browser

Configures booking tasks and distraction levels for Jev Browser and measures booking accuracy.

465|41|Updated Aug 4, 2026
One-click install
npx skills add https://github.com/autonomous-ai/openharness --skill browser-autonomous-ai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: browser
Source: https://github.com/autonomous-ai/openharness/tree/main/store/agents/jev-browser/skills/browser
Command: npx skills add https://github.com/autonomous-ai/openharness --skill browser-autonomous-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Testing how well an autonomous agent navigates a website under realistic conditions is hard to do repeatably. This Skill lets you define booking tasks and tune a distraction dial on the Jev Browser test site, then measures exactly how many bookings come out right and where layout shifts break the agent. ## Core Features & Use Cases - Task Crafting: Write booking tasks in site.json that exercise the full funnel: search, results, details, and review pages. - Distraction Dial: Add decoy buttons, cookie walls, pop-ups, and sponsored rows that shift layouts, scaled from 0 (clean) to 1 (everything moves). - Accuracy Measurement: Run check.mjs and read .harness/verdict.json for bookings completed, share exactly right, steps taken, and clicks lost to layout shifts. - Use Case: Set the distraction dial to 0.6, run the verification, and report at which distraction level the share of correct bookings falls below 9 in 10. ## Quick Start Write a set of booking tasks in site.json, set the distraction dial to 0.25, and run the verification to report how many bookings came out exactly right.

Frequently Asked Questions about browser

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I test an autonomous web agent on a booking site?

Define booking tasks in site.json covering the search, results, details, and review pages, then run node check.mjs. The verdict.json file reports bookings completed, the share exactly right, steps taken, and clicks lost to layout shifts.

How do I simulate pop-ups and layout shifts when testing a web agent?

Set the distraction value in site.json from 0 to 1. Higher values add decoy buttons, a cookie wall, pop-ups between reading and clicking, and a sponsored row that shifts results so clicks land on the wrong element.

What makes a booking count as correct in Jev Browser?

A booking is right only if it matches the exact flight the task describes, such as cheapest, earliest, or latest with nonstop when asked, plus the correct name, email, and bag count with no extras like insurance or Flex upgrade.

Why does the agent book the wrong flight when distraction is high?

At high distraction a sponsored row pushes results down after the page is read, so the click lands on the row above the intended one. This wrong booking is only detectable by carefully checking the next page.

What does the MOCK badge in the pane mean?

The MOCK badge means the offline stand-in is driving the run instead of the real model. You should mention this when reporting results, since the numbers reflect the mock rather than actual model behavior.