Documentation Feedback: "Copy for LLM" context bloat & suggestion for Agent Skills

[cheema-corellian]

I really appreciate the new docs—they are definitely more LLM-friendly than before! However, I wanted to share feedback on two areas that could improve the experience significantly:

1. Context Bloat in "Copy for LLM"
On pages like Getting Started, the content changes based on user selection (e.g., selecting Vite + pnpm). However, clicking the "Copy for LLM" button copies the documentation for all possible scenarios, rather than just the active selection.

• The Issue: This floods the AI context window with irrelevant info, causing the model to get confused and hallucinate errors.
• Current Workaround: I have to manually highlight and copy the specific text I need.
• Suggestion: It would be great if the copy button respected the current state of the page toggles to keep the context focused.

2. Component Discoverability & Agent Skills
Agents often don't know which Spectrum components exist, so they guess (often wrongly) or build custom React components instead.

• Suggestion: Have you considered providing an "Agent Skill"? I've seen tools like Vercel provide skills that work better than MCP in some cases because they use progressive disclosure rather than dumping context.
• Next Steps: I am tempted to create an unofficial S2 skill on GitHub to solve this, but I wanted to see if this is something the team is already thinking about.

[reidbarber]

Agent Skills are definitely on our radar. I think we mainly need to think about maintenance/distribution before we publish anything. If you have any good examples, feel free to share.

[cheema33]

Thanks for the response, @reidbarber! To help explore what this could look like, I went ahead and created an unofficial AI agent skill for Spectrum 2 here: https://github.com/cheema33/react-spectrum2-skill.

It’s intended as a proof-of-concept until an official version is feasible. Feedback and patches are definitely welcome!