The Agent Skills Directory

[SAFE]: The skill does not contain any suspicious command execution, persistence mechanisms, or obfuscated content. The author uses standard instructional language and code examples to demonstrate functionality.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process genome-wide association study (GWAS) data and individual genotype files, which represents a potential surface for indirect prompt injection if external data sources are compromised.
Ingestion points: Retrieves associations from the GWAS Catalog (EMBL-EBI), Open Targets Genetics, and accepts user-provided raw genotype data (e.g., 23andMe, VCF files).
Boundary markers: The skill explicitly defines its scope, stating it is for educational/research purposes and not for diagnosis, which helps set appropriate agent behavior boundaries.
Capability inventory: Employs mathematical functions and specialized GWAS tools for retrieval and risk calculation; no arbitrary system command execution or uncontrolled file writing is present.
Sanitization: While explicit sanitization code is not shown in the instructions, the skill defines a structured workflow for filtering variants by p-value, MAF, and Hardy-Weinberg equilibrium, which serves as a functional validation layer.
[EXTERNAL_DOWNLOADS]: Mentions external academic and research repositories such as the EMBL-EBI GWAS Catalog, PGS Catalog, and the Broad Institute's LD Hub. These are well-known technology and scientific services used for data retrieval and research.

tooluniverse-polygenic-risk-score