This is for when you're building a UI and need to tell backend what data you need without overstepping into their territory. It writes structured markdown docs that describe what information shows on screen, what actions users can take, and what states you need to handle, but deliberately avoids specifying endpoints, field names, or API structure. The format includes sections for uncertainties and explicit invitations for pushback, which honestly feels like teaching people how to communicate rather than just documenting requirements. Everything goes into `.claude/docs/ai/<feature-name>/backend-requirements.md`. Works best if both sides actually read it and update the discussion log, otherwise you're just writing docs into the void.
npx skills add https://github.com/softaworks/agent-toolkit --skill frontend-to-backend-requirements