User repository on GitHub
- Forks
- 60
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Tools & CLIs
Jev-cu
An experimental Codex skill and JavaScript runtime that has Jev pick macOS Accessibility elements and actions from text candidates, while local policy gates hold sensitive operations for human confirmation.
- Stars
- 622
- Language
- JavaScript
- License
- MIT
- Reading time
- 2 min
- Published by
- Community
- Author
- Sac-Y
- Last verified
- Sep 2026
Works forJev for Decision gating
Used inJev for Software & DevTools
Highlights
- Jev chooses the element, action, completion state, and risk from interface text candidates; screenshots are never sent.
- Runs in dry-run by default, and delete, send, pay, authorize, upload, captcha, install, and system-settings actions stop at confirm.
- The app allowlist lives in scripts/policy.mjs, and adding an app requires editing it explicitly.
- Interface text is treated as data rather than instructions, and the skill does not bypass logins, paywalls, or captchas.
- Installs into ~/.codex/skills/jev-cu with npm run install-skill, and npm run p0 measures element-selection accuracy on recorded AX snapshots.
Quickstart
bash
npm run install-skill
echo 'TYPESAFE_API_KEY=YOUR_API_KEY' > .env.local
npm testWatch out
MIT-licensed. Needs Codex desktop's cua_repl runtime and a TYPESAFE_API_KEY; it is experimental, its execution-policy limits require review, and the offline P0 evaluation still calls Jev.
Questions about this entry
Short answers drawn from the linked project or page.
No screenshots are involved at all. Jev receives only text candidates read from the macOS Accessibility layer and returns the element, action, completion state and risk; Codex Computer Use does the reading and the executing. Nothing visual is sent to the model.
Local policy in scripts/policy.mjs does, not Jev. Delete, send, pay, authorize, upload, captcha, install and system-settings actions halt at confirm, and it runs dry-run by default. The app allowlist must also be edited explicitly before a new app is allowed.
Not as a standalone library. The decision loop has to run inside Codex desktop's cua_repl runtime and the driver is constructed around that runtime object. You also need a TYPESAFE_API_KEY, and the README says English goals produce the most accurate results.
No. npm run p0 replays recorded accessibility-tree states and measures element-selection accuracy, but it still calls Jev, so it costs tokens and is not a free check. npm test is the API-free one, and no P0 score is published to compare against.
Run npm run install-skill to copy it into ~/.codex/skills/jev-cu, then start a new session. Put TYPESAFE_API_KEY in .env.local or the environment. The loop itself has to run inside Codex desktop's cua_repl runtime, and English goals give the most accurate results.
Jev receives only text candidates read from the macOS Accessibility layer and returns the element, action, completion state and risk. Codex Computer Use reads the interface and executes, and a screenshot is never sent to the model.
npm run p0 replays recorded accessibility-tree snapshots and measures element-selection accuracy, but it still calls Jev, so it costs tokens and is not a free check. npm test is the API-free one, and no P0 score is published to compare against.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Jev driving real Chrome from Grok Bot
Hacker News: Cua asks which computer-use steps need a frontier model and which are fast decisions.
frabonacci95 pointsShow HN: CUA-S1 – A System One Model for Computer UseRead the thread on Hacker NewsX: Voice-controlled computer use on a Mac
X: A computer assistant that listens
More Jev resources like this
A Chrome side panel that turns the current page's WebMCP tool schemas into typed Jev questions, predicting the tool call, its arguments, confidence, and latency as you type.
Jev Tools & CLIs#community#javascript#chrome-extension
Community
sdras/jev-webmcp-extensionJavaScript
sdras/jev-webmcp-extensionJavaScriptA warden for Claude Code and Codex CLI that keeps an append-only ledger of what an agent did, refuses a done claim when no check passed after the last edit, and uses optional Jev Noul judgments.
Jev Tools & CLIs#community#typescript#claude-code
Community
qkal/CannyTypeScript
qkal/CannyTypeScriptA security hook for coding agents that asks Jev three typed questions before every tool call and scans tool results for prompt injection, with adapters for Claude Code, Codex, Copilot CLI, Gemini CLI, Cursor, pi, OpenCode and ACP.
Jev Tools & CLIs#community#javascript#agents
Community
leepokai/jev-guardJavaScript
leepokai/jev-guardJavaScriptA Claude Code Stop hook that reads the turn transcript locally and spends one four-question Jev call only when files changed with no passing check since, blocking unverified done claims.
Jev Tools & CLIs#community#javascript#claude-code
Community
valentynkit/jev-belayJavaScript
valentynkit/jev-belayJavaScriptClaude Code, Codex, and pi plugin plus MCP server and npm library that hands steps needing no text output to Jev: one batched call answers typed questions about a state, and an opt-in PreToolUse gate can only deny or ask.
Jev Tools & CLIs#community#javascript#claude-code
Community
shitianfang/jev-useJavaScript
shitianfang/jev-useJavaScriptCLI and GitHub Action that turn project-defined semantic preferences from AGENTS.md into Jev Noul or Choice checks over code changes, then map each result to an advisory or blocking outcome. Checks can be tuned on labeled diffs.
Jev Tools & CLIs#community#javascript#cli
Community
doeixd/jev-prefJavaScript
doeixd/jev-prefJavaScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.