research-summarizer

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

• Third-party content exposure detected (high risk: 0.95). The workflow’s “Search + fetch” step uses web_search and fetch_url to ingest arbitrary public web pages’ readable text into the agent for clustering/drafting, which is outsider-authored free text from runtime-fetched web content.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly runs web_search then fetch_url at runtime to retrieve arbitrary external webpages whose contents are injected into the agent context and used to draft the briefing (i.e., the web_search + fetch_url pattern fetching arbitrary external URLs controls prompts), so those fetched URLs are a runtime dependency that can directly influence agent instructions.