live-research

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

• Third-party content exposure detected (high risk: 0.95). This skill’s runtime workflow uses Bright Data discover with --include-content, which fetches and ingests outsider-authored web page text (public URLs returned by Discover) into the agent’s LLM context via the merged corpus.json (results[].content → claim extraction → cited brief).

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly runs Bright Data Discover at runtime (e.g., "bdata discover ... --include-content") to fetch external webpages (results[].link — e.g. https://reuters.com/… or https://eur-lex.europa.eu/…) whose fetched content is injected into the corpus and directly shapes the agent's prompts/output, so external URLs control prompts during runtime.