duckdb
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to download the DuckDB CLI binary from its official GitHub repository releases, which is a well-known and reputable source for this tool.
- [DATA_EXFILTRATION]: The documentation and recipes guide the agent to access local conversation transcripts at paths like
~/.gemini/.../transcript.jsonland~/.claude/projects/.... These files contain sensitive interaction history, and the skill provides SQL patterns to audit this data. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted local data files (JSON, CSV, Parquet) through DuckDB SQL queries.
- Ingestion points: Uses
read_json,read_csv, andread_parquetfunctions to load local files into the analytical engine. - Capability: Utilizes the
duckdbCLI for data processing and supports materializing query results back to the local file system using theCOPY ... TOcommand. - Boundary markers: The instructions do not define specific boundary markers or markers to ignore instructions embedded within the processed data files.
- Sanitization: No explicit sanitization or escaping mechanisms for the input data are described in the provided recipes.
Audit Metadata