The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documents and implements data persistence using the Python pickle module. Specifically, references/workflow_guide.md contains examples of loading analysis results using pickle.load(). Insecure deserialization via pickle can be exploited for arbitrary code execution if the input file originates from an untrusted source or has been tampered with.\n- [COMMAND_EXECUTION]: The skill provides a Python script (scripts/run_deseq2_analysis.py) for executing the differential expression pipeline. The script performs file-based operations and data processing tasks based on command-line arguments.\n- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it ingests untrusted external data from CSV and AnnData files. While no direct command injection was found, the combination of external data ingestion and file-writing capabilities creates a potential path for multi-step exploitation.\n
Ingestion points: Reads count matrices and sample metadata from CSV files (counts.csv, metadata.csv) and AnnData files (data.h5ad).\n
Boundary markers: The instructions lack specific delimiters or guardrails to prevent the agent from interpreting instructions that might be embedded within the processed data files.\n
Capability inventory: The skill includes a script (scripts/run_deseq2_analysis.py) that performs local file system reads and writes for processing genomic data.\n
Sanitization: External data is processed using standard methods such as pandas.read_csv without implementation-specific sanitization or content validation for malicious payloads.\n- [EXTERNAL_DOWNLOADS]: The skill references the pydeseq2 package for installation and points to official documentation and source code hosted at well-known domains, including pydeseq2.readthedocs.io and github.com/owkin/PyDESeq2.

pydeseq2