methodology

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md, the skill parses free-form $ARGUMENTS (any unrecognized text) as a topic used to “search across all reference docs,” and in Step 3 it also performs web-research (WebSearch/WebFetch or Perplexity MCP) based on the user’s domain scoping question, which means outsider-authored text can drive the retrieval/ingestion of additional arbitrary content at runtime.