It converts HTML pages into Markdown, which reduces the token count while maintaining the same semantics, thus preventing the context limit from being reached.
It responds in JSON format, which was necessary when I developed this, as ChatGPT Actions only accepted JSON at the time. (I am not sure if this has changed now.)
The code is available here, and is quite simple, comprising only 60 lines of code.
What is the role of the proxy at https://chaction.csie.cool/? It appears to be a data siphon with no purpose.