Never install a Claude skill Without doing this

@itsmariahbrunner

49.2K followers · usually 2.7K views a post

Views

21.0K

Likes

1.2K

Comments

221

Shares

208

SoundTikTok sound · original sound - itsmariahbrunner

Promoted product

The AI Safety Playbook

Stan | Linksite

2 videos

Why it worked

The video addresses a common concern among AI users: the potential risks of installing third-party code or 'skills'. By providing a clear, actionable safety checklist and offering a free guide, the creator taps into user anxieties and offers a valuable solution, driving engagement through comments.

Summary

This video explains a 4-step safety check process before installing any Claude AI skill. The creator demonstrates how to paste the skill into a regular Claude chat to identify malicious code, then customize it for personal use before moving it to Claude code.

Structure

1Introduction to Claude skill safety check
2Step 1: Paste into regular Claude chat
3Step 2: Run safety prompt
4Step 3: Move to Claude code

Original caption

comment “secure” over on my IG @Itsmariahbrunner and I’ll send you the exact prompts I use :) AI Safety Series part 1!! A downloaded skill can run code on your computer, so before I install anything I paste it into a Claude chat and have Claude flag anything sketchy first. Then customize and save. #claude #aisafety #aitips #aitools #claudeai

More from @itsmariahbrunner

Product placement

The video demonstrates using Claude AI and its 'skills' feature. The creator shows a prompt designed to vet these skills for safety and malicious code before installation. The creator also offers a guide called 'The AI Safety Playbook' for those who comment 'SECURE'.

Call to action

Comment "SECURE" to get the exact prompts for this process.

On-screen text

Never install a Claude skill Without doing this AI SAFETY SERIES (PART 1) AI Safety Series SAFETY CHECK: 1. PASTE INTO REGULAR CLAUDE CHAT 2. RUN SAFETY PROMPT 3. CLEAN? MOVE TO CLAUDE CODE 4. TAILOR IT TO YOU + SAVE comment SECURE

Transcript

Before you install a Claude skill, there is a 30-second check that could save you from a real mess. This is part 1 of my AI safety series. And this is exactly how I vet anything before I bring it into Claude code. Step 1: Paste into regular Claude chat, not Claude code. And this is because that window can't run anything on your computer or touch any of your files. And then just drop in the prompt that I built and it will read every single line and flag anything that's sketchy or malicious. Like code that is built to steal your information or run without permission. Step 2: Run safety prompt. And then once you get the green flag that it's clean, bring it into Claude code. Step 3: Clean? Move to Claude code. And then the first thing that I do is ask my agent, or you can just ask AI to tailor that skill to me. Look at everything the AI knows about me, my writing style, the tools I use, and how I typically ask for things. Then go through every file and line in it, like skill and any scripts in includes, and rewrite any generic instructions so they match my actual workflow and preferences. Swap out placeholder examples for examples that reflect my real work. Don't add anything I'll never use so it always stays true. Add and edit or define you know I always want formatting rules, tone, tools, file locations. Step 4: Tailor it to you and save. Then save it as a skill and trigger it whenever you need it. If you comment secure, I'll send you exactly how to do this.