mass-desire-angle-finder

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md Phase 1 "Reddit Research" (Step 1.1) explicitly requires using web_search and web_fetch to open and read full public Reddit threads and comments (and similarly instructs scraping niche forums, Amazon reviews, and Quora), so the agent ingests untrusted user-generated web content that directly drives angle extraction and subsequent actions.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly requires runtime fetching of external pages (notably reddit threads via site:reddit.com fetched with web_fetch, and also site:amazon.com, site:pubmed.gov, site:quora.com) whose verbatim content is copied into the agent's research and used to drive prompt generation, so these URLs are runtime dependencies that directly control the agent's prompts.