Architecture for layering project-wide AI instructions with user-specific "personal" rules

I am not really an expert for this topic, but assuming there is currently no standard for local or individual AGENTS.md overrides, I see two possible approaches:

1. One could add a section like this:

## Individual rules for submodules
- when there is an AGENTS.local.md file in the root folder of a 
submodule, use the contained instructions to override the information

to the AGENTS.md. How well this works may depend on the specific agents and their capabilities.

2. In case that does not work well, because your AI agents ignore the instructions to process a second file, you could combine the combination from two files into one by yourself: write a script which merges the global information from the global AGENTS.md and a AGENTS.local.md into a combined AGENTS.md file for the submodule (I guess when you structure both files correctly, you can just concatenate them - but if not, you could merge information on a "per section" basis. Make sure the AI agents see only this combined file when they scan your source code folder.