The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external datasets from the data/ directory which are untrusted and could contain malicious instructions.
Ingestion points: Loads all files in the data/ directory (CSV, Parquet) and processes them using pandas.
Boundary markers: Absent. The prompts do not include delimiters or instructions to ignore embedded commands within the datasets.
Capability inventory: The skill utilizes pip install for package management, creates and executes local Python scripts (eval.py), and writes results to the outputs/ directory.
Sanitization: Absent. The skill does not validate or sanitize the content of the data files before processing them.
[DYNAMIC_EXECUTION]: The skill's primary workflow involves generating a Python script (eval.py) at runtime and executing it iteratively to score analysis quality.
[COMMAND_EXECUTION]: The instructions prompt the agent to install external Python packages (such as pandas, scikit-learn, and plotly) via system command lines if they are not present in the environment.

codex-data