Tabbit Agent-Browser Skill

A Tabbit browser skill for agents that support Skills.

• Tabbit is a Chromium-based browser.
• This skill helps an agent discover Tabbit's DevTools / CDP endpoint and hand browser control to agent-browser.
• It does not promise that chrome-devtools MCP will be automatically retargeted to Tabbit.
• Once connected, page opening, navigation, clicking, script execution, and extraction should follow the normal agent-browser workflow.
• It searches for DevToolsActivePort in Tabbit first and Tabbit Browser second.
• It outputs the live DevTools WebSocket details and drives the current Tabbit instance through agent-browser --cdp <wsEndpoint>.

Prerequisites

• python3 is installed locally
• node / npx is installed locally, or agent-browser is already runnable
• Tabbit is installed and open
• Remote debugging is enabled in tabbit://inspect/#remote-debugging

Install with npx skills add

Recommended:

npx skills add Tabbit-Browser/Tabbit-Devtools-Skill

Install only this skill:

npx skills add Tabbit-Browser/Tabbit-Devtools-Skill --skill tabbit-devtools

Restarting the agent after installation is the safest option.

Recommended: install agent-browser too

If your agent host supports skills, install the official agent-browser skill as well:

npx skills add vercel-labs/agent-browser

Even without the extra skill, the wrapper in this repo can launch it as long as agent-browser or npx agent-browser is available locally.

If your environment uses a different command name, set:

export AGENT_BROWSER_BIN="your agent-browser launch command"

For example:

export AGENT_BROWSER_BIN="npx --yes agent-browser"

What is agent-browser

agent-browser is a browser automation CLI for agents.

Its role in this repository is simple:

• this skill discovers the currently available Tabbit wsEndpoint
• the wrapper injects --cdp <wsEndpoint> into agent-browser
• actual page operations are executed by agent-browser

For the full command surface, refer to the official README:

• GitHub: vercel-labs/agent-browser
• Common agent workflow: Usage with AI Agents

Install in Codex

In a Codex conversation:

$skill-installer install https://github.com/Tabbit-Browser/Tabbit-Devtools-Skill/tree/main/skills/tabbit-devtools

Or:

$skill-installer install-skill-from-github --repo Tabbit-Browser/Tabbit-Devtools-Skill --path skills/tabbit-devtools

If you want to install from a terminal:

mkdir -p ~/.agents/skills
python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
  --url https://github.com/Tabbit-Browser/Tabbit-Devtools-Skill/tree/main/skills/tabbit-devtools \
  --dest ~/.agents/skills

Manual install

mkdir -p ~/.agents/skills
ln -sfn /path/to/Tabbit-Devtools-Skill/skills/tabbit-devtools ~/.agents/skills/tabbit-devtools

Restart the agent after installation.

How it works

This skill is best understood as a "connect Tabbit through a Chromium-compatible DevTools endpoint and hand control to agent-browser" layer.

It tells the agent:

• where to look for DevToolsActivePort
• how to derive Tabbit's port and wsEndpoint
• how to hand the rest of the work to agent-browser

It does not implement its own bridge or browser automation layer. Actual page operations belong to agent-browser.

The default search order is:

• ~/Library/Application Support/Tabbit/DevToolsActivePort
• ~/Library/Application Support/Tabbit Browser/DevToolsActivePort

Only when neither location exists should the environment be treated as missing a usable Tabbit browser.

License

This repository is released under the MIT License.