1. Start Pinchtab in headless or headed mode.
2. Navigate to the target URL using the `/navigate` endpoint.
3. Capture the accessibility tree using the `/snapshot` endpoint, optimizing for token usage with `format=compact` and `filter=interactive`.
4. Identify the target element using the snapshot data (refs).
5. Perform an action on the element using the `/action` endpoint (e.g., click, type).
6. Repeat steps 3-5 as needed to complete the task.
7. Extract relevant text using the `/text` endpoint or download files using the `/download` endpoint.
8. Stop the Pinchtab instance when the task is complete.