gen-ai-explainer

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

• Third-party content exposure detected (medium risk: 0.30). Stage 1 “research” in references/research-director.md uses WebSearch and then WebFetch to ingest verbatim outsider-authored content from search results/source pages (including Reddit) into the runtime context.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly requires running WebSearch/WebFetch at runtime and performing targeted fetches of source pages (e.g., site:youtube.com and site:reddit.com), and those fetched pages are ingested to produce research findings that directly control downstream prompts and outputs.