the page title/URL, and views follow the window size on resize. A window closes automatically with its session when the task ends.
 
 **Electron lookup order**: ① `require('electron')` (peer dependency) → ② `ELECTRON_PATH` (explicit override) → ③ the newest among DSH install anchors and pnpm virtual stores. A clear error tells you when none is found.
 
@@ -205,35 +232,39 @@ The browser's **visible view**, the **browser column layout**, and the **column-
 
 | Component | Version |
 | --- | --- |
-| DeepSeek Harness (dsh) | `0.1.0-rc.5` |
-| Electron | `43.4.0` (≥ 40 recommended; 33.x has a compositor defect) |
+| DeepSeek Harness (dsh) | `0.1.1-rc.2` (peer range `^0.1.1-rc.2`) |
+| Electron | `44.0.0` (≥ 40 recommended; 33.x has a compositor defect) |
 | Node.js | `22.20.0` |
-| dsh-builtin-browser | `0.1.11` |
+| dsh-builtin-browser | `0.1.16` |
 | OS | Windows 10 (10.0.26200) |
 
 > The plugin declares `electron >= 30`; it has **only been verified on Windows** (macOS/Linux untested, not yet promised).
 
 ## Known limitations
 
-- Screenshots are PNG-only (CDP JPEG hangs on Electron 43); JPEG awaits a non-CDP conversion path.
+- JPEG screenshots are available only on the self-hosted native path (`capturePage` `toJPEG`); the desktop shell's CDP fallback stays PNG (CDP JPEG hangs on Electron 43).
 - Self-hosted captures prefer Electron's native `capturePage` (CDP `captureScreenshot` can hang with multiple views in the window); the target tab is raised before capturing.
 - `fullPage` capture is flaky under software compositing on some hosts.
 - CAPTCHA cannot be solved automatically: snapshots flag detected challenges; ask the human to complete it in the shared window instead of retrying.
 - Private mode (`privateMode`) is not implemented: it needs Electron session partitioning, which is host-layer territory; this plugin does not promise it.
-- `browser_download` fetches in the page context (keeps logins) and is subject to same-origin/CORS constraints; single files are capped at 256 MB.
+- `browser_download` fetches in the page context (keeps logins) and is subject to same-origin/CORS constraints; HTTP(S) targets only; `savePath` must be absolute (default confined to `~/Downloads`, override with `downloadDir`); single files are capped at 256 MB (streamed with a Content-Length early reject) and are written by the browser child itself (temp file + atomic rename).
+- The self-hosted browser's cookies are stored in plaintext on disk (Electron default); deployments that need encrypted-at-rest should integrate a system keychain / DPAPI at the host layer.
+- `browser_restrict` is a **soft guardrail** against accidental actions, not a security boundary: the model can lift it itself.
+- Popups (`window.open` / `target=_blank`) are re-routed into the current tab instead of opening untracked native windows, so the tab/session model stays intact.
 - The `browser_auth` cookie round-trip does not preserve `hostOnly`/`sameSite` (host-only cookies come back as domain cookies); it is available on the self-hosted browser only.
 - After a self-hosted child crash the browser host restarts automatically, but sessions opened before the crash are gone — call `browser_reset_session` to rebuild.
+- Electron 44+ downloads its binary (~100 MB) on the first window open and needs network; it is never needed again afterwards. If detection runs without network, pre-install a binary and point `ELECTRON_PATH` at it.
 - This plugin contains no browser-column UI — that is host-shell territory; do not treat "browser column" as a plugin feature.
 
 ## Development
 
 ```sh
 # Type-check + build (lib/)
-pnpm run build
-
-# Functional tests: start a local page server + Electron probe (see repo test scripts)
+npm run build
 ```
 
+> Run tests: `npm test` (= `tsc -p tsconfig.json` + `node --test "tests/*.test.mjs"`; fake-host tests, no Electron needed).
+
 Code layout:
 
 | Directory | Responsibility |
@@ -243,6 +274,21 @@ Code layout:
 | `src/tool-browser/` | Model-facing `browser_*` tools |
 | `src/types/` | Electron ambient types (shim; no hard electron type dependency) |
 
+## Update history
+
+> Round-by-round development and fixes (full detail in [CHANGELOG.md](CHANGELOG.md)). Published as of **0.1.16** (tag `v0.1.16`).
+
+| Round | Date | Content |
+| --- | --- | --- |
+| 1 | 2026-08-18 | **Security & robustness**: random-token RPC auth + single connection; download admission (HTTP(S) only, absolute path, `downloadDir`-confined) with streamed caps (Content-Length early reject, 256 MB max); CDP timeout interrupts and click/type timeout key-release recovery; per-task sessions/allow-lists with agent-lifecycle auto-close; history redaction (typed text, replay/execute args not leaked); popup re-routing into the tab |
+| 2 | 2026-08 | **Feature completion + tests + CI**: window title shows the task; flicker-free showView; snapshots/a11y pierce same-origin iframes & Shadow DOM; new `browser_wait`/`scroll`/`back`/`forward`/`key` tools; real `available()` probe; child-side downloads (temp file + atomic rename); constrained Electron lookup; JPEG/scaled screenshots; snapshot perf; test suite + CI |
+| 3 | 2026-08 | **browser-bridge parity + review fixes**: `browser_a11y` a11y tree; 6 form-control tools (`browser_set_value`/`check`/`select`/`clear`/`get_value`/`refresh`); semantic `target` (css/text/xpath); `browser_scrape` structured extraction; independent BrowserWindow + real toolbar (address bar, back/forward/reload, tab strip) routed back into the session model; tool count **20 → 33**; CI switched to npm (no lockfile → pnpm cache broken), README corrections |
+| 4 | 2026-08 | **DSH 0.1.1-rc.2 alignment + review fixes**: peer floor `^0.1.1-rc.2`; fixed `browser_type` dropping text with a target, `browser_key` Space missing CDP `text`, keyUp failure sticking a key, `browser_wait` same-origin URL mis-match, `.part` rename residue, `snapshotMaxElements`/`contentMaxChars` config wiring, missing type exports; 3 regression tests |
+| 5 | 2026-08 | **Electron 44 compatibility**: `available()` is now side-effect free (no more triggering Electron 44 lazy download); `flushAuth` cookie-domain build fix |
+| 6 | 2026-08 | **Windows handshake & tab lookup**: the Electron GUI process never receives piped stdin → RPC token now flows over **stdin + env var**; `browser_switch_tab`/`browser_close_tab` locate tabs across sessions (`locateTab`), `browser_close_tab` no longer fakes success, unknown ids error with the session's actual tab list |
+| 7 | 2026-08 | **Toolbar interaction (Windows focus routing)**: keyboard input only reaches the focused view and the page view grabbed it, so the address bar could not receive input → added `wireFocusRouting` (clicking a view focuses it) + window refocus restores the last-clicked view; verified with real OS input probes |
+| **0.1.16** | 2026-08-26 | **Release**: all seven rounds ship as **0.1.16** (build clean, 21/21 tests pass, `v0.1.16`) |
+
 ## Acknowledgements
 
 Special thanks to the [DeepSeek Harness repository](https://github.com/deepseek-ai/deepseek-harness) and the DeepSeek AI team: the seam, the tool runtime, and the plugin system this plugin builds on all come from that project.

package/README.md

@@ -18,7 +18,7 @@
 | --- | --- |
 | 了解插件为什么存在、与无头方案的区别 | [为什么做共享真实浏览器](docs/why-browser.md) |
 | 安装、配置与日常使用 | [用户指南](docs/user-guide.md) |
-| 全部 20 个工具的参数、输出与示例 | [工具参考](docs/tool-reference.md) |
+| 全部 33 个工具的参数、输出与示例 | [工具参考](docs/tool-reference.md) |
 | 了解 seam / provider / 工具三层与自托管实现 | [架构说明](docs/architecture.md) |
 | 查看全部文档与 README 分工 | [文档索引](docs/README.md) |
 
@@ -28,7 +28,7 @@
 
 - **真实视图,而非转播**:浏览器是原生 `WebContentsView`,用户直接看到 agent 在做什么,随时可以上手接管;
 - **装好即用**:有桌面外壳时嵌入外壳视图;纯 `dsh web` 也能**自托管**——插件自己拉起一个 Electron 窗口,不需要任何额外配置;
-- **一插件即一套工具**:安装后 agent 自动获得 20 个 `browser_*` 工具(打开、查看、操作、填表、截图、下载、登录态管理……)。
+- **一插件即一套工具**:安装后 agent 自动获得 33 个 `browser_*` 工具(打开、查看、无障碍树、等待、语义/坐标操作、滚动、回退、批量/单控件填表、按键、结构化提取、截图、下载、登录态管理……)。
 
 一句话:**安装插件 = 获得一个与用户共享、可被 agent 驱动的真实浏览器。**
 
@@ -131,13 +131,26 @@ dsh plugin --profile web add <本仓库路径>
 | 工具 | 用途 | 守卫 |
 | --- | --- | --- |
 | `browser_open` | 打开 URL(可选新标签),返回页面快照 | ✅ |
-| `browser_snapshot` | 交互元素(输入框/按钮/链接)带编号清单 | – |
+| `browser_wait` | 等待页面加载完成(可选期望 URL / CSS 选择器),返回是否就绪 | – |
+| `browser_snapshot` | 交互元素(输入框/按钮/链接)带编号清单(穿透同源 iframe 与 Shadow DOM) | – |
+| `browser_a11y` | 无障碍树:每个交互节点的语义角色/名称/值/状态 + 坐标(穿透同源 iframe 与 Shadow DOM) | – |
 | `browser_execute` | 在页面执行 JS;参数以 `arguments[0..n]` 传入 | ✅ |
 | `browser_content` | 以 html / markdown / txt / json 抓取页面(selector、maxChars、timeoutMs) | – |
-| `browser_click` | 按视口坐标点击(配合截图做视觉定位) | ✅ |
-| `browser_type` | 向聚焦元素输入文本(CDP `Input.insertText`) | ✅ |
+| `browser_click` | 点击:语义目标(`target`: css/text/xpath,滚动到元素并点中心)或视口坐标(配合截图视觉定位) | ✅ |
+| `browser_type` | 输入文本(可先按 `target` 聚焦元素;CDP `Input.insertText`) | ✅ |
+| `browser_key` | 按命名按键(Enter/Tab/方向键/Home/End 等) | ✅ |
+| `browser_scroll` | 滚动页面(像素增量 / 选择器定位 / 顶部底部) | ✅ |
+| `browser_back` | 页面历史后退一步(无前项时为空操作) | ✅ |
+| `browser_forward` | 页面历史前进一步(无后项时为空操作) | ✅ |
+| `browser_refresh` | 刷新当前页(等价浏览器的刷新按钮) | ✅ |
 | `browser_fill` | 批量填充表单(选择器/名称/标签匹配,受控输入、下拉、单选/复选,可选提交) | ✅ |
-| `browser_screenshot` | PNG 截图,可选 `fullPage` 与 `savePath` | – |
+| `browser_set_value` | 单个控件设值(按 `target` 定位;原生 setter + input/change,React 受控输入可用) | ✅ |
+| `browser_check` | 勾选/取消勾选 checkbox 或 radio(按 `target` 定位) | ✅ |
+| `browser_select` | 选中 `<select>` 的某个选项(按值/文本/索引,按 `target` 定位) | ✅ |
+| `browser_clear` | 清空输入/文本域/contenteditable,或取消勾选(按 `target` 定位) | ✅ |
+| `browser_get_value` | 读取元素当前值(操作后验证用;按 `target` 定位) | – |
+| `browser_scrape` | 结构化提取:容器选择器 + 字段映射(`选择器[@属性]`),静态 CSS 查询、CSP 安全 | – |
+| `browser_screenshot` | 截图,可选 `fullPage`、`savePath`、JPEG(`format`/`quality`)与缩放(`maxWidth`/`maxHeight`) | – |
 | `browser_list_tabs` | 当前会话的标签列表 | – |
 | `browser_switch_tab` | 按 id 切换标签(自托管下同步切换可见视图) | ✅ |
 | `browser_close_tab` | 按 id 关闭标签;关闭活动标签后激活下一个 | – |
@@ -146,13 +159,24 @@ dsh plugin --profile web add <本仓库路径>
 | `browser_reset_session` | 关闭并重建本任务的浏览器会话 | ✅ |
 | `browser_history` | 操作日志(最新在后),含成功/失败与结果摘要 | – |
 | `browser_replay` | 按序号回放某一步(navigate/execute/click/type) | ✅ |
-| `browser_download` | 带会话 cookie 下载 URL 到本地文件(上限 256MB) | ✅ |
+| `browser_download` | 带会话 cookie 下载 HTTP(S) URL 到本地文件(`savePath` 必须绝对路径,上限 256MB) | ✅ |
 | `browser_auth` | 导出/恢复 cookie(登录态持久化,自托管可用) | ✅ |
 | `browser_challenge` | 检测人机验证(CAPTCHA / Cloudflare / reCAPTCHA / hCaptcha / Turnstile) | – |
-| `browser_restrict` | 限制允许的浏览器动作(白名单;空列表解除) | – |
+| `browser_restrict` | 限制允许的浏览器动作(白名单;空列表解除)。**软护栏**,模型可自行解除,非安全边界 | – |
 
 > 「守卫」列:打 ✅ 的动作受 `browser_restrict` 白名单约束;只读工具(snapshot/content/screenshot/list_tabs/session/challenge/history)永不拦截。
 
+### 等待页面就绪
+
+- **`browser_open` 之后、`browser_snapshot` 之前,慢站点请先 `browser_wait`**:传 `url`(你打开的地址)与可选的 `selector`,等它返回 `ready: true` 再拍照——否则拍到的是旧页面或白屏。
+- 页面里看不到的内容先想 iframe / Shadow DOM:快照与无障碍树会穿透同源 iframe 与 shadow root 并标注 `(iframe)`,坐标始终是顶层文档坐标,可直接用 `browser_click`;DOM 选择器则是 frame 作用域的,需用 `browser_execute` 经 `iframe.contentDocument` 访问。
+
+### 语义定位(`target`)与无障碍树
+
+- **`browser_a11y` 是理解页面的首选**:它返回每个交互节点的语义角色(button/textbox/checkbox…)、可访问名称、当前值、状态(enabled/checked/expanded…)与坐标,比编号快照更能说明“这是什么、能做什么”;拿到坐标后可直接 `browser_click`/`browser_type`。
+- **`browser_click`/`browser_type` 支持 `target` 定位**:`{by: css|text|xpath, value, index?}`——`text` 按元素自身可见文本匹配(精确优先、退化包含、最深元素优先);点击会把元素滚动到视口中央再点;输入会先聚焦该元素。
+- **单控件操作用 `browser_set_value`/`browser_check`/`browser_select`/`browser_clear`/`browser_get_value`**,批量用 `browser_fill`,列表页结构化抓取用 `browser_scrape`。
+
 ### 操作纪律(点击/填表)
 
 - **优先用 DOM 语义而非坐标**:表单提交优先 `form.requestSubmit()`;点击优先 `element.click()`;坐标点击是最后手段。
@@ -171,6 +195,7 @@ dsh plugin --profile web add <本仓库路径>
 | `browser-electron` | `httpOnly` | 布尔 | `true` | 仅允许 HTTP(S) 导航;其余协议(如 `file:`/`data:`)拒绝(`BROWSER_NAVIGATION_BLOCKED`) |
 | `browser-electron` | `snapshotMaxElements` | 数字 | `60` | 快照最多收录的交互元素数,超出截断 |
 | `browser-electron` | `contentMaxChars` | 数字 | `100000` | 内容抓取默认字符上限 |
+| `browser-electron` | `downloadDir` | 字符串 | `~/Downloads` | 限定 `browser_download` 保存路径必须位于该目录内(防 agent 写任意路径);默认收敛到系统下载目录,可改为沙箱目录 |
 | `tool-browser` | `timeoutMs` | 数字 | `60000` | 工具协作超时(ms) |
 | `tool-browser` | `tabTools` | 布尔 | `true` | 是否注册标签管理工具(`browser_list_tabs` 等) |
 
@@ -186,9 +211,11 @@ agent (browser_* 工具)
 
 - **seam 层**(`browser` 行)提供 `ctx.browser` 服务:provider 注册、会话生命周期、错误码,与具体实现解耦;
 - **provider 层**(`browser-electron` 行)通过 `ElectronBrowserViewHost` 接缝操作视图(创建/销毁/显示/`sendCommand`),由真实外壳用 Electron 对象实现;
-- **工具层**(`tool-browser` 行)提供模型侧的 20 个 `browser_*` 工具,按调用方任务(DSH 会话)维护独立的浏览器会话。
+- **工具层**(`tool-browser` 行)提供模型侧的 33 个 `browser_*` 工具,按调用方任务(DSH 会话)维护独立的浏览器会话。
+
+**自托管模式**:没有桌面外壳时,插件自己拉起一个 Electron 子进程(`host-main.js`),通过本机 TCP JSON-RPC 驱动。RPC 带随机 token 认证,token 经 **stdin + 环境变量双通道**传递——Windows 上 Electron 是 GUI 子系统进程、收不到 piped stdin,环境变量兜底保证握手稳定。子进程崩溃会自动重启;截图优先走 Electron 原生 `capturePage`(CDP 截图在多视图下会挂起),并自动选择环境中**最新版本**的 Electron(33.x 有合成器缺陷,建议 ≥ 40;44+ 首次使用自动下载二进制,需联网)。
 
-**自托管模式**:没有桌面外壳时,插件自己拉起一个 Electron 子进程(`host-main.js`),通过本机 TCP JSON-RPC 驱动,窗口标题 `dsh-browser`。子进程崩溃会自动重启;截图优先走 Electron 原生 `capturePage`(CDP 截图在多视图下会挂起),并自动选择环境中**最新版本**的 Electron(33.x 有合成器缺陷,建议 ≥ 40)。
+**自托管浏览器就是一台真正的浏览器**:每个任务(DSH 会话)拥有**独立的浏览器窗口**,窗口自带完整工具栏——地址栏、后退/前进/刷新按钮、标签条(新建/切换/关闭标签)。人可以直接像用 Chrome 一样使用它:在地址栏输入网址(自动补 `https://`)、点标签切换页面、开新标签;键盘焦点跟随点击——**点地址栏即可输入、点页面即可操作**(Windows 焦点路由,修复了点击不转移焦点导致地址栏无法输入的问题)。agent 与人的操作都汇入**同一个会话模型**(同一套标签、历史与导航),窗口标题实时显示当前任务标识与页面标题/URL,窗口缩放时视图自动跟随。任务结束后窗口随会话自动关闭。
 
 **Electron 定位顺序**:① `require('electron')`(peer 依赖)→ ② `ELECTRON_PATH`(显式覆盖)→ ③ DSH 安装锚点与 pnpm 虚拟仓库中**版本最新**者。找不到时工具会报清晰的错误提示。
 
@@ -205,35 +232,39 @@ agent (browser_* 工具)
 
 | 组件 | 版本 |
 | --- | --- |
-| DeepSeek Harness(dsh) | `0.1.0-rc.5` |
-| Electron | `43.4.0`(推荐 ≥ 40;33.x 存在合成器缺陷) |
+| DeepSeek Harness(dsh) | `0.1.1-rc.2`(peer 声明 `^0.1.1-rc.2`) |
+| Electron | `44.0.0`(推荐 ≥ 40;33.x 存在合成器缺陷) |
 | Node.js | `22.20.0` |
-| dsh-builtin-browser | `0.1.11` |
+| dsh-builtin-browser | `0.1.16` |
 | 操作系统 | Windows 10 (10.0.26200) |
 
 > 插件声明 `electron >= 30`;**当前仅在 Windows 环境实测**(macOS/Linux 未验证,暂不承诺)。
 
 ## 已知限制
 
-- 截图仅 PNG(CDP JPEG 在 Electron 43 上挂起);JPEG 等待非 CDP 转换路径。
+- JPEG 截图仅自托管原生路径可用(`capturePage` 的 `toJPEG`);桌面壳的 CDP 回退路径仍为 PNG(CDP JPEG 在 Electron 43 上挂起)。
 - 自托管截图优先走 Electron 原生 `capturePage`(CDP `captureScreenshot` 在多视图下会挂起);截图前自动把目标标签置顶。
 - 部分主机在软件合成下 `fullPage` 截图不稳定。
 - 人机验证(CAPTCHA)无法自动解决:快照会标注检测到的挑战,此时应请用户在共享窗口中人工完成,而不是反复重试。
 - 无痕模式(`privateMode`)未实现:它需要 Electron 的 session 分区能力,属于宿主层,本插件不承诺。
-- `browser_download` 在页面上下文内 `fetch`(带登录态),受同源/CORS 约束;单文件上限 256MB。
+- `browser_download` 在页面上下文内 `fetch`(带登录态),受同源/CORS 约束;仅允许 HTTP(S) 目标;`savePath` 必须为绝对路径(默认限定在 `~/Downloads`,可用 `downloadDir` 覆盖);单文件上限 256MB(流式限流,按 Content-Length 提前拒绝),文件由浏览器子进程直接落盘(临时文件 + 原子改名)。
+- 自托管浏览器的 cookie 在磁盘上以明文存储(Electron 默认行为);需要加密落盘的部署应在宿主层接入系统钥匙串 / DPAPI。
+- `browser_restrict` 是防误操作的**软护栏**,不是安全边界:模型可以自行解除白名单。
+- 页面弹窗(`window.open` / `target=_blank`)会被重定向到当前标签页内打开,不创建独立窗口,以免破坏标签/会话模型。
 - `browser_auth` 的 cookie 往返不保留 `hostOnly`/`sameSite` 字段(host-only cookie 恢复后变成 domain cookie);仅自托管浏览器可用。
 - 自托管浏览器子进程崩溃后会自动重启,但崩溃前已打开的会话视图已失效,调用 `browser_reset_session` 重建即可。
+- Electron 44+ 首次打开浏览器窗口时会自动下载二进制(约 100MB,需网络);之后不再需要。若探测时网络不可用,可预装 `ELECTRON_PATH` 指定的二进制。
 - 本插件不含浏览器列 UI——那是宿主外壳的配套,别把"浏览器列"当成插件能力。
 
 ## 开发
 
 ```sh
 # 类型检查 + 构建(lib/)
-pnpm run build
-
-# 功能测试:启动本地页面服务器 + Electron probe(见仓库测试脚本)
+npm run build
 ```
 
+> 运行测试: `npm test`(= `tsc -p tsconfig.json` + `node --test "tests/*.test.mjs"`,假 host 测试,无需 Electron)。
+
 代码结构:
 
 | 目录 | 职责 |
@@ -243,6 +274,21 @@ pnpm run build
 | `src/tool-browser/` | 模型侧 `browser_*` 工具 |
 | `src/types/` | electron 环境类型(shim,避免强制依赖 electron 类型) |
 
+## 更新记录
+
+> 按轮次记录的开发与修复历程(完整明细见 [CHANGELOG.md](CHANGELOG.md))。**0.1.16** 起随版本发布(tag `v0.1.16`)。
+
+| 轮次 | 日期 | 内容 |
+| --- | --- | --- |
+| 第一轮 | 2026-08-18 | **安全与健壮性修复**:RPC 随机 token 认证 + 单连接强制;下载准入(仅 HTTP(S)、绝对路径、`downloadDir` 限定)与流式限流(Content-Length 提前拒绝,256MB 上限);CDP 超时打断与 click/type 超时松键恢复;会话/白名单改为每任务作用域并随 agent 生命周期自动关闭;操作历史脱敏(输入文本、replay/execute 参数不泄露);弹窗重定向回标签页 |
+| 第二轮 | 2026-08 | **功能补全 + 测试 + CI**:窗口标题显示任务标识、showView 无闪烁;快照/无障碍树穿透同源 iframe 与 Shadow DOM;新增 `browser_wait`/`scroll`/`back`/`forward`/`key` 工具;真实 `available()` 探测;下载改由子进程直接落盘(临时文件 + 原子改名);Electron 定位收敛;JPEG/缩放截图;快照性能优化;新增测试套件与 CI |
+| 第三轮 | 2026-08 | **对标 browser-bridge 的功能 + 审查修复**:`browser_a11y` 无障碍树;表单控件 6 件套(`browser_set_value`/`check`/`select`/`clear`/`get_value`/`refresh`);语义定位 `target`(css/text/xpath);`browser_scrape` 结构化提取;独立 BrowserWindow + 真实工具栏(地址栏/后退/前进/刷新/标签条),工具栏操作路由回会话模型;工具总数 **20 → 33**;CI 改 npm(无 lockfile 不兼容 pnpm cache)、README 修正等审查项 |
+| 第四轮 | 2026-08 | **DSH 0.1.1-rc.2 对齐 + 复查修复**:peer 下限对齐 `^0.1.1-rc.2`;修复 `browser_type` 带 target 丢文本、`browser_key` 空格缺 CDP `text`、keyUp 失败卡键、`browser_wait` URL 同源误匹配、download `.part` rename 残留、`snapshotMaxElements`/`contentMaxChars` 配置接线、导出类型补齐;新增 3 个回归测试 |
+| 第五轮 | 2026-08 | **Electron 44 兼容**:`available()` 改为无副作用探测(不再触发 Electron 44 懒下载);`flushAuth` cookie-domain 构建错误修复 |
+| 第六轮 | 2026-08 | **Windows 握手与标签定位**:Electron GUI 进程收不到 piped stdin → RPC token 改 **stdin + 环境变量双通道**;`browser_switch_tab`/`browser_close_tab` 跨会话按 id 定位(`locateTab`),`browser_close_tab` 不再静默假成功,未知 id 报错附带现有标签列表 |
+| 第七轮 | 2026-08 | **工具栏交互(Windows 焦点路由)**:键盘输入只进有焦点的 view,页面 view 抢占焦点导致地址栏无法输入 → 新增 `wireFocusRouting`(点击即聚焦该 view)+ 窗口 refocus 恢复上次点击的 view;真机 OS 输入探针验证 |
+| **0.1.16** | 2026-08-26 | **发布**:以上七轮全部随 **0.1.16** 发布(构建零错误、21 项测试全绿,`v0.1.16`) |
+
 ## 特别感谢
 
 特别感谢 [DeepSeek Harness 原始仓库](https://github.com/deepseek-ai/deepseek-harness) 与 DeepSeek AI 团队:本插件的 seam、工具运行时与插件体系都构建在这个项目之上。

package/docs/README.md

@@ -6,7 +6,7 @@
 | --- | --- |
 | 了解插件为什么存在、与无头方案的区别 | [为什么做共享真实浏览器](why-browser.md) |
 | 安装、配置、日常使用与常见问题 | [用户指南](user-guide.md) |
-| 全部 20 个工具的参数、输出与示例 | [工具参考](tool-reference.md) |
+| 全部 25 个工具的参数、输出与示例 | [工具参考](tool-reference.md) |
 | 了解 seam / provider / 工具三层与自托管实现 | [架构说明](architecture.md) |
 | 回到项目首页 | [README](../README.md) |
 

package/docs/architecture.md

@@ -37,11 +37,13 @@ agent (browser_* 工具)
 
 ### 工具层(`src/tool-browser/`)
 
-20 个 `browser_*` 工具,按**调用方任务**(`exec.agent.id`)维护独立浏览器会话:
+25 个 `browser_*` 工具,按**调用方任务**(`exec.agent.id`)维护独立浏览器会话:
 
 - 会话缓存 `sessionsByTask`:同一任务复用同一会话,并发首开去重;
 - `browser_reset_session` 关闭本任务会话并遗忘映射(即使 close 抛错也清除,下次调用重建);
-- `browser_restrict` 维护模块级白名单,守卫所有非只读工具;
+- 会话生命周期绑定 agent 作用域 ctx:agent(DSH 会话)销毁时自动关闭对应浏览器会话,任务结束后不再泄漏窗口/视图;
+- 会话/开启动态/白名单都是**每插件实例(每 context)作用域**的,多 context 互不共享、互不污染;
+- `browser_restrict` 维护本实例白名单,守卫所有非只读工具;
 - 输出 schema 与返回值严格一致(DSH 运行时会校验,`additionalProperties: false` 下多一个字段都会报错)。
 
 ## 自托管实现(纯 `dsh web`)
@@ -57,9 +59,10 @@ RemoteElectronViewHost  ──TCP JSON-RPC──▶  host-main.js
 ```
 
 - **协议**:本机 loopback TCP,每行一个 JSON(`{ id, op, ... }` ↔ `{ id, ok, result|err }`);
+- **认证**:每次 spawn 生成随机 token(经 `--rpc-token` 传入),子进程首条消息必须回传该 token(`hello`);服务端只接受**一个**连接,其余连接直接断开——本机其他进程无法伪冒子进程或注入回复;
 - **Electron 定位**:① peer 依赖 → ② `ELECTRON_PATH` → ③ 锚点与 pnpm store 中**版本最新**者(33.x 有合成器缺陷,建议 ≥ 40);
-- **稳健性**:子进程/套接字都有 `error` 监听(否则未捕获事件会炸掉整个 DSH 进程);子进程退出自动重启;物化失败可重试;下载有 256MB 上限与 60s 超时;cookie 导出/恢复有 30s 超时;
-- **视图可见性**:多标签/多会话时,`showView` 隐藏其他视图并把目标视图置顶(remove+re-add),确保用户看到的是活动标签;
+- **稳健性**:子进程/套接字都有 `error` 监听(否则未捕获事件会炸掉整个 DSH 进程);子进程退出自动重启;物化失败可重试;下载仅限 HTTP(S)、`savePath` 必须绝对路径(默认限定 `~/Downloads`,可配置 `downloadDir` 覆盖)、流式限流 + Content-Length 提前拒绝、256MB 上限与 60s 超时,文件由**子进程直接落盘**(临时文件 + rename,不再经 RPC 传 base64);cookie 导出/恢复有 30s 超时;
+- **视图可见性**:多标签/多会话时,`showView` 隐藏其他视图并把目标视图置顶(remove+re-add),确保用户看到的是活动标签;目标是当前可见视图时**跳过** remove/re-add(否则每次操作都闪烁);窗口标题实时显示当前可见会话的任务标识 + 页面标题/URL(`showView` 携带会话 label,子进程读 `getTitle`/`getURL`);窗口 resize 时视图 bounds 自动跟随;
 - **孤儿防护**:父进程断开时子进程自动退出,不留僵尸窗口;
 - **cookie 落盘**:子进程使用独立 userData 目录(`<DSH_HOME>/dsh-builtin-browser-host`),登录态跨重启保留(另有 `browser_auth` 手动导出/恢复)。
 

package/docs/tool-reference.md

@@ -1,13 +1,15 @@
 # 工具参考
 
-全部 20 个 `browser_*` 工具。守卫列:✅ 表示该动作受 `browser_restrict` 白名单约束;只读工具永不拦截。
+全部 33 个 `browser_*` 工具。守卫列:✅ 表示该动作受 `browser_restrict` 白名单约束;只读工具永不拦截。
 
 ## 页面与导航
 
 | 工具 | 参数 | 输出 | 守卫 | 说明 |
 | --- | --- | --- | --- | --- |
 | `browser_open` | `url`(必填), `newTab?` | 快照(url/title/elements/truncated/challenge) | ✅ | 打开 URL,返回带编号元素的快照;`newTab: true` 在新标签打开 |
-| `browser_snapshot` | – | 快照 | – | 交互元素(输入框/按钮/链接)编号清单,供定位与点击 |
+| `browser_wait` | `timeoutMs?`, `url?`, `selector?` | `{ ready, reason }` | – | 等待页面加载完成(可选期望 URL / CSS 选择器);未就绪不抛错,返回原因 |
+| `browser_snapshot` | – | 快照 | – | 交互元素(输入框/按钮/链接)编号清单,供定位与点击;穿透同源 iframe 与 Shadow DOM,iframe 内元素标注 `frame` |
+| `browser_a11y` | `includeHidden?`, `maxNodes?`(10-5000,默认 500) | `{ url, title?, count, nodes[], truncated }` | – | 无障碍树:每个交互节点的 `role`/`name`/`value`/`states`/`depth`/`tag`/坐标;优先 Chrome `computedRole`/`computedName`,穿透同源 iframe 与 Shadow DOM;坐标可直接喂 `browser_click`/`browser_type` |
 | `browser_content` | `format`(html/markdown/txt/json,必填), `selector?`, `maxChars?`, `timeoutMs?` | `{ content, truncated }` | – | 抓取页面内容;`selector` 限定区域 |
 | `browser_challenge` | – | `{ blocked, kind?, reason?, hint? }` | – | 检测人机验证(CAPTCHA/Cloudflare/reCAPTCHA/hCaptcha/Turnstile);阻塞时请用户处理 |
 
@@ -16,9 +18,20 @@
 | 工具 | 参数 | 输出 | 守卫 | 说明 |
 | --- | --- | --- | --- | --- |
 | `browser_execute` | `script`(必填), `args?` | `{ ok, value? / exception? }` | ✅ | 在页面执行 JS;脚本以 `return` 开头或作为表达式;`args` 以 `arguments[0..n]` 传入 |
-| `browser_click` | `x`, `y`(必填) | `{ clicked }` | ✅ | 视口坐标点击(配合截图做视觉定位) |
-| `browser_type` | `text`(必填) | `{ typed }` | ✅ | 向聚焦元素输入文本(CDP `Input.insertText`) |
+| `browser_click` | `target?`(css/text/xpath), `x?`, `y?`(target 与坐标二选一) | `{ clicked }` | ✅ | 语义目标点击:按 `target` 定位、滚动到视口中央再点中心;或视口坐标点击(配合截图做视觉定位) |
+| `browser_type` | `text`(必填), `target?` | `{ typed }` | ✅ | 输入文本;传 `target` 先聚焦该元素(CDP `Input.insertText`) |
+| `browser_key` | `key`(必填,枚举) | `{ pressed }` | ✅ | 按命名按键:Enter/Tab/Escape/Backspace/Delete/方向键/Home/End/PageUp/PageDown/Space |
+| `browser_scroll` | `deltaX?`, `deltaY?`, `selector?`, `toTop?`, `toBottom?` | `{ scrolled }` | ✅ | 滚动页面:像素增量 / 选择器定位 / 顶部底部 |
+| `browser_back` | – | `{ back }` | ✅ | 页面历史后退一步(无前项时为空操作) |
+| `browser_forward` | – | `{ forward }` | ✅ | 页面历史前进一步(无后项时为空操作) |
+| `browser_refresh` | – | `{ refreshed }` | ✅ | 刷新当前页(等价浏览器刷新按钮) |
 | `browser_fill` | `fields`(必填,数组), `submit?` | `{ fields[], submitted }` | ✅ | 批量填表;字段按 `selector`/`name`/`label`/`placeholder` 匹配,值支持字符串/数字/布尔;单个字段失败不影响其余;`submit: true` 提交表单 |
+| `browser_set_value` | `target`(必填), `value`(必填) | `{ method, value }` | ✅ | 单个控件设值(按 css/text/xpath 定位);原生 setter + input/change,React 受控输入可用;select 按值/文本 |
+| `browser_check` | `target`(必填), `checked?` | `{ checked }` | ✅ | 勾选/取消勾选 checkbox 或 radio(按 target 定位) |
+| `browser_select` | `target`(必填), `optionValue?`/`optionText?`/`optionIndex?`(三选一) | `{ value, text }` | ✅ | 选中 `<select>` 的某个选项(按 target 定位) |
+| `browser_clear` | `target`(必填) | `{ cleared }` | ✅ | 清空输入/文本域/contenteditable,或取消勾选 checkbox/radio |
+| `browser_get_value` | `target`(必填) | `{ value?, checked?, selectedText? }` | – | 读取元素当前值,用于操作后验证 |
+| `browser_scrape` | `item`(必填), `fields`(必填,映射), `timeoutMs?` | `{ count, items[] }` | – | 结构化提取:容器选择器 + 字段映射(如 `{"title": "h3", "url": "a@href"}`);静态 CSS 查询、不执行任意代码、CSP 安全;等待 item 出现(默认 5s) |
 
 ## 标签与会话
 
@@ -36,27 +49,27 @@
 | 工具 | 参数 | 输出 | 守卫 | 说明 |
 | --- | --- | --- | --- | --- |
 | `browser_history` | – | `{ entries[] }` | – | 操作日志(最新在后),含 seq/action/ok/params/result/error |
-| `browser_replay` | `seq`(必填) | `{ replayed }` | ✅ | 按序号回放某一步(navigate/execute/click/type) |
-| `browser_download` | `url`(必填), `savePath`(必填) | `{ path }` | ✅ | 带会话 cookie 下载到本地(上限 256MB,受 CORS 约束) |
+| `browser_replay` | `seq`(必填) | `{ replayed }` | ✅ | 按序号回放某一步(navigate/execute/click/type/scroll/key) |
+| `browser_download` | `url`(必填), `savePath`(必填) | `{ path }` | ✅ | 带会话 cookie 下载到本地(仅 http(s);`savePath` 必须为绝对路径,默认限定在 `~/Downloads`,可用 `downloadDir` 覆盖;上限 256MB,受 CORS 约束;由子进程直接落盘) |
 
 ## 登录态与安全
 
 | 工具 | 参数 | 输出 | 守卫 | 说明 |
 | --- | --- | --- | --- | --- |
 | `browser_auth` | `action`(flush/restore,必填), `cookies?` | `{ cookies[]? / restored? }` | ✅ | 导出/恢复 cookie(自托管可用);flush 返回 cookie 列表,restore 带列表写回 |
-| `browser_restrict` | `allowed?` | `{ restrictedTo[] }` | – | 设置动作白名单;空列表解除;未知工具名报错 |
+| `browser_restrict` | `allowed?` | `{ restrictedTo[] }` | – | 设置动作白名单;空列表解除;未知工具名报错。**软护栏**——模型可自行解除,非安全边界 |
 
 ## 截图
 
 | 工具 | 参数 | 输出 | 守卫 | 说明 |
 | --- | --- | --- | --- | --- |
-| `browser_screenshot` | `fullPage?`, `savePath?` | `{ dataUrl, path? }` | – | PNG 截图;`savePath` 落盘供视觉模型读取 |
+| `browser_screenshot` | `fullPage?`, `savePath?`, `format?`(png/jpeg), `quality?`, `maxWidth?`, `maxHeight?` | `{ dataUrl, path? }` | – | 截图;PNG 默认,JPEG 仅自托管原生路径;`maxWidth`/`maxHeight` 等比缩放降低视觉模型开销;`savePath` 落盘供视觉模型读取 |
 
 ## 常用组合
 
 **调研一个网站**
 ```
-browser_open https://site → browser_content format=markdown → browser_snapshot → 逐页浏览
+browser_open https://site → browser_wait(url=...) → browser_content format=markdown → browser_snapshot → 逐页浏览
 ```
 
 **登录并下载文件**

package/docs/user-guide.md

@@ -33,6 +33,7 @@ dsh plugin --profile web add <本仓库路径>
 | `browser-electron` | `httpOnly` | 布尔 | `true` | 仅允许 HTTP(S) 导航;`file:`/`data:` 等拒绝 |
 | `browser-electron` | `snapshotMaxElements` | 数字 | `60` | 快照最多收录的交互元素数 |
 | `browser-electron` | `contentMaxChars` | 数字 | `100000` | 内容抓取默认字符上限 |
+| `browser-electron` | `downloadDir` | 字符串 | `~/Downloads` | 限定 `browser_download` 保存路径必须位于该目录内;默认收敛到系统下载目录,可改沙箱目录 |
 | `tool-browser` | `timeoutMs` | 数字 | `60000` | 工具协作超时(ms) |
 | `tool-browser` | `tabTools` | 布尔 | `true` | 是否注册标签管理工具 |
 
@@ -40,11 +41,12 @@ dsh plugin --profile web add <本仓库路径>
 
 ```
 1. browser_open 打开 https://example.com
-2. browser_snapshot 查看页面有哪些可交互元素(带编号)
+2. 慢站点先 browser_wait(url=…) 等页面就绪,再 browser_snapshot 查看可交互元素
 3. 需要填表时用 browser_fill(按 name/label/placeholder 匹配,一次填多个字段)
-4. 需要截图确认时用 browser_screenshot(可 savePath 存文件)
-5. 遇到验证码(browser_challenge 或快照标注 CHALLENGE)时,停下请用户处理
-6. 每次操作后告知用户你在页面上做了什么
+4. 需要截图确认时用 browser_screenshot(可 savePath 存文件,大页面用 maxWidth 缩小)
+5. 需要滚动/回退/按键时用 browser_scroll / browser_back / browser_forward / browser_key
+6. 遇到验证码(browser_challenge 或快照标注 CHALLENGE)时,停下请用户处理
+7. 每次操作后告知用户你在页面上做了什么
 ```
 
 ## 操作纪律
@@ -79,10 +81,10 @@ dsh plugin --profile web add <本仓库路径>
 窗口标题为 `dsh-browser`(自托管)。若子进程崩溃会自动重启;重启后旧会话失效,调用 `browser_reset_session` 重建。
 
 **Q:下载报 CORS 错误?**
-`browser_download` 在页面上下文内 `fetch`,受同源/CORS 约束;跨域文件请先在同源页面内操作,或直接请求用户提供。
+`browser_download` 在页面上下文内 `fetch`,受同源/CORS 约束;跨域文件请先在同源页面内操作,或直接请求用户提供。仅支持 HTTP(S) URL;`savePath` 必须为绝对路径(配置 `downloadDir` 后限定在该目录内)。
 
 **Q:如何禁止 agent 乱点?**
-`browser_restrict` 设置白名单(如只允许 `browser_snapshot`/`browser_content`);传空列表解除。
+`browser_restrict` 设置白名单(如只允许 `browser_snapshot`/`browser_content`);传空列表解除。注意它是防误操作的**软护栏**,模型可自行解除,不是安全边界。
 
 ## 故障排查
 

package/lib/browser/runtime.d.ts

@@ -8,9 +8,9 @@
  */
 import { Context, Service } from '@deepseek-ai/cordis';
 import z from '@deepseek-ai/schemastery';
-import type { BrowserClickRequest, BrowserContentRequest, BrowserContentResult, BrowserDownloadRequest, BrowserExecuteRequest, BrowserExecuteResult, BrowserFillRequest, BrowserFillResult, BrowserHistoryEntry, BrowserNavigateRequest, BrowserOpenRequest, BrowserProvider, BrowserScreenshotRequest, BrowserScreenshotResult, BrowserSessionId, BrowserSnapshotResult, BrowserTab, BrowserTypeRequest, BrowserChallenge, ExportedCookie } from './types.ts';
-export { BrowserError, } from './types.ts';
-export type { BrowserChallenge, BrowserClickRequest, BrowserContentFormat, BrowserContentRequest, BrowserContentResult, BrowserDownloadRequest, BrowserExecuteRequest, BrowserExecuteResult, BrowserFillField, BrowserFillRequest, BrowserFillResult, BrowserHistoryEntry, BrowserNavigateRequest, BrowserOpenRequest, BrowserProvider, BrowserScreenshotRequest, BrowserScreenshotResult, BrowserSessionId, BrowserSnapshotElement, BrowserSnapshotResult, BrowserTab, BrowserTypeRequest, ExportedCookie, } from './types.ts';
+import type { BrowserA11yRequest, BrowserA11yResult, BrowserCheckRequest, BrowserClearRequest, BrowserClickRequest, BrowserContentRequest, BrowserContentResult, BrowserDownloadRequest, BrowserElementTarget, BrowserExecuteRequest, BrowserExecuteResult, BrowserFillRequest, BrowserFillResult, BrowserGetValueRequest, BrowserGetValueResult, BrowserHistoryEntry, BrowserNavigateRequest, BrowserOpenRequest, BrowserProvider, BrowserScrapeRequest, BrowserScrapeResult, BrowserScreenshotRequest, BrowserScreenshotResult, BrowserSelectRequest, BrowserSelectResult, BrowserSessionId, BrowserSetValueRequest, BrowserSetValueResult, BrowserSnapshotResult, BrowserTab, BrowserTypeRequest, BrowserWaitRequest, BrowserWaitResult, BrowserScrollRequest, BrowserKeyRequest, BrowserChallenge, ExportedCookie } from './types.js';
+export { BrowserError, } from './types.js';
+export type { BrowserA11yNode, BrowserA11yRequest, BrowserA11yResult, BrowserChallenge, BrowserCheckRequest, BrowserClearRequest, BrowserClickRequest, BrowserContentFormat, BrowserElementTarget, BrowserGetValueRequest, BrowserGetValueResult, BrowserScrapeField, BrowserScrapeRequest, BrowserScrapeResult, BrowserSelectRequest, BrowserSelectResult, BrowserSetValueRequest, BrowserSetValueResult, BrowserWaitRequest, BrowserWaitResult, BrowserScrollRequest, BrowserKeyRequest, BrowserContentRequest, BrowserContentResult, BrowserDownloadRequest, BrowserExecuteRequest, BrowserExecuteResult, BrowserFillField, BrowserFillRequest, BrowserFillResult, BrowserHistoryEntry, BrowserNavigateRequest, BrowserOpenRequest, BrowserProvider, BrowserScreenshotRequest, BrowserScreenshotResult, BrowserSessionId, BrowserSnapshotElement, BrowserSnapshotResult, BrowserTab, BrowserTypeRequest, ExportedCookie, } from './types.js';
 declare module '@deepseek-ai/cordis' {
     interface Context {
         browser: BrowserRuntime;
@@ -53,7 +53,7 @@ export declare class BrowserRuntime extends Service {
     /** Resolve the selected provider or throw the matching {@link BrowserError}. */
     private resolveProvider;
     /** Open a new browser session through the selected provider. */
-    open(): Promise<BrowserSessionId>;
+    open(label?: string): Promise<BrowserSessionId>;
     /** Open a URL through the selected provider, optionally in a new tab. */
     openUrl(session: BrowserSessionId, request: BrowserOpenRequest, signal?: AbortSignal): Promise<void>;
     /** List the session's tabs through the selected provider. */
@@ -68,16 +68,50 @@ export declare class BrowserRuntime extends Service {
     navigate(session: BrowserSessionId, request: BrowserNavigateRequest, signal?: AbortSignal): Promise<void>;
     /** Execute JS in the session's page context through the selected provider. */
     execute(session: BrowserSessionId, request: BrowserExecuteRequest, signal?: AbortSignal): Promise<BrowserExecuteResult>;
+    /** Wait until the session's page is ready (and optional URL/selector match). */
+    waitFor(session: BrowserSessionId, request: BrowserWaitRequest, signal?: AbortSignal): Promise<BrowserWaitResult>;
     /** Produce an AI-friendly snapshot of the session's page. */
     snapshot(session: BrowserSessionId, signal?: AbortSignal): Promise<BrowserSnapshotResult>;
+    /** Read the session's page accessibility tree through the selected provider. */
+    a11y(session: BrowserSessionId, request: BrowserA11yRequest, signal?: AbortSignal): Promise<BrowserA11yResult>;
+    /** Reload the session's active tab through the selected provider. */
+    reload(session: BrowserSessionId, signal?: AbortSignal): Promise<void>;
     /** Fetch page content in a requested format. */
     content(session: BrowserSessionId, request: BrowserContentRequest, signal?: AbortSignal): Promise<BrowserContentResult>;
-    /** Click at viewport coordinates through the selected provider. */
-    click(session: BrowserSessionId, request: BrowserClickRequest, signal?: AbortSignal): Promise<void>;
-    /** Type into the focused element through the selected provider. */
-    type(session: BrowserSessionId, request: BrowserTypeRequest, signal?: AbortSignal): Promise<void>;
+    /** Click at viewport coordinates (or a located element) through the selected provider. */
+    click(session: BrowserSessionId, request: BrowserClickRequest | {
+        readonly target: BrowserElementTarget;
+    }, signal?: AbortSignal): Promise<void>;
+    /** Type into the focused element (or a located one) through the selected provider. */
+    type(session: BrowserSessionId, request: BrowserTypeRequest | {
+        readonly target: BrowserElementTarget;
+    }, signal?: AbortSignal): Promise<void>;
+    /** Scroll the session's page through the selected provider. */
+    scroll(session: BrowserSessionId, request: BrowserScrollRequest, signal?: AbortSignal): Promise<void>;
+    /** Go back in the session's page history through the selected provider. */
+    back(session: BrowserSessionId, signal?: AbortSignal): Promise<void>;
+    /** Go forward in the session's page history through the selected provider. */
+    forward(session: BrowserSessionId, signal?: AbortSignal): Promise<void>;
+    /** Press one named key through the selected provider. */
+    key(session: BrowserSessionId, request: BrowserKeyRequest, signal?: AbortSignal): Promise<void>;
     /** Fill a form's fields in one batch through the selected provider. */
     fillForm(session: BrowserSessionId, request: BrowserFillRequest, signal?: AbortSignal): Promise<BrowserFillResult>;
+    /** Set one element's value through the selected provider. */
+    setValue(session: BrowserSessionId, request: BrowserSetValueRequest, signal?: AbortSignal): Promise<BrowserSetValueResult>;
+    /** Check or uncheck one checkbox/radio through the selected provider. */
+    check(session: BrowserSessionId, request: BrowserCheckRequest, signal?: AbortSignal): Promise<{
+        readonly checked: boolean;
+    }>;
+    /** Select one option of a `<select>` through the selected provider. */
+    selectOption(session: BrowserSessionId, request: BrowserSelectRequest, signal?: AbortSignal): Promise<BrowserSelectResult>;
+    /** Clear one element through the selected provider. */
+    clearField(session: BrowserSessionId, request: BrowserClearRequest, signal?: AbortSignal): Promise<{
+        readonly cleared: boolean;
+    }>;
+    /** Read one element's value through the selected provider. */
+    getValue(session: BrowserSessionId, request: BrowserGetValueRequest, signal?: AbortSignal): Promise<BrowserGetValueResult>;
+    /** Extract structured data through the selected provider. */
+    scrape(session: BrowserSessionId, request: BrowserScrapeRequest, signal?: AbortSignal): Promise<BrowserScrapeResult>;
     /** Capture the current page through the selected provider. */
     screenshot(session: BrowserSessionId, request?: BrowserScreenshotRequest, signal?: AbortSignal): Promise<BrowserScreenshotResult>;
     /** Check for a human-verification challenge on the active tab. */