How actually playwright-interactive skill returns coordinates for next step action?

does the `codex.emitImage()` do this? or it just add it to model context, the model gpt5.4 will auto reasoning against the image and output target x/y value in the context and the model will take the next step to click based on x/y coordinates?