The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references biological model weights and code from the official repository of the Wang Lab at the University of Toronto on GitHub. This is an established source for the research community.
[COMMAND_EXECUTION]: The skill orchestrates the execution of Python scripts on a remote compute provider using the platform's compute API. This is the primary mechanism for processing large-scale biological datasets that require high-VRAM GPUs.
[INDIRECT_PROMPT_INJECTION]: The skill processes single-cell expression data stored in AnnData (H5AD) format, which represents a potential ingestion surface for untrusted data.
Ingestion points: Data is read using the anndata.read_h5ad function as described in the loading instructions.
Boundary markers: No specific delimiters or instructions are provided to the agent to ignore potential commands embedded within data fields.
Capability inventory: The skill has the capability to perform network operations through the remote compute API and write files via artifact saving tools.
Sanitization: The skill relies on standard biological data processing libraries and does not explicitly implement additional validation or sanitization of the input file contents.

scgpt