jev-browser-skill

Chinese-language Codex skill adapted from jev-ultrafast that lets Codex hand multi-step Chrome tasks to Jev with preset inputs.

Agent skillbetaMIT

View on GitHub ↗

Overview

Jev Browser Skill by wanghai673 is a Codex skill, documented in Chinese, that lets Codex drive Chrome through Jev for multi-step browser tasks such as searching, filtering, filling and opening pages. It is adapted from Browser Use's jev-ultrafast runtime, adding preset inputs, continuous execution, configuration self-checks and result recording.

Its speed and cost come from design choices: Codex supplies text such as search terms in advance so no extra text-generation call is needed; page text and controls are read from the DOM rather than screenshots; and Jev performs a sequence of clicks, scrolls and page changes on its own before Codex responds, reducing handoffs. The README cites a 7.1 s Google Flights search from the upstream demo, and Jev's quoted 70-500 ms latency and $0.042 per million input tokens, noting that full tasks also include Codex, page-load and retry overhead. It supports macOS with Chrome, Python 3.12+ and uv, and is MIT licensed with upstream notices.

What the decision model decides

Type	Question	Answers
Choice	Which page action and target should be used next?(inferred)	—

How it works

Codex installs the skill and supplies goal and inputs
Skill opens Chrome and reads the DOM
Jev performs consecutive actions toward the goal
Results are recorded and returned to Codex

Strengths

Fewer handoffs between Codex and the browser
No extra text-generation calls for preset inputs
Reuses your Chrome

Caveats

macOS and Chrome only
Demo timing comes from the upstream project

Getting started

Ask Codex to install the skill from https://github.com/wanghai673/jev-browser-skill, or run `scripts/jev.py run --url ... --goal ... --inputs ...`.

At a glance

Type: Agent skill
Status: beta
Language: Python
Runtime: Codex skill
Has code: yes
Has tests: yes
License: MIT

Use cases and integrations

Agent Decisions340

Codex262

How curated lists describe it

This Codex Skill lets Codex drive Chrome through Jev to complete multi-step browser tasks from a goal description with preset inputs.
logicrw/awesome-jev-projects

Frequently asked questions

What does Jev decide?

Each click, scroll and page navigation in the task.

Where does typed text come from?

Codex supplies it in advance as preset inputs.

Which platforms are supported?

macOS with Chrome.

Related projects

jev-browse (0x7067)
Agent-plugin browser tool where Jev picks one operation and one row from a numbered page table; a small LLM only writes typed text.
ProjectbetaTypeScript
jev-browser-control
Chrome extension plus MCP server so Claude or Codex can drive your real Chrome, with Jev picking each click in about half a second.
MCP serverbetaTypeScript
jev-browser-use
Codex browser skill where Jev handles navigation, clicks, toggles and scrolling while Codex handles text input and verification.
Agent skillbetaTypeScript
jev-browser-use
Local browser automation workbench with an Edge/Chrome sidebar and Codex MCP sharing one Jev-driven task engine, with user confirmation.
MCP serverbetaTypeScript
jev-max
Jev browser agent that pierces shadow DOM and iframes, adds canvas fallback and full actions, with one Jev request per cycle.
MCP serverbetaPython
jev-mcp (legostin)
MCP server that lets Claude Code or Codex drive real Chrome with Jev picking elements and checking steps, plus safety gates and traces.
MCP serverbetaTypeScript