The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied genomic files (BAM, VCF, etc.), which represents a potential surface for indirect prompt injection.\n
Ingestion points: Bundled scripts such as alignment_qc.py and variant_summary.py read user-provided genomic files from the local filesystem.\n
Boundary markers: Data parsing relies on HTSlib's implementation of standard biological file formats; the skill does not implement specific LLM boundary markers for these data streams.\n
Capability inventory: The skill has the ability to write files and execute bioinformatics commands through the pysam.samtools and pysam.bcftools wrappers.\n
Sanitization: File parsing is performed by the HTSlib library. The skill documentation explicitly recommends using parameterized API calls rather than constructing shell commands from untrusted input.\n- [COMMAND_EXECUTION]: The skill utilizes the pysam library to execute samtools and bcftools commands.\n
Evidence: Documented use of pysam.samtools.sort, pysam.samtools.index, and pysam.bcftools.index across the reference guides and scripts.\n
Context: These calls are standard bioinformatics operations. The skill includes explicit security guidance to prevent shell injection by passing arguments as separate strings and avoiding shell command evaluation.\n- [EXTERNAL_DOWNLOADS]: The skill documents how to access remote genomic data and reference sequences.\n
Evidence: Mentions of REF_PATH, REF_CACHE, and HTTPS URLs for data access are found in the performance and CRAM reference documentation.\n
Context: These are standard industry practices for genomic analysis. The skill author notes that the underlying library has disabled implicit remote fetching by default to improve security and auditability.

pysam