The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs users and provides code for using pickle.load() to restore saved analysis states.
Evidence: In references/workflow_guide.md, a code block demonstrates loading a saved result using pickle.load(f). Additionally, SKILL.md and scripts/run_deseq2_analysis.py implement pickle.dump() for saving data objects.
Risk: The Python pickle module is inherently insecure. If a user is tricked into loading a malicious pickle file (e.g., from a shared research repository or a public dataset), it can execute arbitrary code on the user's system.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an ingestion surface for untrusted external data files (CSV, TSV, AnnData).
Ingestion points: The skill uses pd.read_csv() and ad.read_h5ad() in SKILL.md, scripts/run_deseq2_analysis.py, and references/workflow_guide.md to load gene count matrices and experimental metadata.
Boundary markers: Absent. There are no delimiters or specific warnings to ignore embedded instructions within the processed data files.
Capability inventory: The skill has the capability to write files to the local file system (to_csv, pickle.dump) and generate plots. The analysis script run_deseq2_analysis.py allows specifying an output directory.
Sanitization: While scripts/run_deseq2_analysis.py performs basic numeric validation (checking for negative values), non-numeric metadata fields are not sanitized, creating a surface for potential exploitation if data is subsequently used in a prompt-based context.

pydeseq2