That's one hell of a task. First question is how fine-grained your capability system will be. Both in terms of capabilities and who they are granted for. Not fine-grained enough and everything will need everything, e.g. access to various clocks could be used to DoS you or as a side channel attack. Unsafe memory access might speed up your image parsing but kills all safety. Similar problems with scope. If per dependency, forces library authors to remove useful functionality or break up their library into tiny pieces. If per function and module you'll have a hard time auditing it all. Lastly, it's a huge burden on devs to accurately communicate why their library/function needs a specific capability. We know from JavaScript engines, containerization and WASM runtimes what's actually required for running untrusted code. The overhead is just to large to do it for each function call.