Security Review Lens

Security Review Lens is a Claude Code skill that scans third-party agent skills before you install them — so you find out a skill tries to phone home or steal credentials before it's running inside your environment, not after. Instead of spitting out a pass/fail verdict, it acts as a highlighter: it surfaces risky patterns and lets you decide. The skill runs 35 detection rules across four severity tiers — 8 CRITICAL (reverse shells, credential theft, prompt injection, cryptojacking), 8 HIGH (script execution, package installation, persistence), 10 MEDIUM (filesystem access, network requests, dynamic code generation) and 9 LOW (overly broad triggers, missing metadata, scope mismatches). On top of that it runs 7 semantic checks for indirect data leakage, remote access, hidden functionality, self-modification, obfuscation, chained attacks and trust-boundary violations. Its 9-step workflow covers file location, skill-type classification, frontmatter validation, rule execution, context-aware analysis, semantic checks, attack-surface mapping, red-team analysis and a final review. It deliberately bans the word 'safe' so it never gives you false assurance. It's MIT-licensed and needs no MCP servers to run.