The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents administrative setup procedures that involve system-level commands. This includes instructions for creating system-wide cache directories and managing permissions using sudo (e.g., sudo mkdir -p /system/settings and sudo chmod 755 /shared/cache/lamindb in references/setup-deployment.md). These are standard for multi-user system configuration and are documented neutrally for deployment purposes.\n- [CREDENTIALS_UNSAFE]: The documentation provides templates and instructions for managing authentication credentials. It includes placeholder connection strings for PostgreSQL databases (e.g., postgresql://user:password@hostname:5432/dbname) and guidance on setting environment variables for cloud service access (AWS S3, Google Cloud Storage). It also references the local storage of API tokens in ~/.lamin/ for authentication management.\n- [EXTERNAL_DOWNLOADS]: The skill describes workflows that involve installing Python packages from official registries and interacting with remote storage. Instructions include using pip to install packages such as lamindb, bionty, and various integration modules. It also details connecting to external data sources via cloud storage URIs and HTTP/HTTPS endpoints.\n- [SAFE]: The skill is designed to process external biological datasets, which constitutes a surface for indirect prompt injection. \n
Ingestion points: External data enters the context through loading functions such as ad.read_h5ad and pd.read_csv in SKILL.md and reference files.\n
Boundary markers: The instructions do not define specific delimiters for external data input.\n
Capability inventory: The agent can perform file system operations, database interaction, and connectivity with MLOps platforms like Weights & Biases or MLflow.\n
Sanitization: The skill focus is explicitly on data curation and validation using curators (e.g., AnnDataCurator), which effectively sanitizes data against predefined schemas.

lamindb