duckdb

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill prompt contains examples that embed real-looking secrets and passwords directly into SQL and command strings (e.g., AWS KEY_ID/SECRET, MinIO credentials, and a PostgreSQL password), which requires the LLM to handle or reproduce secret values verbatim and therefore poses a high exfiltration risk.

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow described is DuckDB directly executing user-authored SQL to read data files (e.g., read_parquet/read_csv/read_json from paths like s3://... or rest+http://...), meaning any outsider-provided free text that becomes part of those inputs/queries could be ingested by the engine at execution time.