NewJev Browser Skill
A Playwright-based browser skill lets an agent run a bounded web task while Jev selects page actions.
Added to Jevfast
NewHow it uses Jev
The selected choice is checked against the offered criteria and resolved to a live observed target. The run loop handles REVIEW, BLOCKED and DONE, enforces its probability floor, executes supported actions and observes again.
What Jev decides
Illustrative source-defined Jev questionExample answers · not a recorded Jev response · Source ↗
Question 1 · action
YOUR APP
INSTRUCTION
Choose the single operation and target that best advances the goal. Complete visible required choices BEFORE scrolling. If any color is permitted and none is selected, choose an available color now. Compare clicking each specific target against scrolling. An informational help link does not select a configuration option.
STATE
State is JSON with `page` (current observation including visible text and observed targets) and `recentActions` (last 10 actions). The goal is included in question instructions. Representative page has visible text and no actionable targets or eligible scrolling; the chart shows the resulting control choices.
JEV · CHOICE
- WAIT
- BLOCKED
- REVIEW
- DONE
App workflow
- Observe
Collect page text, field values and recent action history; omit password, file and hidden inputs.
- Decide and act
Ask Jev to choose a page action, execute it through isolated Playwright Chromium, then observe again until the step limit or goal.
Why it is interesting
It combines a general typed decision model with a browser executor and explicit origin limits, while keeping each task goal narrow.