Rams reviews UI code (React, Vue, Svelte, CSS, SwiftUI) for visual design and accessibility. It reads the files you send it and returns findings with severity, category and file:line. It does not edit files, run code, open pull requests or touch any repository. Applying a fix is always your step.
Walk the user through these four steps in order. Keep each step short and show the real tool output.
1. Connect
The first Rams tool call asks the user to sign in at rams.ai with Google or GitHub and approve access. There is no key to paste. If a tool call fails with an authorization error, ask the user to connect Rams from the plugin settings and try again.
2. Run a quick check
Ask for one UI file the user is working on, or a component they can paste.
Call quick_review with the file path and its full content. It answers in
about ten seconds with the top issues. It returns no score and no patch.
Show the issues as a short list: severity, what is wrong, and file:line.
3. Run a full review
When the user wants a score or ready-made fixes, call review_files once on
the same files. It returns a 0-100 score, a one-line summary, and issues with
fixes. Many issues include a unified diff in patch that applies with
git apply.
Show the score first, then the issues from most to least severe. Offer to apply the patches. Only apply one after the user agrees.
4. Verify the fixes
After fixes are applied, call verify_fixes with the updated files and the
issues from step 3. It reports which issues are fixed and which are still
present.
Usage
If the user asks how many reviews they have left, call usage and relay the
answer as written.
Limits
- Rams judges UI code only. For backend code, configs or tests, say Rams has nothing to review there.
- Rams only sees the content sent in the call. It cannot read a repository on its own.
- Rams cannot deploy, merge, change billing or act on anyone's account.