The Agent Skills Directory

[COMMAND_EXECUTION]: The skill demonstrates the use of the host.compute API to execute Python scripts (e.g., pipeline.py) on remote compute instances. This is a standard platform feature utilized to offload large-scale genomic data processing to specialized hardware.
[DATA_EXFILTRATION]: The remote compute workflow involves transferring local datasets (.h5ad files) to an external compute provider (byoc:modal). This transmission is a necessary component of the skill's primary function to facilitate high-performance analysis on remote infrastructure.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface through the ingestion of AnnData files.
Ingestion points: The agent is instructed to read external files via sc.read_h5ad("dataset.h5ad") in SKILL.md.
Boundary markers: No explicit delimiters or boundary markers are used to separate untrusted data in the file's metadata from the agent's instructional context.
Capability inventory: The skill allows remote script execution via host.compute.submit_job and local file creation/modification via .write_h5ad().
Sanitization: The provided instructions and helper scripts do not include sanitization or validation logic for data read from the .h5ad metadata fields.

scvi-tools