The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructions recommend that the agent "Write Python code via Bash for every step" to handle complex data wrangling, which involves the dynamic generation and execution of code at runtime based on task context and external data inputs.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and parse data from a wide array of untrusted external sources including public scientific APIs and complex genomics/omics file formats, creating a potential surface for indirect prompt injection from maliciously crafted data files.
Ingestion points: Scientific data is retrieved via requests and parsed using libraries like pandas, BioPython, and pysam as demonstrated in SKILL.md and references/specialized-domains.md.
Boundary markers: The provided parsing logic lacks explicit delimiters or instructions to the LLM to ignore potentially executable or instructional content embedded within the data.
Capability inventory: The skill utilizes network operations, file system access, and command-line code execution.
Sanitization: No specific sanitization or strict schema validation is described for the external data being processed.- [EXTERNAL_DOWNLOADS]: The skill fetches scientific configuration and data from well-known and reputable organizations including the National Center for Biotechnology Information (NCBI), European Bioinformatics Institute (EBI), National Cancer Institute (NCI), and the Centers for Disease Control and Prevention (CDC).

tooluniverse-data-wrangling