security-vetlisted

# Security Vet **When to use:** anything third-party is about to be installed. A skill, a plugin, an npm or pip package, an editor extension, a cloned repo. This is a standing gate, not a suggestion: **scan first, hold and ask on any flag.** The order is **scan → verdict → user approval → install.** Never install first and scan later. Once code has run on your machine, a scan tells you what happened, not what will happen. ## Why this is worth the five minutes Agent skills are instructions a model will follow with your tools and your credentials. A malicious or careless one does not need an exploit — it just needs you to install it and then ask it to do something. The blast radius is your filesystem, your tokens, and anything your agent can reach. That is a different risk shape from a normal dependency, and it deserves its own gate. ## Step 1 — Reconnaissance Before reading a line of code: - **Who is the maintainer?** Real identity, or anonymous? Account age? Other repos with history? - **Adoption:** stars, downloads, forks — and how fast they appeared. A week-old repo with 4,000 stars is a flag, not a credential. - **Maintenance:** last commit, release cadence, open issue response. - **Signals of care:** is there a `SECURITY.md`, CI, tests, a license? An anonymous one-month-old maintainer is **not a block**. It changes the *install method* — see Step 5. ## Step 2 — Static scan of the actual code Read what it does. Specifically: - **Network egress.** Every URL a