filesystem-context

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

• Third-party content exposure detected (high risk: 0.80). This skill explicitly instructs agents to persist and later read large tool outputs from the open web (e.g., "Web search returns 8000 tokens" written to scratch/search_results_001.txt) and to grep/read those files on demand, which means untrusted third‑party content from web searches can be ingested and interpreted by the agent.