The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external bioinformatics data files, creating a vulnerability surface for indirect prompt injection if those files contain malicious instructions or unexpected metadata strings.
Ingestion points: sample.bam, regions.bed, genome.fa, and genes.bed are utilized as data inputs in the provided Bash, Python, and R code examples.
Boundary markers: Absent. There are no instructions to the agent to disregard potential instructions embedded within the metadata or content of the genomic files.
Capability inventory: The skill directs the execution of CLI tools like nucleoatac, danpos.py, and bedtools, and utilizes Python/R libraries for deep file parsing.
Sanitization: Absent. The skill does not provide methods for sanitizing the input data before processing.
[EXTERNAL_DOWNLOADS]: The skill requires downloading and installing research software from external GitHub repositories.
Evidence: Instructions include cloning scPrinter from github.com/buenrostrolab/scPrinter and using DANPOS3 from github.com/sklasfeld/DANPOS3.
Note: These are reputable academic sources within the field of genomics, but the pattern of cloning and execution represents a potential supply chain risk if the repositories were compromised.
[COMMAND_EXECUTION]: The skill facilitates the execution of various local shell commands for scientific computation.
Commands include nucleoatac run, bedtools slop, and python danpos.py dpos.
This is expected behavior for bioinformatic analysis but grants the agent direct access to the shell and file system.

bio-atac-seq-nucleosome-positioning