The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data fetched from external biological databases (GWAS Catalog and Open Targets) which could theoretically contain malicious instructions embedded in metadata fields.\n
Ingestion points: The python_implementation.py script fetches data using gwas_get_associations_for_trait and OpenTargets_search_gwas_studies_by_disease. Documentation in SKILL.md also provides patterns for using requests.get to fetch TSV and JSON data from ebi.ac.uk.\n
Boundary markers: Absent. The skill logic assumes the integrity and structure of the data provided by the external APIs without using delimiters or instructions to ignore embedded content.\n
Capability inventory: The skill is limited to scientific data aggregation, p-value filtering, and ranking. It does not possess capabilities for file system modification, arbitrary shell execution, or network exfiltration of local sensitive data.\n
Sanitization: No explicit sanitization or instruction-filtering is applied to the string fields (such as trait descriptions or study metadata) before the data is processed or presented to the agent context.\n- [EXTERNAL_DOWNLOADS]: The skill documentation and implementation scripts reference downloading large biological datasets from established scientific repositories.\n
Evidence: SKILL.md provides code to download a 37MB TSV dataset from the European Bioinformatics Institute (EBI) at https://www.ebi.ac.uk/gwas/api/search/downloads/alternative using pandas and requests.\n
Evidence: Implementation scripts interact with Open Targets Genetics and EBI infrastructure via standard API tools.\n- [COMMAND_EXECUTION]: The documentation includes standard environment setup and diagnostic tools.\n
Evidence: SKILL.md suggests using pip install tooluniverse to manage dependencies.\n
Evidence: Several test scripts (test_gwas_tools.py, test_skill_comprehensive.py) are included to verify tool and workflow functionality within the execution environment.

tooluniverse-gwas-trait-to-gene