The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a workflow where Python scripts are used to generate bash script templates which are then marked as executable and run on the host system. This pattern, found in SKILL.md and scripts/workflow_generator.py, allows for dynamic command generation and execution.\n- [PROMPT_INJECTION]: The scripts/workflow_generator.py script is vulnerable to indirect prompt/command injection. It uses f-strings to interpolate user-provided arguments, such as BAM file paths and genome sizes, directly into bash script templates without any form of sanitization or shell-escaping. If an agent processes a directory containing maliciously named files (e.g., a file named sample;rm -rf.bam), the generated workflow script would execute those unintended commands when run.

deeptools