The Agent Skills Directory

[SAFE]: The skill utilizes standard Python libraries such as zipfile, tempfile, json, and csv for file manipulation and environment management, following secure development practices for temporary directory creation.
[SAFE]: External references and model checkpoints are hosted on well-known platforms (Hugging Face) and originate from the vendor (NVIDIA), which are recognized as trusted sources.
[SAFE]: The instructions mandate strict validation of input CSV data, including DNA character normalization (A/C/G/T) and sequence length checks, before any execution occurs. This significantly reduces the risk of malformed data causing unexpected behavior.
[INDIRECT_PROMPT_INJECTION]: The skill processes external CSV data which serves as input for model scoring. While this presents a potential ingestion surface, the risk is mitigated by explicit instructions for the agent to manually validate every row against specific schema and sequence constraints before command preparation or execution. (Severity: SAFE)

codonfm-score