cherry-pick

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

• Third-party content exposure detected (high risk: 0.95). The skill is explicitly designed to process a pasted block from an external source (e.g., Perplexity/another Claude session) that the user did not author, and that pasted free text is the runtime input the agent must mine and judge—so outsider-authored prose is fed into the LLM context.