01
Identify what can actually change
The current implementation draws the image to a canvas and exports PNG. It does not provide a palette-reduction optimizer, a selectable PNG compression level or a target-file-size field. The generic Quality slider does not turn PNG encoding into a lossy quality adjustment.
Width is the meaningful size control. Height follows the same proportions. Reducing a 2000-pixel-wide screenshot to 1000 pixels removes detail, even though the resulting PNG encoding itself is lossless. Inspect small text and one-pixel lines before deciding that the resized version meets the purpose.
02
Check transparency and the saved bytes
Open the downloaded file against both a light and dark background if transparency matters. Check edge pixels and any labels. Compare its actual size with the source; if dimensions are unchanged and the output is larger, keeping the original can be the better result.
This route is not an archival preservation tool for all PNG metadata or animation. Do not assume an animated PNG retains every frame through a canvas image conversion. Keep the original and use a dedicated animation or PNG optimizer when those properties are required. Very large images also consume much more decoded memory than their compressed file size suggests.