The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The references/workflows.md file contains a code example for caching results that uses pickle.load(). Deserializing data with the pickle module is inherently unsafe; if a user is tricked into loading a maliciously crafted cache file, it can lead to arbitrary code execution on their system.
[PROMPT_INJECTION]: The skill ingests data from external biological databases and user-supplied sequence files (FASTA, CSV). This establishes a surface for indirect prompt injection, where malicious instructions could be embedded in database records or sequence metadata to influence the agent's behavior during processing.
[EXTERNAL_DOWNLOADS]: Modules such as gget ref and gget setup facilitate the download of genomic datasets, model parameters, and local databases from external scientific repositories (e.g., Ensembl, UniProt, and DeepMind's AlphaFold models). These are expected operations for the tool's bioinformatics use case.
[DATA_EXFILTRATION]: The skill connects to multiple external APIs, including various genomic database providers and the OpenAI API via the gget gpt module. Users should be aware that search terms, gene identifiers, and sequences are transmitted to these services for processing and analysis.

gget