Use This SkillGet Pro
Safety scan · 1,403 skills scanned

What we check before you install.

A skill is a folder of instructions and files that an agent will follow. Before a skill is listed, and again every week, every file in its folder is scanned for the patterns a skill should never need. The result sits on every skill page and in the MCP connector, with the file, line and matched text for each finding, so you can judge it yourself.

Clear 1,360Needs review 35Flagged 8
The rules

Nine patterns, two levels.

A flag means the pattern has no place in a skill and you should read the file before installing. Review means the pattern is sometimes legitimate; the finding shows you the exact line so you can decide.

Downloads and executes code in one step

flag

curl or wget piped straight into a shell, or Invoke-Expression on a download. Common in installers, and still the single easiest way to run anything on your machine.

Destructive command against the system or home directory

flag

rm -rf on /, ~ or $HOME, fork bombs, mkfs, dd to a disk. rm -rf on a project folder such as node_modules does not match.

Sends data to an outside endpoint

flag

Slack or Discord webhooks, request-catcher services, ngrok tunnels and similar collectors.

Obfuscated or encoded code that is then executed

flag

base64 decoded into a shell, eval of decoded strings, python -c with exec.

Contains rules-override phrasing

review

"Ignore previous instructions", "do not tell the user" and similar. Security documents quote these as warnings, so this is a review item: read the line.

References credential files or secrets on disk

review

~/.ssh, id_rsa, AWS credentials, reading or printing .env files, the keychain.

Asks for elevated privileges

review

sudo, chmod 777, chown to root. Often legitimate for installing a system package.

Installs something that runs on its own later

review

crontab, launchctl, systemctl enable, Windows Run keys.

Rewrites shared git history

review

git push --force, git reset --hard origin.

What it is not

A scan, not a verdict.

This is a static pattern match over text files. It cannot run the skill, it cannot see what a model will do with the instructions, and it will miss things a determined author hides. A "Clear" result means none of the nine patterns appeared, nothing more. Skills belong to their authors; read the source, check the license, and install into a project you can afford to reset.

Found something the scan missed? Write to hello@usethisskill.com with the skill and the line.

Currently flagged

8 skills to read first.

Most of these are install instructions that pipe a download into a shell. That is normal for a CLI installer and still worth reading before you run it.