It is fairly straightforward to wrap Claude Code in an OpenAPI Compatible HTTP API. It seems like any prompt could also be interpreted as a coding problem - for example, instead of wrapping like I did, you could watch edits to an output file while Claude Code reads an input file. How do you think Anthropic will stop this, with their new rules - recognizing and blocking specific prompts?
I find myself wanting to build simple, minimal/no-styling versions of web-pages for agents, and then full versions for people, built this little tool to help with that, thoughts?