I build Workshop, a local app that puts a repo’s skills and Cursor rules on a bookshelf. You can switch between Cursor, Claude Code and Codex to see which files each can read. A wider spine means more text loaded before a task.
The short video shows task search, saved shelves, the Library, and installing a skill through Discover with a scan and review before it writes files. It also shows comparing a newer skill edition and merging it with local edits.
One limitation for Cursor users: Workshop’s Usage view reads local Claude Code and Codex session logs only. It doesn’t track Cursor usage. A logged skill read also doesn’t prove the agent followed its instructions. The usage counts and merge example in the video are sample data.
To try it, download the free sample and unzip it. With Node 20 or newer, run this from the download’s top folder:
node Library/app/workshop.mjs
Then choose a project folder. The app and six skills are free; I also sell an optional library of 150 skills. The demo and download are at Rival Workshop: finish and review your agent’s work.
I’m interested in where this gets your Cursor rules wrong, especially rules that apply only to particular files.