[Feature Request] Add pure SKILL.md for usage without MCP dependecies

[DennisAtConceito]

Describe the problem

I really love this idea of setting up AI tools to standardize Svelte development. This is really great an amazing. Thanks for your efforts!!
I was hoping to find somewhere pure SKILL.md files that do not mention MCP, so that I could just use them locally.

Describe the proposed solution

Add a section with pure SKILL.md with best practices, especially project/repo architecture, for Svelte developement that are independent from the MCP server.

Importance

nice to have

[paoloricciuti]

What do you mean by pure SKILL.md? We just released the svelte-core-bestpractices skill that has no mention of the MCP. Is that what you are looking for?

[DennisAtConceito]

Yes exactly. So based on the svelte-core-bestpractices, i was thinking about a suite of SKILLs that describe common coding practices/use cases such as they are listed in Prompts

[paoloricciuti]

How would that differ from svelte-core-bestpractices?

[DennisAtConceito]

Not differ, exactly in that way, just more documentation content transfered into SKILLs. In https://svelte.dev/docs/ai/prompts there is a huge list of document for different use cases. So their could be many SKILLs for all kind of use cases. Usage of SKILLs is very straightforward because you just add it to the agents context. In generall I have the feeling that SKILLs are becoming a fundamental standard in context engineering, so it would be cool to have curated set of SKILLs that is directly provided by the svelte team. I think there will be more and more third-party provider of SKILL.md such as Github and others. Why not go ahead and deliver them from the svelte team itself?

[paoloricciuti]

Ohhh gotcha... I don't think have every single doc page as a SKILL.md would be beneficial. It would add too much noise.

Also you can basically get the same result by:

• invoking the MCP server
• using the CLI
• appending /llms.txt to a documentation page

We could have a skill to search the documentation though that will guide the LLM on where to find the relevant links and how to properly fetch them. This will also guarantee that is always up to date.