by woniu9524
Drive your real Chrome from any AI agent — logged-in sessions, real fingerprint, live tabs. CLI-based, works with OpenCode, Pi, or any Harness.
# Add to your Claude Code skills
git clone https://github.com/woniu9524/open-web-bridgeGuides for using ai agents skills like open-web-bridge.
open-web-bridge is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by woniu9524. Drive your real Chrome from any AI agent — logged-in sessions, real fingerprint, live tabs. CLI-based, works with OpenCode, Pi, or any Harness. It has 53 GitHub stars.
open-web-bridge's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/woniu9524/open-web-bridge" and add it to your Claude Code skills directory (see the Installation section above).
open-web-bridge is primarily written in JavaScript. It is open-source under woniu9524 on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh open-web-bridge against similar tools.
No comments yet. Be the first to share your thoughts!
Unlocks once the catalog security scan passes (runs nightly).
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
An agent that opens its own clean browser can only reach what is public.
An agent driving your browser can read the article behind your subscription, check the dashboard you are already signed into, and finish the flow that needs your identity. That is the whole point.
npm i -g open-web-bridge # CLI + extension files + agent skill, one command
owb setup # walks you through the one step you must do yourself
owb # self-check — two ✓ and you are ready
Then just tell your agent: "open Hacker News and summarize the top three stories."
AI agent Claude Code · Codex · Kimi Code · anything with a shell
│
│ owb <command> the CLI is the entire integration surface — zero client config
▼
local daemon Node.js on 127.0.0.1:43917
│
│ WebSocket
▼
MV3 extension installed in the browser you actually use
│
│ Chrome DevTools Protocol
▼
your live tab your cookies · your session · your fingerprint
Two deployment shapes:
| What you get | |
|---|---|
| 🔍 Semantic snapshots | read_page assigns stable @eN refs to interactive elements so click/fill/screenshot reference them directly. since_last returns only what changed — the difference between a session that survives and one that drowns in tokens. article mode extracts body text as clean markdown |
| ⏳ Waiting primitives | wait_for on a selector, text, URL or network idle, instead of polling with evaluate |
| 🖱️ Real mouse + visible cursor | mouse_click dispatches genuine CDP Input events (isTrusted), with an in-page bezier cursor animation so a user sitting beside you can see what is happening |
| 🤝 Human handoff | handoff / wait_user give the tab back to you for a captcha or a QR login, and the agent resumes automatically once you are done |
| 📎 The awkward interactions | download/upload (uploads go through an in-page DataTransfer, so no filesystem access is needed), print_pdf, list_frames + evaluate frame_pattern for iframe-targeted evaluation |
| 🌍 Environment emulation | emulate/emulate_reset covers viewport, network throttling, geolocation, timezone, locale, permissions and UA in one call |
| 📡 Network capture | Full requests and responses including headers and bodies, plus get_initiator for the call stack that produced a request |
| 🎞️ Session recording (HAR) | record_start/stop writes standard HAR 1.2 (timing, WebSocket, actively collected bodies, url/resource_type filters, multi-tab merge), with console archives, storage change streams and a navigation screenshot timeline; daemon_task_end files the HAR automatically |
| 🔁 HAR processing | daemon_har_to_replay (→ python/curl/node replay scripts, with dynamic signature headers marked as placeholders), daemon_har_diff (drift between two recordings), daemon_har_assert |
| 🧩 Tasks and workflows | daemon_task_begin/end for archiving and tab grouping; daemon_workflow_save/run turns a working flow into deterministic replay |
| 🔐 Per-site session store | daemon_state_save/load <name> saves and restores a login (cookies + localStorage + IndexedDB) |
| 🛠️ Debugging and analysis | On demand: hook presets (xhr/fetch/crypto), breakpoints and call-frame inspection, script patching, offline function verification, TLS fingerprint replay |
Requirements: Node.js ≥ 18 and a Chromium-based browser (Chrome or Edge).
Install open-web-bridge for me. Do these in order and tell me the result of each:
- Run
npm i -g open-web-bridge- Run
owb setup, read me the "install the browser extension" step, and wait until I confirm I've done it- Run
owband confirm both the daemon and the extension show ✓. If the extension is not connected, tell me to click the extension icon in my browser toolbar and check its status- Run
owb skill install, then tell me to start a new sessionAfter that you can drive my browser with the
owbcommand;owb helplists everything.
npm i -g open-web-bridge # CLI + extension files + skill, in one command
owb setup # walkthrough: extension path, skill install, self-check
owb setup tells you how to install the extension. That is the one step you
have to do yourself — the extension has to go into the browser you actually use,
because that is where your sessions are, which is the entire point of this
project. No command line can do it for you.
Then run owb once as a self-check; two ✓ means you are ready. Finally
owb skill install installs the skill for every agent it detects — Claude
Code, Codex CLI, Kimi CLI, Qwen Code, iFlow CLI, Gemini CLI, OpenCode
(~/.config/opencode/), Cursor (project-level only), and the shared
~/.agents/skills/. Detection is not a fixed list: any ~/.<tool>/skills/ or
~/.config/<tool>/skills/ that already exists counts too, so an agent released
after this version still gets the skill without waiting for an owb release.
Use --to kimi,codex to pick agents yourself — any name works, one owb has
never heard of installs to ~/.<name>/skills/ — --project to install into the
current project only, or --dir <path> for anywhere else. --link symlinks the
package's skill source instead of copying it, so a later
npm i -g open-web-bridge@latest updates every installed copy on its own — it
falls back to a copy wherever symlinks are unavailable. For an agent with no
skills directory at all, owb skill print writes SKILL.md to stdout, ready to
paste into whatever rules file it does read.
npx skillsThe skill is a plain Agent Skill, so npx skills —
which knows 75+ agents — can install it straight from this repository:
npx skills add woniu9524/open-web-bridge -g -y
That installs the skill and nothing else; the agent bootstraps the rest
(npm i -g open-web-bridge, then owb setup) the first time it reads it.
Later, owb update check compares your install against npm and prints the
upgrade steps when a newer version exists — the skill also teaches agents to
run it at the end of a session, so you hear about updates without asking.
Progressive disclosure, and skill install copies the whole directory:
owb/SKILL.md the trunk — enters context on every trigger
owb/references/commands.md arguments for all 80 commands
owb/references/recipes.md longer per-task procedures
owb/references/debugging.md capture / HAR / hooks / breakpoints / reverse engineering
owb/references/field-notes.md real-world oddities, indexed by symptom
owb/references/relay.md relay mode setup
Only SKILL.md is always loaded; the agent reads the rest when it needs them.
Other agents can simply concatenate these markdown files into their rules or
system prompt — there is no proprietary format.
Tell your agent something like "open Hacker News and summarize the top three stories for me". Or drive it yourself:
owb open https://example.com
owb page # semantic snapshot: interactive elements numbered @eN
owb click @e1 # act by number
owb help # all commands
Running from source (development): `