The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains instructions for installing the DuckDB CLI using a shell command: curl -fsSL https://install.duckdb.org | sh. This pattern involves downloading a script from a remote server and executing it directly in the shell. The source domain belongs to the official DuckDB project.
[CREDENTIALS_UNSAFE]: The documentation includes several examples for configuring access to protected resources using credentials.
The PostgreSQL attachment example uses a password=secret parameter.
The S3 configuration example includes AWS credential placeholders such as KEY_ID 'AKIAIOSFODNN7EXAMPLE' and SECRET 'wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY'.
These are identified as standard documentation placeholders rather than actual hardcoded secrets.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing third-party libraries and extensions.
Commands for installing Python packages duckdb, pandas, and pyarrow are included.
The skill demonstrates how to install DuckDB extensions such as httpfs, iceberg, delta, and postgres from their official registry.
[PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted external data files (Parquet, CSV, JSON, Iceberg) which represents a surface for indirect prompt injection.
Ingestion points: Data is loaded into the agent context through functions like read_parquet(), read_csv(), and read_json() in SKILL.md.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands within the files are provided in the usage snippets.
Capability inventory: The skill provides capabilities for writing files (COPY TO), network operations through database extensions, and shell command execution for installation.
Sanitization: The Python API examples demonstrate the use of parameterized queries (using ? placeholders), which is a recommended practice for preventing SQL injection.

duckdb