If the kernel identifies that a file you’re trying to execute has a known sequence of magic bytes, it will happily execute it. However, you might want to run certain formats without needing to pass them to another tool. In as much as these are usually binary tools, you can prepend a magic-bytes preamble (or use an existing sequence, if your binary format already declares one) and then associate that with a certain loader in the running kernel using binfmt_misc. Access is usually via procfs (currently /proc/sys/fs/binfmt_misc/register). Cross-execution with a virtualized process is a common use-case for this.

Note that this doesn’t really apply to text files (it’s all data) because you’d need to inject gibberish at the top of your file to use this functionality when using a shebang (which is meant for this case) should work just fine.

This is a walkthrough of how to do this for compiled Python or LUA source-code: https://twdev.blog/2024/01/docker_multi_platform

Since both of the example compilation formats include magic bytes, you can just write a tool to acquire those from libraries (at least with Python) and call binfmt_misc in one step.

You can also write configs to be automatically loaded at boot. There appears to be no way to list registered formats, with the only recourse being to sift the config paths (probably empty by default) or to check your kernel config to see which supported formats were included in that build.