The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The gget gpt module accepts a user-supplied prompt for text generation via OpenAI. Because this tool is intended to work with bioinformatics data (like gene descriptions or sequence analysis outputs) which may originate from external, untrusted databases, it constitutes an indirect prompt injection surface.
Ingestion points: The prompt parameter in gget.gpt() as documented in SKILL.md and references/module_reference.md.
Boundary markers: None found in the provided scripts or documentation for isolating untrusted genomic data within the LLM prompt.
Capability inventory: The skill can perform network operations, write files, and execute local binaries (DIAMOND, Muscle5).
Sanitization: No evidence of prompt sanitization or escaping was found in the example scripts.
[COMMAND_EXECUTION]: The skill interfaces with several local bioinformatics binaries, such as Muscle5 for sequence alignment and DIAMOND for local protein alignment. These are executed as subprocesses or via library wrappers.
Evidence: Usage of gget muscle and gget diamond across SKILL.md, references/module_reference.md, and scripts/batch_sequence_analysis.py.
[EXTERNAL_DOWNLOADS]: The gget setup module facilitates the download of large external resources, including AlphaFold2 model parameters (~4GB) and the Eukaryotic Linear Motif (ELM) database. It also uses curl to fetch genomic reference files from Ensembl.
Evidence: gget setup alphafold, gget setup elm, and gget ref -d in SKILL.md and references/module_reference.md.

gget