What should your AI be allowed to do on its own, and what must always wait for you?
- Module
- 04/05
- To read
- ~10 min
In 2025, a malicious npm package ran on developers’ machines during install. Many of those machines had AI coding CLIs installed and logged in. What do you think the malware did with them?
In the module you make your guess first, then read what happened and why.
Grant by task, not by convenience
Deny wins, and denial must be enforced
Never auto-approve these
Tools from outside are untrusted input
Your AI’s permission list
A spec for your own app, as a text file you download. The module also gives you the prompt for Claude Code, Codex or Antigravity, a checklist to check what it built, and recommended skills.
- Defense in depth
- Separate environments and credentials
- Branch protection