What problem does it solve? Installing third-party agent skills from uploaded archives or GitHub repositories carries risks: malicious scripts, missing licenses, credential leaks, duplicate names, and provider-specific behavior that cannot run locally. This Skill provides a structured review workflow that treats external content as untrusted data and produces an accept, adapt, defer, or reject recommendation before any files are written. ## Core Features & Use Cases - Safe archive inventory: Lists archive entries, detects path traversal, symlinks, and suspicious compression before any extraction, and reconciles manifest entries against actual SKILL.md implementations. - Provenance and license gating: Records source identity, hashes, and license evidence, blocking installation when licensing is absent or unclear. - Compatibility mapping: Classifies every external dependency as portable, adaptable, or unsupported, and flags provider-specific behavior such as credential discovery or restart instructions. - Use Case: A user uploads a zip bundle of community skills. The Skill inventories the candidates, detects one duplicate of an existing project-owned skill, redacts credential-like fields in a bundled config, and recommends accept for two licensed skills while deferring the rest. ## Quick Start Review the uploaded skill archive and tell me whether each candidate skill is safe to install, without installing anything yet.