Do these in Claude Code or Codex (the desktop apps or the terminal), or in Terminal yourself. Ordinary chat can’t run commands. Haven’t installed yet? Install SkillCanary first.
STEP 1
Turn on protection
Paste this into Claude Code or Codex:
Set up SkillCanary: run canary setup.
Or run canary setup in Terminal.
A Mac window asks which protection level you want. We recommend Guard: agents can then install skills only through SkillCanary. Guard and Lockdown ask for your Mac password in the standard macOS window; your agent never sees it. Compare the levels.
If your Mac asks to install Apple’s command-line developer tools, choose Install, then run setup again.
STEP 2
Install your first skill through SkillCanary
Try it with Anthropic’s public frontend-design skill. Paste this:
Use Canary to install https://github.com/anthropics/skills/tree/main/skills/frontend-design
SkillCanary downloads the skill into quarantine, checks every file, and shows you a Mac dialog with its verdict and what the skill can do. Choose Install, and the skill is ready in Claude Code and Codex from your next session. Your agent only learns the outcome; it never reads the skill.
The second check runs on your own Claude Code login or an Anthropic or OpenAI API key. Without one, SkillCanary still checks every file, and marks the result “needs your review” so you decide.
ANY TIME
See what you’re protected against
Run canary doctor, or ask your agent to. It shows your level, anything that weakens it, and exactly what SkillCanary can claim on your Mac. To change level, run canary setup again.