The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions and reference materials describe the use of subprocess.check_call to execute bioinformatics tools like samtools, bwa, and quality_filter. This is a core capability intended to facilitate genomic data processing.
[EXTERNAL_DOWNLOADS]: The skill manages dependencies by downloading official packages from the DNAnexus GitHub repository, Ubuntu package registries via apt-get, and container images from Docker Hub. These activities target well-known and trusted service providers.
[DYNAMIC_EXECUTION]: The skill utilizes a platform-specific plugin architecture where Python entry points are defined using @dxpy.entry_point decorators. It also supports spawning subjobs and workflows, which is standard for distributed cloud computing environments.
[INDIRECT_PROMPT_INJECTION]: The skill has ingestion points for untrusted data through genomics files (FASTQ, BAM, VCF) and structured records.
Ingestion points: Data enters the context via dxpy.download_dxfile() and record.get_details().
Boundary markers: Explicit delimiters for embedded natural language are not documented, though genomics data is primarily binary or structured scientific data.
Capability inventory: The skill has access to shell execution (subprocess), network operations, and project-level API calls.
Sanitization: While scientific files are not typically sanitized for natural language instructions, the risk of cross-domain injection in this context is negligible.

dnanexus-integration