duckdb

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

• Third-party content exposure detected (high risk: 0.70). The SKILL.md explicitly shows reading external, potentially untrusted files/URLs (e.g., "SELECT * FROM read_parquet('https://example.com/data.parquet')" and the "Remote Files (httpfs)" section referencing HTTP(S) and s3:// paths), so the agent would ingest and act on third-party web-hosted data as part of its workflow.