audit-expo-skill

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). The required workflow clones and reads outsider-authored free text from public GitHub checkouts (e.g., https://github.com/clerk/javascript.git and https://github.com/clerk/clerk-docs.git) into the audit context, then re-reads their .mdx/CHANGELOG.md/source files to extract claims for LLM processing—an indirect prompt-injection risk via fetched docs/SDK text.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill specifies shallow-cloning remote repos at runtime (git clone --depth 1 https://github.com/clerk/javascript.git and https://github.com/clerk/clerk-docs.git) and then reads those checkouts as the authoritative source that directly controls the audit outputs/prompts, so these URLs are runtime dependencies that influence agent instructions.