The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents various command-line arguments and dot commands for the DuckDB CLI, including the ability to run SQL queries directly (-c), execute scripts from files (-f, .read), and access an external editor (.edit).
[DATA_EXFILTRATION]: The skill provides instructions for exporting data to files via the COPY command and dot commands like .output, which could be used to write sensitive information to attacker-controlled locations or local files.
[PROMPT_INJECTION]: The skill facilitates the processing of untrusted external data, creating a surface for indirect prompt injection.
Ingestion points: The documentation in SKILL.md shows how to read CSV, Parquet, and JSON files using functions like read_csv_auto and read_json_auto.
Boundary markers: No specific boundary markers or instructions are provided to help the agent distinguish between data and potentially malicious instructions embedded within these files.
Capability inventory: DuckDB possesses capabilities for file system access (read/write) and shell command execution (evidenced by the mention of the .sh command in the Safe Mode description in SKILL.md).
Sanitization: There are no sanitization or validation steps described for the data being ingested.
[NO_CODE]: This skill is entirely comprised of markdown documentation and does not contain any executable code files or scripts.

duckdb-en