The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses pysam.samtools and pysam.bcftools to execute bioinformatics command-line tools. These dispatchers use tokenized string lists for arguments rather than shell strings, preventing command injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill instructions direct the installation of the pysam library via uv pip install "pysam==0.24.1". The use of a pinned version of a well-known bioinformatics library from a standard registry is considered a safe practice.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted genomic data files (SAM, BAM, CRAM, VCF), which constitutes an attack surface for indirect prompt injection via malicious metadata or headers.
Ingestion points: Input files processed by AlignmentFile, VariantFile, and bundled scripts such as scripts/alignment_qc.py.
Boundary markers: Not explicitly present for these binary and highly structured genomic formats.
Capability inventory: The skill possesses capabilities for file writing (AlignmentFile.write), wrapped subprocess execution (pysam.samtools), and network I/O (HTSlib remote access support).
Sanitization: The skill uses tokenized arguments for command dispatchers and utilizes json.dumps for metadata report generation to prevent injection during output.
[SAFE]: The skill implements proactive safety checks in its bundled scripts, such as using exclusive creation mode ('x') to prevent accidental file overwrites and validating that output paths do not collide with input files or existing indexes.

pysam