ai-research-agent

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime “Trend Monitoring / Deep Research” workflow explicitly scans external sources (e.g., Hacker News, Twitter/X, Reddit, Substack, LinkedIn, Arxiv, Anthropic/OpenAI blogs, Product Hunt, Indie Hackers), whose fetched page text and scraped content would be ingested into the agent’s LLM context as free-form outsider-authored prose.