The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data files (CSV, Parquet, JSON, Excel), which constitutes a potential surface for indirect prompt injection. This risk is effectively addressed by the skill's security guidelines.
Ingestion points: Data is loaded via read_csv_auto, read_parquet, read_json_auto, and read_xlsx functions as described in SKILL.md and references/file-ingestion.md.
Boundary markers: SKILL.md includes explicit instructions to "Validate file paths exist before reading" and "reject paths outside the working directory."
Capability inventory: The skill utilizes file write capabilities through DuckDB's COPY TO command and Python library export functions (documented in references/export-patterns.md). It also performs SQL transformations and data profiling.
Sanitization: Path validation is mandatory. Data content is processed using DuckDB's native schema inference and parsing engines, which provide structured handling of external inputs.
[SAFE]: The skill promotes security best practices by advising against hardcoding credentials and recommending the use of .gitignore for data files.
[SAFE]: All identified Python dependencies are well-known, reputable packages used extensively in the data science community for file handling and analysis.

duckdb