The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/convert_rds_to_csv.py utilizes the subprocess.run module to execute shell-level commands (calling R and Rscript). This creates a risk of command injection if the input arguments are manipulated by an attacker to include shell-breaking characters or unauthorized commands.
[REMOTE_CODE_EXECUTION]: Within scripts/convert_rds_to_csv.py, the skill dynamically constructs an R script using f-strings that interpolate the rds_path and csv_path variables. This generated script is then written to a temporary file and executed via the system shell. Because these paths are sourced from user-controlled arguments and are not sanitized, an attacker could inject arbitrary R commands into the execution context.
[PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it processes external, untrusted data files. 1. Ingestion points: Data enters the context through scripts/load_count_matrix.py and the conversion script. 2. Boundary markers: There are no explicit instructions or delimiters defined to prevent the agent from following instructions embedded within the count matrices or metadata. 3. Capability inventory: The skill has permissions to write results to disk and execute shell commands via the R bridge. 4. Sanitization: The skill does not implement filtering or validation to detect or escape instruction-like strings within the genomic datasets.

tooluniverse-rnaseq-deseq2