The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to construct an awk command using a variable <n> provided by the user without validation or escaping.
Evidence: In Step 1, the instruction uses: awk -F'\t' 'BEGIN {OFS="\t"} {print $1, $2, $3, $<n>/100}'. A malicious user could provide input for <n> that terminates the print statement and executes arbitrary system commands via awk's system() function.
[PROMPT_INJECTION]: The skill processes external genomic data files, which serves as an ingestion point for potentially untrusted content.
Ingestion points: The skill reads input files (sample.bed) provided by the user in Step 1.
Boundary markers: There are no markers or instructions to treat the file content as purely data or to ignore embedded instructions.
Capability inventory: The skill utilizes shell commands (awk, sort) and several external tools (mcp__methyl-tools__generate_metilene_input, mcp__methyl-tools__run_metilene, mcp__methyl-tools__filter_dmrs, etc.) which could be subverted if file content influences their execution.
Sanitization: No validation or sanitization of the input file content is specified before the data is passed to processing tools.

differential-methylation