← ClaudeAtlas

skillvetlisted

Audit an Agent Skill (Claude Code / Cowork skill) for malicious code, data exfiltration, prompt-injection instructions, dangerous shell commands, obfuscation, over-broad permissions, and supply-chain risk BEFORE installing or trusting it. Use this whenever the user wants to scan, vet, review, audit, or check the safety of a skill, a SKILL.md, a .skill package, a plugin, or a skills marketplace/repo they downloaded or are about to install — even if they just say "is this skill safe?", "check this skill for malware", or paste a skill folder. Always run it before installing a third-party skill.
V3r7ig0/skillvet · ★ 2 · AI & Automation · score 73
Install: claude install-skill V3r7ig0/skillvet
# skillvet — security scanner for Agent Skills Agent Skills are code plus instructions that get loaded into an agent's context and run with the user's own permissions. A malicious skill can steal secrets, exfiltrate data, run remote payloads, or silently steer the agent against the user. This skill audits a target skill and produces a clear risk report so a human can decide whether to trust it. Your job is to run the static engine, then apply **semantic judgment** on top of it: a pattern match is a reason to look, not a proof of malice, and a clean scan is never a proof of safety. You add the context the regex engine can't. ## When to use Trigger whenever the user wants to vet a skill's safety: "scan this skill", "is this SKILL.md safe", "check this plugin for malware", "audit this skills repo before I install it", or when they paste/point at a skill directory or a `.skill`/`.zip` package. Prefer running this before any third-party skill is installed. ## Workflow Follow these steps in order. Do not skip the static engine — it is deterministic and high-recall, and it anchors your review in specific lines. ### 1. Locate the target Identify the path to scan. It can be: - a skill directory containing `SKILL.md`, - a packaged `.skill` or `.zip` file, - a whole repo / marketplace containing several skills (scan each skill dir). If the user gave you a repo with multiple skills, list the `SKILL.md` locations first and scan each skill directory separately so findings stay att