publish-research-site

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

• Third-party content exposure detected (high risk: 0.75). The required workflow invokes the $deep-research skill (step 2), which typically fetches public web pages and ingests their readable text (outsider-authored content) into the agent’s LLM context via citations/evidence.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs using "npx vercel@latest" at runtime if the local vercel CLI is missing, which will fetch and execute remote code from the npm registry during the deployment step (npx vercel@latest).