The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/workflow_generator.py script uses Python f-strings to generate Bash shell scripts based on user-supplied parameters. The skill's documentation in SKILL.md instructs users to make these generated scripts executable via chmod +x and run them, which is a common pattern for automation but introduces risk if the generation process is not secure.
[COMMAND_EXECUTION]: The generated workflows execute several command-line utilities, including multiBamSummary, plotCorrelation, plotPCA, bamCoverage, bamCompare, computeMatrix, and samtools. These tools perform intensive file system operations and data processing on genomic alignment files.
[INDIRECT_PROMPT_INJECTION]: The skill contains a vulnerability surface in scripts/workflow_generator.py regarding how external data is handled.
Ingestion points: The script accepts user-controlled strings for BAM filenames (--input-bam, --chip-bams), directory names (--output-dir), and BED files (--genes-bed, --peaks-bed) via command-line arguments.
Boundary markers: There are no boundary markers or validation routines to sanitize these strings before they are interpolated into the Bash script templates.
Capability inventory: The skill has the capability to write new files and generate shell scripts that execute complex command sequences with subprocess privileges.
Sanitization: Sanitization is absent. The input parameters are directly embedded into the shell script templates. Because these variables are not quoted during expansion within the generated shell script (e.g., using $INPUT_BAM instead of "$INPUT_BAM"), an attacker could provide a filename containing shell metacharacters (like semicolons or backticks) to execute arbitrary commands when the generated script is run.

alterlab-deeptools