Marketplace for trusted AI agent skills

Skill Detail

Jet Browser

Install, verify, integrate, or troubleshoot the Jet Browser WPE WebKit runtime for isolated agent browser sessions with ordered JSONL automation, native input, screenshots, and reproducible container checks.

Browser AutomationCodex
Browser Automation Codex Published
⭐ 2 GitHub stars
COPY SKILL INSTRUCTIONS (OPTIONAL)
npx skills add agentskillexchange/skills --skill jet-browser Copy
Uses the third-party skills CLI, not an ASE-owned installer. Check your agent’s compatibility. This copies instructions; complete the upstream tool setup below separately.
View source Documentation
At a glance
Tools required
Docker for the standalone runtime; Docker and Node.js 24+ for source checkout and SDK workflows.
Install & setup
codex plugin marketplace add masakaai/jet-browser –ref main && codex plugin add jet-browser@masaka
Author
masakaai
Last updated
Oct 13, 2026
Quick brief

# Jet Browser

How it works

What this skill actually does

Jet Browser is an open-source, local or self-hosted WPE WebKit runtime for AI-agent browser work. Use this skill when a repository needs an isolated browser process with ordered JSONL commands, native pointer and keyboard input, screenshots, explicit profile or download mounts, and a reproducible container acceptance check. It keeps model choice, reasoning, credentials, and orchestration in the calling project. The runtime is WebKit rather than Chromium: do not assume Chrome DevTools Protocol, Chrome extensions, or Chromium-only behavior. It is also not an anti-bot service, hosted proxy network, or replacement for a user’s already-running personal Chrome profile.

## Installation

### OpenAI Codex Plugin

codex plugin marketplace add masakaai/jet-browser --ref main
codex plugin add jet-browser@masaka

### Direct repository checkout

git clone https://github.com/masakaai/jet-browser.git
cd jet-browser
npm ci
npm run standalone

The standalone check requires Docker and Node.js 24 or newer. Treat its JSON result as the acceptance gate: local-page loading, JavaScript marker verification, native text input, screenshot capture, network isolation, and clean teardown must all pass. For source changes, also run `npm test` and `cargo test –all-targets`.

## Integration guidance

– Read `docs/demo.md` for the container lifecycle and `docs/agent-tools.md` for the model-facing declarations.
– Reuse the versioned tool definitions exported by `sdk/tools.mjs`; do not invent command names or response fields.
– Send one JSON command per line, consume one JSON response per line, preserve ordering, and close the session in guaranteed cleanup.
– Keep one isolated browser session per container. Mount profile and download paths explicitly, and do not pass unrelated host credentials into the container.
– Use semantic evidence for target selection and screenshots for visual verification when both are available.

Benchmark claims must retain their published scope. The repository’s benchmark and technical report describe a pinned offline fixture; they do not establish public-site success, stealth, CAPTCHA handling, model accuracy, or long-running stability.