What problem does it solve? Installing third-party AI skills into a legal workspace carries real risk: a malicious or sloppy SKILL.md can inject instructions, wire up unvetted MCP connectors, or write files outside its scope. This Skill enforces a defense-in-depth install pipeline so nothing is written to disk until the user has seen the raw source and explicitly approved it. ## Core Features & Use Cases - Allowlist gating: Reads an administrator-controlled allowlist.yaml before fetching anything, refusing unlisted registries, publishers, MCP connectors, and licenses in restrictive mode. - Raw source transparency: Displays the complete raw SKILL.md with flagged injection patterns, runs a structural trust check on hooks, MCP servers, tool permissions, file-write paths, and network calls, then runs a skills-qa quality scan. - License and freshness validation: Extracts SPDX license identifiers by strict pattern matching, verifies LICENSE files against registry metadata, and validates freshness frontmatter before injecting a freshness-gate preamble at install. - Use Case: A paralegal wants to install a community contract-review skill from a GitHub registry. The installer checks the allowlist, shows the full raw SKILL.md, flags an external URL in the file, runs the QA scan, and only copies files after the supervising attorney-approved user types yes. ## Quick Start Install the contract-review skill from the watched registry and show me the raw SKILL.md before writing anything.