How it uses Jev
TipTour pairs a typed task and recent action history with locally detected screen controls. Jev chooses a target and click kind, and also judges completion and whether the needed control is absent; the app validates and executes the selected click.
What Jev decides
Choose how to act on the selected screen elementExample answers · not a recorded Jev response · Source ↗
Question 1 · kind
YOUR APP
INSTRUCTION
How should that element be acted on?
STATE
A screen task with the current task, step number, already-completed actions, and detected screen elements. The action-kind question is fixed regardless of the dynamic target list.
JEV · CHOICE
- click
- double_click
- right_click
App workflow
- Observe local controls
Refresh locally detected screen targets and convert them to stable candidate IDs and descriptions.
- Build a grounded request
Send the task, step number, completed actions, and candidate descriptions with completion, absence, target, and click-kind questions.
- Validate the decision
Map Jev’s returned target and click kind to the candidate pool and stop if completion or no-target conditions apply.
- Execute and observe again
Run the allowed click through the shared action engine, record history, then refresh targets for the next step.
Why it is interesting
The request binds candidate IDs and descriptions to the same screen state, and the loop uses an explicit `none` option and bounded stopping conditions.