Skills don’t work if your setup rejects them. Skillproof finds what you need and fits it into your setup — seamlessly, with your permission.
Paste one short prompt into any agentic AI — it installs Skillproof and starts your first session.
For Claude Code. Downloads six files to disk so you can read them — nothing is piped into a shell, and it never edits your setup without showing you the plan and getting your yes.
For Cursor, Claude Desktop, and any other MCP-capable agent — the catalog as four read-only tools your agent can query mid-task.
✓ Every skill here was checked to make sure it isn’t malicious before it was listed.
Skill — a small folder of instructions your AI reads so it does one job well (write honest tests, design cleaner pages). Agent — the AI (Claude, ChatGPT, Cursor…) doing the work for you. MCP — a plug that lets other tools talk to it. Your setup — the config files those tools already read (CLAUDE.md, installed skills, rules).
Add items to see what your stack covers.
A small folder of instructions your AI reads so it does one job well — honest tests, cleaner design, tighter research. Skillproof finds the right one and fits it into what you already have.
Only with your yes. It shows you the whole plan first — every file it would touch — keeps a backup, and tells you how to undo.
Every listed skill was checked for malicious code before it was listed, and re-checked whenever its code changes — anything flagged is pulled and never shown. The Skillproof skill also reads a skill’s source before it runs a command from it, and nothing is ever piped into a shell. Still read anything before you run it.
Free and open source. No account.
Any AI chat via the prompt (Claude, ChatGPT, Grok…); Claude Code via the skill; Cursor, Claude Desktop and any MCP-capable agent via the server.