The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's documentation in references/workflow_guide.md provides examples and instructions for using the pickle module to save and load analysis states (e.g., pickle.load(f)). Because Python's pickle deserialization is inherently unsafe and can execute arbitrary code, this practice poses a significant risk if the agent or user loads a malicious or untrusted .pkl file.
Evidence in references/workflow_guide.md: with open("counts.pkl", "rb") as f: counts_df = pickle.load(f) and with open("dds_result.pkl", "rb") as f: adata = pickle.load(f).
[PROMPT_INJECTION]: The skill facilitates the processing of external data provided by the user, such as RNA-seq count matrices and metadata in CSV, TSV, or H5AD formats. This creates a surface for indirect prompt injection where instructions hidden in the data could influence the agent's behavior.
Ingestion points: pd.read_csv() and ad.read_h5ad() are used to load external data in SKILL.md, scripts/run_deseq2_analysis.py, and references/workflow_guide.md.
Boundary markers: Absent; there are no specific delimiters or instruction-ignore warnings implemented to isolate the data from the agent's command processing.
Capability inventory: The skill has local file-write capabilities via to_csv and savefig in scripts/run_deseq2_analysis.py.
Sanitization: While scripts/run_deseq2_analysis.py includes basic validation for numeric counts and index matching, it lacks sanitization to detect or neutralize natural language instructions embedded within the datasets.

pydeseq2