dsh-tabbit 0.2.3 → 0.3.2

package/CHANGELOG.md

@@ -0,0 +1,133 @@
+# Changelog
+
+## 0.3.2
+
+- `tabbit_browser` gains `list_tasks: true`: a zero-side-effect readback of
+  the browser tasks this session already has open (which one an omitted
+  `task` would target next), read from the in-memory session registry only.
+  A memory jog for when the model has lost track after a gap in the
+  conversation, instead of guessing a task name or opening a redundant one.
+- `claim_tabs` now works when the task already exists, not only when the
+  same call creates it. Runtime Service silently drops a creation-time
+  `--claim-tab` on a reused task, so previously the plugin threw
+  `CLAIM_REQUIRES_NEW_TASK` and forced starting a fresh task just to attach
+  a tab the user pointed at mid-conversation. It now falls back to the
+  standalone `claim` CLI command for tasks the session already knows about,
+  so an already-open tab can join ongoing browser state instead of always
+  starting over.
+- Fixed the bundled `SKILL.md`'s frontmatter parser assuming LF-only line
+  endings: a copy checked out or published with CRLF endings would have had
+  its raw YAML header served to the model as skill content instead of being
+  stripped.
+
+## 0.3.1
+
+- Fixed the web client plugin never activating on current DSH hosts — the web
+  UI showed `dsh-tabbit: pending (waiting for service: conversationEvents)`.
+  The DSH client runtime refactor (already part of `0.1.2-alpha.1`) renamed the
+  `conversationEvents` service to `uiConversation` and moved node registration
+  from `service.register()` to `service.events.register()`. The client plugin
+  now injects `uiConversation`; the `/tabbit-info` status card, the `@tab`
+  composer source, and the watching-instance hint all had been blocked behind
+  that pending inject and work again. The `conversation.chat.node` keyed slot
+  registration and the Definition contract (`match`/`start`/`update`/
+  `buildViewNode`) are unchanged.
+
+## 0.3.0
+
+- Major upgrade: browser automation now runs through the native
+  `tabbit_browser` tool — real Playwright code executing in the user's Tabbit
+  Browser profile with shared login state, screenshots into model context,
+  automatic interruption recovery, and multi-instance resolution — plus a
+  browser-backed `web_fetch` provider, `@tab` page mentions in the dsh web
+  composer, a dedicated page-access permission, and a `/tabbit` diagnostics
+  command. The bundled skill now teaches the tool workflow instead of the
+  `tabbit-cli` shell workflow. Requires Node 22.19 or newer (was 20).
+- Adopted the renamed `tabbit-cli` launcher that current Tabbit Browser
+  builds register (with fallback to the legacy `tabbit-playwright` name), and
+  the per-instance `.product` labels in the instance registry.
+- New model-free tab inventory: `ctx.tabbit.listAllTabs()` and the loopback
+  route `GET /tabbit/tabs` list every tab of the running browser profile
+  (including the user's own tabs) by talking to the Runtime Service endpoint
+  directly — no launcher process, no task, no side effects, ~1ms steady-state.
+  Requires a 1.11.16+ Tabbit Browser; the browser is never launched for a
+  listing, offline instances report as unavailable instead.
+- The `@` mention menu in the dsh web composer now lists the user's own
+  browser tabs alongside agent task pages (active tab first; only http(s)
+  pages are offered, since chrome://, extension, and blank/new-tab pages
+  cannot deliver content when mentioned), under a "网页标签" group shown
+  below DSH's built-in file/session mentions, with up to 50 candidates
+  instead of 8. Mentioning a user tab fetches a fresh copy of that URL's
+  content through the shared browser fetch task — the user's actual tab is
+  never claimed, moved, or focused. The extracted page text no longer lands
+  in the sent message itself: the composer sends a short `@Title` mention,
+  and the full text is delivered to the model as a separate, collapsed
+  "Context injection" entry via an `agent/pre-step` hook — the same delivery
+  path DSH's own `@file`/`@session` mentions use — so a mentioned page's
+  content never floods the visible chat bubble. `tabbit_browser` gained
+  `list_tabs: true` for the model-facing version of the same inventory, paired
+  with `claim_tabs` to attach a specific user tab to a new task.
+- Windows support: the launcher is discovered under
+  `%LOCALAPPDATA%\Tabbit\LocalAgent\bin\tabbit-cli.exe`, instance selection
+  defers to the native CLI when the instance registry is not readable, and the
+  environment preflight falls back to runtime-process detection.
+- The environment preflight (`tabbit_browser_install`), the daily plugin
+  update check, and the `tabbit_plugin_update` tool carry over from 0.2.x.
+  The update check now reads the npm registry for the latest version (with
+  release notes served from the published tarball), and stays silent for
+  browser-managed (preinstalled) copies, which update together with Tabbit
+  Browser.
+- The bundled skill gained references for the runtime interaction helpers
+  (`tabbit.observe`, `pasteText`, `triggerAndObserve`, …) and AI-mode ARIA
+  snapshot recipes for canvas-backed and visually complex surfaces. It is now
+  named `tabbit`, deliberately matching the shared skill Tabbit Browser
+  installs into `~/.agents/skills/tabbit/`: DSH's same-name precedence
+  prefers that browser-managed copy, and the bundled one is the fallback for
+  machines without a current browser. The `/tabbit` diagnostics command was
+  renamed to `/tabbit-info` to leave the `/tabbit` slash entry to the skill.
+- Task finishing adapted to the new runtime semantics (plain `finish` now
+  retains tabs as a resumable group): the plugin always passes an explicit
+  `--keep` or `--discard`, which behaves correctly on both CLI generations,
+  so `finish` without `keep_tabs` and end-of-session cleanup reliably close
+  task tabs.
+- DSH `0.1.2-alpha.1` also ships its own direct-HTTP `web_fetch` provider and
+  pins the `web` service to it; the bundle patch now re-pins `fetchProvider`
+  to this plugin's browser-backed provider, keeping JS rendering, login
+  state, the page-access/intranet permission gates, and the browser's proxy
+  environment (the built-in fetcher connects directly to resolved public IPs
+  and refuses every domain on fake-ip proxy setups).
+- Requires DSH `0.1.2-alpha.1` or newer, whose standard agent preset ships
+  with `web_fetch` enabled — the bundled "Tabbit mode" preset is gone, and
+  plugin activation removes a previously installed managed copy (marker
+  protocol respected: user-owned copies are never touched).
+
+## 0.2.3
+
+- Fixed the bundle's `cordis.patch.yml` loader entry to use the npm package
+  name `dsh-tabbit` instead of the non-existent package `tabbit-browser`. This
+  resolves the `Cannot find package 'tabbit-browser'` load failure that
+  affected fresh installs after the package was renamed in 0.2.1.
+
+## 0.2.2
+
+- Install and update commands now use the npm package (`dsh plugin add
+  dsh-tabbit`) as the primary route — prebuilt tarball, CDN-served, and
+  counted in npm download stats; the `github:` source remains documented as
+  a fallback.
+
+## 0.2.1
+
+- Renamed the package to `dsh-tabbit` and published it to npm; updated all
+  remaining references (daily update-check changelog URL, install commands in
+  the plugin, skill, and READMEs) from `Tabbit-Browser/dsh-plugin` to
+  `Tabbit-Browser/dsh-tabbit`.
+
+## 0.2.0
+
+- Added a daily plugin update check. Loading the skill compares the installed
+  plugin version with the published changelog — cached for 24 hours, at most
+  one request per day, silent when offline — and, when outdated, asks the user
+  whether to update, showing what the new version added.
+- Added the `tabbit_plugin_update` tool. It records a version the user declined
+  so the skill stops announcing it, and can force a recheck after a plugin
+  update or a connectivity change.

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Tabbit-Browser
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.en.md

@@ -0,0 +1,141 @@
+# dsh-tabbit — Tabbit's Official DeepSeek Harness Plugin
+
+**English** | [简体中文](README.md) | [Changelog](CHANGELOG.md)
+
+![Tabbit Browser for DeepSeek Harness](assets/dsh-tabbit-banner.png)
+
+A DeepSeek Harness (dsh) plugin bundle for Tabbit Browser. Through this
+plugin, dsh calls on Tabbit to complete agent tasks: real pages, real login
+state, real interactions, driven through a native code-first tool (not shell
+round-trips). Use it for web automation, information extraction, QA, and
+benchmarks.
+
+## What you get
+
+| Component | Description |
+|---|---|
+| `tabbit_browser` | Read and operate web pages using Tabbit's real, built-in CLI mode. |
+| Browser-backed `web_fetch` | Re-points dsh's `web_fetch` to Tabbit itself (dsh's built-in direct fetcher has no JS rendering, no login state, and bypasses the system proxy — under a fake-IP proxy setup it refuses every domain). |
+| "Web tabs" `@` mention | Type `@` in the dsh Web UI input box to list pages open in this session's browser tasks **and every tab in the user's browser** — pick one to add as context. |
+| Dedicated "page access" permission | Full permission controls. Because dsh shares the user's cookies when it reaches Tabbit pages, it confirms `pageAccess` (asked once per session by default) and `intranetFetch` (per-request approval when `web_fetch` targets an intranet address). |
+| `tabbit_browser_install` tool | Environment preflight: detects an installed stable Tabbit build and verifies the launcher and Runtime Service; downloads Tabbit as a dsh background job when it's missing or outdated. |
+| `tabbit_plugin_update` tool | Plugin update check: asks npm for the latest release at most once a day, and silently installs a suitable version in the background. |
+| `/tabbit-info` command | Type `/tabbit-info` in the dsh input box for diagnostics: launcher, instance list (with product names), effective instance and its source, permission settings, task occupancy. |
+| `tabbit` skill | Teaches the model best practices for using Tabbit. Defaults to Tabbit's own official skill at `~/.agents/skills/tabbit/` (it evolves with the browser runtime); this plugin bundles a fallback copy. |
+
+## Installation
+
+### Prerequisites
+
+- A stable Tabbit Browser build ([international edition](https://www.tabbit.ai) or [China edition](https://www.tabbit.com/), `1.9.0` or newer) that has been launched at least once (this registers the CLI launcher on first launch).
+- Node.js `>=22.19` and dsh `>=0.1.1-rc.2` (install dsh with `npm install -g @deepseek-ai/dsh`). On hosts below `0.1.2-alpha.1`, the built-in `web_fetch` tool is unavailable in Web app sessions (`tabbit_browser` is unaffected).
+
+### Install dsh-tabbit
+
+```bash
+dsh plugin --profile web add dsh-tabbit                 # primary npm route
+```
+
+### Other install methods
+
+```bash
+dsh plugin --profile web add github:Tabbit-Browser/dsh-tabbit # fallback when npm is unreachable
+dsh plugin --profile web add link:/path/to/dsh-tabbit   # local development
+```
+
+> This package supersedes the earlier `tabbit-browser` skill-only plugin, and
+> continues on from the 0.2.x line published on npm — 0.2.x users get an
+> upgrade notice from the daily update check, and simply re-running the
+> install command upgrades in place.
+
+## Community & Support
+
+Scan the QR code below to join the **dsh-tabbit Developer Group** to share feedback, ask questions, and discuss new features:
+
+![dsh-tabbit Developer Group](assets/dsh-tabbit-developer-community-qr.png.jpg)
+
+## Settings
+
+### Basic configuration
+
+dsh Settings → `tabbit`, or `$DSH_HOME/settings.yaml`
+
+```yaml
+tabbit:
+  instance: ""            # explicit 16-hex instance id (/tabbit-info lists them); usually leave empty
+  launcherPath: ""        # override; default discovers tabbit-cli, falls back to tabbit-playwright; %LOCALAPPDATA%\Tabbit\LocalAgent\bin\tabbit-cli.exe on Windows
+  pageAccess: ask         # ask (once per session) | always | never
+  intranetFetch: ask      # web_fetch to intranet/loopback targets: ask (once per session+origin) | always | never
+```
+
+### Instance resolution priority
+
+Priority order when this machine has more than one Tabbit build installed:
+
+1. an explicit `tabbit.instance` setting;
+2. **the Tabbit instance currently viewing dsh-web** (auto-detected: the
+   client plugin pings `/tabbit/instance-hint` on page load, and the server
+   traces the loopback socket's peer process up its parent chain to match the
+   instance registry's `browserPid`; macOS only — naturally misses when a
+   non-Tabbit browser has the page open) — "execute in whichever Tabbit
+   you're viewing dsh in";
+3. an inherited `TABBIT_PLAYWRIGHT_INSTANCE` environment variable (the
+   authoritative channel in the embedded form: Tabbit injects its own
+   instance id when it launches its bundled dsh);
+4. automatic registry selection (the single online instance; an ambiguity
+   error listing the candidates otherwise — on Windows, when the registry
+   isn't readable, the native CLI picks for itself).
+
+Check the currently effective source anytime with `/tabbit-info`
+(`execution instance: ... (via ...)`).
+
+**Full access note**: dsh's `danger-full-access` permission preset writes the
+session's approval policy as `never` (dsh's definition: auto-deny every ask).
+This plugin's permission gate detects that override and **auto-allows**
+instead of issuing an ask that's guaranteed to be denied — full access means
+full access (bash is already unrestricted in that mode, so gating only the
+browser tool has no defensive value). Exception: a deployment-level default
+of `never` (not a session override) isn't visible to the public API and still
+gets denied — that combination gets a denial message pointing at the
+`tabbit.pageAccess: always` escape hatch.
+
+## Permissions & Security
+
+- Because the agent shares the user's real login state, `pageAccess` is a
+  dedicated permission independent of filesystem/sandbox permissions, and
+  asks for user confirmation by default.
+- Authorization is remembered: once a tool call succeeds, this session won't
+  ask again; failures aren't recorded (a retry after failure asks again).
+
+## Development
+
+Developing and testing this project depends on a local
+[deepseek-harness](https://github.com/deepseek-ai/deepseek-harness) checkout.
+The `@deepseek-ai/*` packages published on npm generally lag too far behind
+to install directly as dependencies, so you need to build a harness checkout
+locally first, then point the repo's `.dsh-harness` symlink at it with the
+script below (that path is already ignored in `.gitignore`). That's the only
+place you configure it — neither `tsconfig.json` nor `package.json` needs
+any changes.
+
+Get and build deepseek-harness (skip if you already have one):
+```bash
+git clone https://github.com/deepseek-ai/deepseek-harness && cd deepseek-harness && pnpm install && pnpm build && cd ..
+```
+
+Point this repo at it — replace `/path/to/deepseek-harness` with your actual checkout path:
+```bash
+npm run link-harness -- /path/to/deepseek-harness   # or set DSH_HARNESS_PATH
+```
+
+Install and build:
+```bash
+pnpm install && pnpm build   # tsc → lib/
+npm test                     # build + node --test tests/
+```
+
+## Known limitations / Roadmap
+
+- Mentioning bookmarks/favorites isn't supported yet.
+- Screenshots entering context require a model route that accepts image input.
+- Windows regression testing is limited.

package/README.md

@@ -1,113 +1,107 @@
-# tabbit-browser for DeepSeek Harness
+# dsh-tabbit — Tabbit 的 DeepSeek Harness 官方插件
 
-**English** | [简体中文](README.zh-CN.md)
+[English](README.en.md) | **简体中文** | [Changelog](CHANGELOG.md)
 
-![Tabbit Browser for DeepSeek Harness](tabbit-for-dsh.png?v=2)
+![Tabbit Browser for DeepSeek Harness](assets/dsh-tabbit-banner.png)
 
-A plugin for DeepSeek Harness (DSH) that gives the agent control over your Tabbit Browser: real pages, real login state, and real interactions, driven through `tabbit-cli` — the task-isolated Playwright CLI owned by the browser itself. Use it for web automation, information extraction, QA, and benchmarks.
+Tabbit Browser 的 DeepSeek Harness（dsh）插件包（bundle）。dsh 可以通过此插件调用 Tabbit 完成 Agent 任务：真实页面、真实登录态、真实交互，经原生 code-first 工具驱动（不走 shell 转发）。适用于网页自动化、信息提取、QA 与评测。
 
-## What you get
+## 能力
 
-| Component | Description |
-| --------- | ----------- |
-| `tabbit-browser` skill | The working guide for browser automation: persistent task spaces, locators and waits, screenshots, receipts and recovery. Discovered and loaded automatically with the plugin — no separate skill install. The model loads it via `skill({ name: "tabbit-browser" })` or `/tabbit-browser`. |
-| `tabbit_browser_install` tool | Environment preflight: detects installed stable Tabbit editions, requires version `1.9.0` or newer, and verifies the `tabbit-cli` resident runtime. When Tabbit is missing or outdated, it starts a DSH background job that downloads the region-appropriate installer. |
-| `tabbit_plugin_update` tool | Plugin update check: compares the installed plugin version with the published changelog at most once a day, silently skips offline failures, and records a version the user declined. When a newer release exists, the skill loads with an update notice showing what the new version added. |
+| 能力 | 说明 |
+|---|---|
+| **`tabbit_browser`** | 用户使用真实的 Tabbit 内置的 CLI 模式执行网页读取和网页操作。 |
+| **浏览器代理的 `web_fetch`** | 把 dsh 的 `web_fetch` 改接到 Tabbit 上（dsh 自带的直连抓取器无 JS 渲染、无登录态、不走系统代理——在 fake-ip 代理环境下所有域名都会被它拒绝）。 |
+| **"网页标签"输入框提及 `@`** | 在 dsh Web UI 输入框敲 `@`，会列出本会话浏览器任务中打开的页面**以及用户浏览器里的全部标签页**，可以选中成为上下文。 |
+| **独立"页面读取"权限** | 完善的权限控制。因为与用户共享 cookie，dsh 访问 Tabbit 网页时会确认权限`pageAccess`（默认每会话询问一次）+ `intranetFetch`（web_fetch 访问内网目标默认逐次审批）。 |
+| **`tabbit_browser_install` 工具** | 环境预检：检测已装稳定版 Tabbit、校验 launcher 与 Runtime Service；缺装/过旧时以 dsh 后台任务下载 Tabbit。 |
+| **`tabbit_plugin_update` 工具** | 插件更新检查：每天最多查一次 npm 上的最新发布，有合适版本会离线静默安装； |
+| **`/tabbit-info` 命令** | 在 dsh 输入框内输入 `/tabbit-info` 可诊断：launcher、实例列表（含产品名）、生效实例及来源、权限设置、任务占用。 |
+| **`tabbit` skill** | 告知模型使用 Tabbit 的最佳实践。默认使用 Tabbit 自带的 `~/.agents/skills/tabbit/` 的官方 skill（随浏览器 Runtime 同步演进）。本插件内为兜底版本。 |
 
-## Installation
+## 安装
 
-### 1. Check or install DeepSeek Harness
+### 前置条件
 
-Check whether DSH is already installed:
+- 已安装稳定版 Tabbit Browser（[国际版](https://www.tabbit.ai) 或 [国内版](https://www.tabbit.com/)，≥1.9.0），且启动过（首次启动时注册 CLI launcher）；
+- Node ≥ 22.19、dsh ≥ 0.1.1-rc.2（dsh 安装：`npm install -g @deepseek-ai/dsh`）。低于 0.1.2-alpha.1 的宿主上，网页版会话里内置的 `web_fetch` 工具不可用（`tabbit_browser` 不受影响）。
 
-```sh
-dsh --version
-```
-
-If the command prints a version number, continue to the next step. If it is not found, install it for your operating system.
-
-#### macOS
 
-Install Node.js 20 or newer, then install DSH:
-
-```sh
-brew install node
-npm install -g @deepseek-ai/dsh
+### 安装 dsh-tabbit
+```bash
+dsh plugin --profile web add dsh-tabbit                 # npm 主路线
 ```
 
-#### Windows
-
-Install Node.js LTS in PowerShell:
+### 其他安装模式
 
-```powershell
-winget install OpenJS.NodeJS.LTS
+```bash
+dsh plugin --profile web add github:Tabbit-Browser/dsh-tabbit # npm 不可达时的回退
+dsh plugin --profile web add link:/path/to/dsh-tabbit   # 本地开发
 ```
 
-Reopen PowerShell after the installation, then install DSH:
+> 本包取代早期的 `tabbit-browser` skill-only 插件；npm 上的 0.2.x 版本也由本版本接续——0.2.x 用户经每日更新检查会收到升级提示，重跑安装命令即可原地升级。
 
-```powershell
-npm install -g @deepseek-ai/dsh
-```
+## 社区与交流
 
-Run `dsh --version` again to confirm DSH works.
+欢迎扫描下方二维码加入 **dsh-tabbit 开发者交流群**，交流使用心得、反馈问题与探讨新特性：
 
-### 2. Install the tabbit-browser plugin
+![dsh-tabbit 开发者交流群](assets/dsh-tabbit-developer-community-qr.png.jpg)
 
-```sh
-dsh plugin --profile web add dsh-tabbit
-```
+## 设置
 
-This installs the npm package (prebuilt tarball, CDN-served). If the npm
-registry is unreachable, fall back to the GitHub source:
+### 基本配置
+dsh Settings → tabbit，或 `$DSH_HOME/settings.yaml`
 
-```sh
-dsh plugin --profile web add github:Tabbit-Browser/dsh-tabbit
+```yaml
+tabbit:
+  instance: ""            # 显式指定 16 位大写 hex 实例 id（/tabbit-info 可列出）；通常留空即可
+  launcherPath: ""        # 默认自动发现：优先 ~/.local/bin/tabbit-cli，回退 tabbit-playwright；Windows 为 %LOCALAPPDATA%\Tabbit\LocalAgent\bin\tabbit-cli.exe
+  pageAccess: ask         # ask（每会话询问一次）| always | never
+  intranetFetch: ask      # web_fetch 访问内网/回环目标：ask（每会话每 origin 询问一次）| always | never
 ```
 
-### 3. Start DSH
-
-```sh
-dsh web
-```
+### 实例解析优先级
 
-## How it works
+当本机有多个 Tabbit 版本时的优先级配置
 
-After installation, the bundle automatically registers its skill provider. The model loads the skill via `skill({ name: "tabbit-browser" })` or `/tabbit-browser`. Before the first browser operation in a task, the skill calls `tabbit_browser_install`:
+1. `tabbit.instance` 显式设置；
+2. **正在查看 dsh-web 的 Tabbit 实例**（自动检测：页面加载时 client 插件向 `/tabbit/instance-hint` 打点，服务端由 loopback socket 对端进程沿父链匹配实例注册表的 `browserPid`；仅 macOS，非 Tabbit 浏览器打开时自然不命中）——"在哪个 Tabbit 里看，就在哪个 Tabbit 里执行"；
+3. 继承的 `TABBIT_PLAYWRIGHT_INSTANCE` 环境变量（嵌入形态：Tabbit 启动打包的 dsh 时注入自己的实例 id，即为权威通道）；
+4. 注册表自动选择（唯一在线实例；歧义时报带实例列表的引导错误。Windows 上注册表不可读时委托原生 CLI 自行选择）。
 
-- **`ready`** — a stable Tabbit edition at `1.9.0` or newer is installed and the runtime is running; the agent continues with the `tabbit-cli` workflow.
-- **`restart-required`** — the installed version is sufficient, but the `tabbit-cli` runtime is not running; the user is asked to restart Tabbit Browser once.
-- **`background`** — no stable edition is installed, or none reaches `1.9.0`; the tool starts a DSH background job that reads the operating system's configured region (macOS reads the system locale, Windows calls the system region API) and downloads the matching stable installer: the domestic build from `tabbit.com` for mainland China, or the international build from `tabbit.ai` for every other or unknown region. It selects the right Windows x64, macOS Apple Silicon, or macOS Intel package, saves it to the user's `Downloads` folder, reports download progress, and notifies the absolute installer path on completion.
+当前生效来源可用 `/tabbit-info` 命令查看（`execution instance: ... (via ...)`）。
 
-The environment check also:
+Full access 说明：dsh 的 `danger-full-access` 权限预设会给会话写入审批策略 `never`（dsh 定义为"自动拒绝一切询问"）。本插件的权限门检测到该覆盖时**自动放行**，不再发出注定被拒的询问——Full access 就是完全访问（该模式下 bash 本就不受限，单拦浏览器工具没有防御价值）。例外：部署级把审批默认配成 `never`（非会话覆盖）时公开 API 读不到，仍会被拒；无审批通道的组合（如某些 headless 编排）会得到指明 `tabbit.pageAccess: always` 出路的拒绝信息。
 
-- Treats the runtime as available when multiple Tabbit instances are running; the agent sets `TABBIT_PLAYWRIGHT_INSTANCE` from the CLI's hint instead of reporting the instance ambiguity as an unavailable runtime.
-- Diagnoses the DSH sandbox mode required to invoke the CLI on the current platform: Windows reports `cliSandboxMode: danger-full-access`, other platforms report `default`.
-- Caches a successful environment check per agent session and re-checks only after a Runtime/launcher failure or an installation change, via `refresh: true`.
+## 权限及安全
 
-## Requirements
+- 基于对用户数据安全和用户登录态权限的考量，设计了独立的 `pageAccess` 权限，该权限独立于文件系统/沙箱权限，默认需用户确认。
+- 授权记录：某工具调用成功后，本会话后续调用不再重复询问；失败不记入（失败后重试会再次询问）。
 
-- A stable Tabbit Browser at version `1.9.0` or newer: either the international `Tabbit` or the domestic `Tabbit Browser` — installing either one is enough. If it is missing or outdated, the plugin downloads the installer for you.
-- The current DSH profile provides `ctx.skills`, `ctx.tools`, and `ctx.jobs` together with the corresponding model tools.
-- `dsh-tool-jobs` provides background job control and completion notifications for the current agent.
-- The current DSH profile provides a Bash/Shell tool running on the same host machine as Tabbit Browser.
-- The shell's execution environment can reach the Browser-owned Runtime Service.
-- On Windows, DSH's `read-only` and `workspace-write` restricted tokens cannot write to the Runtime named pipe. The skill first runs the normal `tabbit-cli tasks` connection probe and requests no permission at all when it succeeds. Only when the Browser, launcher, and Runtime processes are all detected but the connection returns `BROWSER_RUNTIME_UNAVAILABLE` does it ask the user to switch the current DSH session to Full Permission — and it then stops the task immediately, without retrying or continuing browser operations.
+## 开发
 
-## Notes and limitations
+项目的开发和测试依赖本地的 [deepseek-harness](https://github.com/deepseek-ai/deepseek-harness) 环境。
+由于 `@deepseek-ai/*` 系列包在 npm 上的发布版本普遍滞后，不能直接作为依赖安装，因此需要先在本地构建好一份 harness 检出，再通过下方脚本将仓库内的 .dsh-harness 符号链接指向它（该路径已在 .gitignore 中忽略）。
+整个流程只需配置这一处，`tsconfig.json` 和 `package.json` 均无需改动。
 
-- Mainland China uses the domestic `tabbit.com` download source; all other regions use the international `tabbit.ai` source.
-- The background download reports progress and notifies the absolute installer path when it finishes. It never opens the `.dmg`/`.exe` automatically.
-- Development builds are not detected.
-- The plugin does not provide native browser tools such as `tabbit_browser_evaluate`.
-- If DSH's Bash runs in a sandbox such as E2B or a remote container that cannot access the local GUI browser, this skill cannot make Tabbit automation work there.
+Deepseek Harness 代码获取及安装（如果已安装请忽略）
+```bash
+git clone https://github.com/deepseek-ai/deepseek-harness && cd deepseek-harness && pnpm install && pnpm build && cd ..
+```
 
-## Development
+将本机的 Deepseek 环境路径修改并 ln 到本项目，将 `/path/to/deepseek-harness` 改为你的实际目录。
+```bash
+npm run link-harness -- /path/to/deepseek-harness   # 或设置环境变量 DSH_HARNESS_PATH
+```
 
-```sh
-npm test
-npm pack --dry-run
+安装和编译
+```bash
+pnpm install && pnpm build   # tsc → lib/
+npm test                     # 构建 + node --test tests/
 ```
 
-## License
+## 已知限制 / 路线图
 
-MIT
+- 收藏夹/书签的提及暂不支持。
+- 截图进上下文要求当前模型路由支持图像输入。
+- Windows 版本回归测试有限。