dsh_use_browser
Identifieddsh_use_browser
English | 中文
Codex-style local browser automation for DeepSeek Harness: open pages, interact by @e references, extract content, screenshot, and run page JS — with two backends: managed headless Playwright (fast, invisible) or CDP connected to your own Chrome (visible, and loads Chrome extensions that headless cannot).
Compatibility
This revision requires DeepSeek Harness core 0.1.2-alpha.5 or later within the 0.1.x line. It uses the Settings service API introduced in that core release.
Capabilities
| Tool | Purpose |
|---|---|
browser_open | Open a URL → page summary + interactive @e refs |
browser_click / browser_type | Interact by @e ref |
browser_extract | Page text (body or selector) |
browser_screenshot | Save a PNG → pair with analyze_image (the vision plugin) to "see" the page |
browser_eval | Run JavaScript in the page |
Requirements & notes
- Runtime: Node 22+;
playwright ~1.61.0is the only runtime dependency and is resolved from the deploying harness (peer of the deployment, not bundled). - One session shares a single browser page:
browser_opennavigates the current page; later calls act on that page. - Screenshots land in
/tmp/dsh-browser-*.png— pair withanalyze_image(the vision plugin) to "see" the page, since the model context stays text. - Headless mode cannot load Chrome extensions; use
mode: cdpfor extensions or for visible, watchable automation.
Install into an official DeepSeek Harness (no repo modification)
The package declares dsh.bundle, so dsh plugin add mounts it automatically:
dsh plugin --profile web add file:/path/to/dsh_use_browser/packages/use-browser
# or, once published
dsh plugin --profile web add @deepseek-ai/dsh-tool-use-browser
[WARN] Issues with peer dependenciesis expected — peers come from the deployment. Restart the harness; the sixbrowser_*tools then appear in every preset.
Configuration (Settings → browser)
| Field | Default | Meaning |
|---|---|---|
mode | playwright | playwright = headless Chromium; cdp = connect to your Chrome |
cdpUrl | http://localhost:9222 | CDP endpoint for mode: cdp |
headless | true | Headless for mode: playwright |
timeoutMs | 30000 | Default operation timeout |
Start a Chrome with extensions for CDP mode:
open -na "Google Chrome" --args --remote-debugging-port=9222 --user-data-dir=/tmp/dsh-chrome
Requirements
playwright(installed with the plugin) + a browser binary:npx playwright install chromium(or rely on an existing~/Library/Caches/ms-playwrightcache; the pinned~1.61.0matches chromium-1228).- CDP mode additionally needs your own Chrome running with
--remote-debugging-port.
Development
- Backend-agnostic: one session manager serves both modes; the tools never change.
- Screenshots land in
/tmp/dsh-browser-*.pngand pair with the vision plugin'sanalyze_image. - Workflow guidance ships as
SKILL.md.
License
MIT
Comments
Loading…
From the same category
DeepSeek Harness plugin for Reactive Resume: bridges your resumes and job applications into a Harness session over MCP.
★ 41.7k
↓ 215/wk
MIT
Aug 24, 2026
dsh plugin --profile web add dsh-plugin-reactive-resumeby dsh-market
The plugin market inside DeepSeek Harness — browse, search, one-click install · DSH 可视化插件市场
★ 10.1k
↓ 361.7k/wk
MIT
TypeScript
Nov 1, 2026
dsh plugin --profile web add dshmarketby Tencent
Let AI agents use your real, logged-in browser without interrupting your work. CLI + extension for browser automation across any shell-capable AI agent.
★ 9.9k
↓ 12.3k/wk
MIT
TypeScript
Nov 1, 2026
dsh plugin --profile terminal add @wxg-prc-cpg/browser-skill-dsh-pluginby yjh051108
dsh-routing-suite — injector + router-standard kit: install the runtime injector first, then the task-aware reasoning-mode router preset (measured P1-P23).
★ 7k
MIT
JavaScript
Sep 18, 2026
dsh plugin --profile web add @dsh-external/dsh-super-injectorby superdesigndev
OpenRouter for agent tools. Join community here: https://discord.gg/6mQYYfFMAn
★ 6.9k
NOASSERTION
Python
Nov 1, 2026
dsh plugin --profile web add treg-dshby Q00
Agent OS: the agent gets smarter on its own. We just hold the line: Interview-gated, staged evaluation, budgeted evolution loop. MCP server, 14 runtimes: Claude Code, Codex CLI, Gemini CLI, OpenCode,
★ 6.3k
MIT
Python
Oct 30, 2026