Single-folder vs Rule-per-folder

[TomLucidor]

This is a bit of a nitpick, but something worth considering. There are two ways of writing code files

1. Have .cursor/rules (or whichever IDE) capture a microcosm of what the file structure should look like, and then glob accordingly. This way we can have one single "power folder" that can be easily inserted into ANY new projects.
2. Have rule file for every folder, making things easier to browse but harder to transplant to other codebases.

1 is better than 2 in the long term, and it will be useful if it absorbs best practices. Maybe even we can have one "mother" folders for many sub-projects! Still, 2 can allow for minor tweaks in case 1 fails.

Reference to https://github.com/intellectronica/ruler?tab=readme-ov-file#nested-rule-loading

[intellectronica]

I'm not sure I understand the value of this, other than replicating the structure used by Cursor (and Copilot)? If there's a vendor-specific mechanism that you love using, why not just use that?

My tendency now is to follow the AGENTS.md standard (and any additional widely adopted standards that emerge). Regardless of opinion (I actually like it better, but I can see how others might prefer the apply rules solution), if that's the emergent standard, it should be prioritised.

[TomLucidor]

AGENTS.md is frankly a bit of a weak standard, we need something that has more sub-files so that the agent context is not completely blown up. Not sure if there are other open standards up for the occasion tho

[intellectronica]

Yeah, love it or hate it, that's the standard that got the most support so far, so I guess we're stuck with that.

[TomLucidor]

Someone else will probably standardize .agent/rules/example.md (Cursor. Windsurf, KiloCode, Trae, AugmentCode, AmazonQ, etc.) but not sure how that can be handled for collectives