The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill documents how to ingest and process external data using powerful platform capabilities. Ingestion points: External files and records retrieved via 'dx download' and the 'dxpy.DXFile' class. Boundary markers: No delimiters or input-ignoring instructions are present in the provided documentation or code samples. Capability inventory: Jobs executed in isolated Linux VMs with root access, full internet access, subprocess execution ('subprocess.check_call' in 'app-development.md'), and the ability to spawn remote subjobs ('dxpy.new_dxjob'). Sanitization: No input validation or sanitization logic is demonstrated for external data.
External Downloads (LOW): The documentation contains instructions for installing the 'dxpy' library and other bioinformatics tools via 'pip' and 'conda' from sources outside the defined trusted scope.
Remote Code Execution (LOW): The skill details standard platform procedures for remote deployment and execution of code via 'dx build' and 'dxpy.DXApplet.run'.
Command Execution (LOW): Multiple examples illustrate the use of the 'dx' CLI and Python 'subprocess' module to execute system commands within the execution environment.

dnanexus-integration