The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external files which presents a vulnerability surface for indirect prompt injection attacks.
Ingestion points: The skill reads from local data files using duckdb.sql() (e.g., 'data.csv', 'sales.parquet', 'users.json') and third-party DataFrame libraries as demonstrated in SKILL.md and references/integration.md.
Boundary markers: The provided code snippets and instructions do not include any delimiters or specific warnings to ignore potential instructions embedded within the data files.
Capability inventory: The skill enables the agent to read from the file system, perform complex SQL operations, and write results back to the file system using the COPY command (documented in SKILL.md).
Sanitization: There is no evidence of data sanitization or validation before the data is processed or the results are presented to the agent.
[COMMAND_EXECUTION]: The installation and setup sections instruct the agent to execute shell commands to manage dependencies.
Evidence: Commands such as python3 -c "import duckdb; print(duckdb.__version__)" and pip3 install duckdb 'polars[pyarrow]' are included in the documentation.
[EXTERNAL_DOWNLOADS]: The skill requires downloading external Python packages from the official Python Package Index (PyPI).
Evidence: The skill instructs the installation of duckdb, polars, and pyarrow. These are recognized as well-known and legitimate packages in the data science ecosystem.

duckdb