Currently, Sandwich guess the list of URLs of resources to prefetch from the main document by cross referencing devtools request's initiator field and trace events. How ever the initiator field is know as being unreliable.

Because of this, Sandwich needs to deal with issues where the list of URLs discovered by the HtmlPreloadScanner might be flaky despite the fact that the fetched main resource were the same.

The idea is to implement a tool similar as cachetool or content_decoder_tool to parse a document it would received on the stdin and output discovered URLs in the stdout.