The Agent Skills Directory

[SAFE]: The skill provides documentation and code snippets for the legitimate scvi-tools library and related scientific packages. It includes installation instructions using the standard uv package manager and provides examples for common data analysis tasks such as dimensionality reduction, batch correction, and differential expression.\n- [COMMAND_EXECUTION]: The skill includes shell commands for installing necessary software via uv pip install scvi-tools. These are standard setup procedures for the described functionality and target a well-known public library.\n- [PROMPT_INJECTION]: The skill possesses an ingestion surface for indirect prompt injection as it processes external biological data files through functions like sc.read_h5ad(), sc.read_csv(), and sc.read_10x_mtx().\n
Ingestion points: Data loading functions in workflows.md and models-spatial.md.\n
Boundary markers: No explicit delimiters or instructions are used to separate data content from agent instructions.\n
Capability inventory: The skill allows shell execution for installation (SKILL.md) and file system writes for saving models and results (workflows.md).\n
Sanitization: No validation or sanitization of the content of the biological data files is performed before processing.\n
Despite these factors, the risk is negligible as the skill is designed for structured scientific data analysis and does not provide a mechanism for data to influence the agent's control flow in a malicious way.

scvi-tools