ai-tools

This adds a new skill, the skill, that instructs the agent to file an issue on this repo in case the user corrects them on a Svelte mistake so that we can catch as many of these as possible. This is a bit of a weird skill because it is completely opt-in. We don't ship it with plugins and that is because, obviously, this can be disruptive for work. We want people to actually opt in to use this so either they need to download it or they need to install it with their favourite skills manager. I've also tweaked a bit of configuration. I had to install Node types and I created a specific so that the scripts could actually import TS files, but the rest of the application should not. Finally, I tweaked the documentation skill script to also include the front matter because people also need that to be able to actually use the skill.

Is the instruction "...MUST be used whenever creating, editing or analyzing any Svelte component (.svelte) or Svelte module (.svelte.ts/.svelte.js)" really necessary? Some models repeatedly re-read the skill even when prompted for a minor follow-up adjustment in the same session where the skill has already been loaded. Every time I ask for an adjustment to a component within a session where the skill was already fetched, the model reads the skill all over again. This leads to unnecessary token waste. Could we optimize this prompt instruction or the caching/activation behavior so it doesn't trigger redundant skill reads during follow-up messages in the same conversation?