Jev Browser Use
Codex skill: Jev clicks and scrolls on accessibility text; Codex types, interprets visuals, and verifies. Author claims about 5 to 10x faster browser ops at EZCollegeApp.
Overview
jev-browser-use (github.com/wy-coliney/jev-browser-use) installs as a Codex skill or plugin. Jev receives accessibility text from the existing Computer Use connection, chooses clicks, navigation, toggles, and scrolls, then hands back to Codex for typing, images, and final verification. README attributes roughly 5 to 10 times faster browser operations in EZCollegeApp workflows as an author claim. Provider config supports TypeSafe or OpenRouter Decisions with keys kept in a local dotenv file named by ~/.config/jev-browser-use/config.json.
Problem: Codex Computer Use sessions spend full model turns on repetitive clicks and scrolls when the host model plans every micro-action in prose.
Built for: Codex (and Claude Code skill) users who want Jev to drive navigation and controls over accessibility text while the main agent types, judges visuals, and verifies outcomes.
First indexed on Jev Directory: 2026-09-26
Creator and team
- Name
- Coliney
- Organization
- EZCollegeApp
- Handle
- @wy-coliney
How Jev is used
- Role in the product flow
- Browser action Choice loop inside Codex Computer Use without a separate driver
- Primitives
- Choice
- State in
- Accessibility text and control inventory from the connected Chrome or in-app browser session, plus the active goal from Codex.
- Decision out
- Next browser operation executed through Computer Use; Jev completion signal returns control to Codex for verification.
- Codex sets a browser goal and invokes the skill
- Observe controls through the existing Computer Use channel
- Jev selects the next action from structured candidates
- Browser executes; session retains progress across handoffs
- Codex verifies page state before trusting completion
This is the Codex plus Jev browser skill lane, distinct from browser-use-jev-ultrafast, which is Browser Use's Python hybrid with DOM tables and a tiny LLM only on TYPE_TEXT. Here the split is intentional teamwork: Jev burns cheap decisions on controls; Codex keeps judgment on text and visuals. sac-y-jev-cu is the parallel Codex skill with explicit dry-run and risk gates on sensitive ops. moritzkremb-jev-voice-browser adds microphone debouncing. Install with npx skills add wy-coliney/jev-browser-use; no extra npm driver beyond Node 22+ and an existing Computer Use browser connection.
Sourced performance claims
- README author claim: about 5 to 10 times faster browser operations in EZCollegeApp workflows (server-side processing excluded).Source: github.com/wy-coliney/jev-browser-use README
- README cost table: Jev 1.13 at $0.042 per 1M input tokens versus GPT-5.6 Terra at $2.00 per 1M input (48x) as of September 18, 2026.Source: github.com/wy-coliney/jev-browser-use README
- Public GitHub repo wy-coliney/jev-browser-use had about 545 stars and 33 forks when this listing was drafted.Source: GitHub API September 2026
Features and stack
Features
- npx skills add wy-coliney/jev-browser-use for Codex or Claude Code agents
- Optional codex plugin marketplace path documented in README
- Shared ~/.config/jev-browser-use/config.json across install methods
- TypeSafe or OpenRouter provider docs under skills/jev-browser-use/references
- INSTALL.md agent-driven setup without pasting keys into chat
Stack
- Node.js 22+
- Codex Computer Use
- TypeSafe or OpenRouter Jev
- Agent skills CLI
Pricing: MIT skill; Jev input pricing per README table; Codex usage bills separately.
Links
FAQ
- Does this add a second browser driver?
- README: it uses your existing Computer Use connection to Chrome or the in-app browser; no extra driver npm dependencies.
- Can I use OpenRouter instead of TypeSafe?
- Yes. provider-configuration.md documents typesafe versus openrouter keys; they are not interchangeable even though both expose Jev.
- Is the 5 to 10x speedup a benchmark guarantee?
- No. README labels it an approximate author claim from EZCollegeApp workflows with server-side processing excluded.
- Does Jev see screenshots?
- README how-it-works: Jev receives accessibility text; Codex handles visual interpretation and typing.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
- Jev vs LLM classification
When to gate with System One probabilities instead of asking a chat model to label things.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.