suno

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

• Third-party content exposure detected (high risk: 0.85). SKILL.md defines an “on-the-fly external research” action (research <topic>) that uses runtime WebFetch/Perplexity to retrieve public web content and then synthesizes it into the LLM’s prompt (“Translate findings into the 6-layer formula” / output includes research-derived facts and citations), so outsider-authored free text can enter the model context via those web lookups.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's "research" action explicitly performs on-the-fly WebFetch/Perplexity lookups (e.g., help.suno.com/en/articles/11362369) at runtime and synthesizes fetched content into Suno style/lyrics prompts, so external pages are used during runtime to directly control prompt construction.