personalization-6-buckets

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

• Third-party content exposure detected (high risk: 0.85). This skill instructs connecting to ColdIQ MCP/API and using calls like mcp__coldiq__extract_post_engagement and mcp__coldiq__fetch_page_content, which at runtime will ingest outsider-authored web/social/profile text into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill tells users to run remote code via the npx command "COLDIQ_API_KEY= npx -y @coldiq/mcp@latest", which will fetch and execute an external package at runtime and is required for the mcp__coldiq__* connectors the skill references.