The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a local Python script (scripts/aggregate_results.py) for data processing. The script is written defensively, implementing checks to ensure that input and output files are handled safely and do not overwrite each other or follow dangerous symlinks.
[DATA_EXFILTRATION]: While the skill processes evaluation results, it includes explicit warnings in the documentation and workflow against including private data, customer information, or credentials in evaluation sets.
[INDIRECT_PROMPT_INJECTION]: The Python script ingests external data in JSONL format. This data is parsed strictly as JSON and used for statistical calculations. The script does not evaluate or execute strings contained within the input data as code or instructions.

agent-evaluation