The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/prepare_ribonn.py fetches a public dataset from GitHub (raw.githubusercontent.com/CenikLab/TE_classic_ML). This is a legitimate data retrieval operation from a well-known service for biological research purposes.
[COMMAND_EXECUTION]: The skill provides standard Python command templates for data preparation and model fine-tuning. These commands use local scripts and a project-specific runner (src.runner), which is consistent with the intended use case for machine learning workflows.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect injection as it processes external biological data (CSV/TSV files). However, scripts/prepare_ribonn.py includes robust validation logic, such as checking for expected column headers, validating that sequences only contain DNA bases (ACGT), and ensuring label values are finite numbers, which effectively mitigates common data-driven risks.
[DATA_EXFILTRATION]: No evidence of unauthorized data transfer or sensitive file access was found. Network operations are limited to downloading public research assets.

codonfm-finetune