The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted experimental data from CSV and TSV files for processing. Maliciously crafted data could potentially be used to influence the agent's summary, interpretation, or subsequent actions based on the analysis results. \n
Ingestion points: The pd.read_csv() function is utilized in scripts/run_deseq2_analysis.py and the SKILL.md documentation to load count matrices and sample metadata. \n
Boundary markers: There are no explicit boundary markers or isolation instructions to protect the analysis logic from embedded instructions within the data. \n
Capability inventory: The skill possesses the capability to read and write local files, perform intensive statistical calculations, and generate visualizations through Python script execution. \n
Sanitization: Sanitization is minimal, limited to checking for non-negative integers in the count matrix; metadata strings and identifiers are not validated against injection patterns. \n- [COMMAND_EXECUTION]: The provided analysis script and workflow documentation promote the use of the pickle module for data persistence. Deserializing data using pickle from an untrusted or external source is a known security risk that can lead to arbitrary code execution. \n
Evidence: The script scripts/run_deseq2_analysis.py implements pickle.dump() to save the DeseqDataSet object, and the references/workflow_guide.md provides explicit code examples for using pickle.load() to restore analysis states.

alterlab-pydeseq2