Claude Code skills and sub-agents, with a safety check before each install
I added skills and helper agents to Claude Code. Before I install anything, the AI checks if it is safe.
- Why
- Anyone can publish a skill. I wanted the useful ones without the risky ones.
How I built it
- I asked Claude to check each skill before I installed it: who made it, how many people use it, and what it can touch.
- I installed the ones that passed, like a skill finder and a competitor research skill.
- The installer showed a checklist of 72 apps. Claude found a single command that skipped it.
- I set up helper agents in plain words, saved so they work in every project.
- I asked Claude to review a paid skill bundle before buying. It found most of it was free elsewhere. I did not buy it.
- To copy it you need
- Claude Code. The gotcha: check the source before you install, every time.
Try it yourself
Copy this prompt and paste it into ChatGPT or Claude to try it yourself.
Marty Stevens-Heebner has her AI check every skill before she installs it (see her What Are We Building? page). I want to do the same. Here is a link to a skill I am thinking about. Tell me who made it, how many people use it, what license it has, and what it can read or change on my computer. Then tell me plainly: install or skip. If I install, give me one command with no interactive prompts. Start by asking for the link. Ask me for anything you need.
Tools I used for this project
