The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-provided genomic data files in CSV, TSV, and AnnData (H5AD) formats.
Ingestion points: pd.read_csv and ad.read_h5ad are utilized in the instructions and the accompanying script scripts/run_deseq2_analysis.py to load count matrices and experimental metadata.
Capability inventory: The skill executes statistical computations and writes results to the local filesystem using to_csv and pickle.dump.
Sanitization: The scripts/run_deseq2_analysis.py script includes validation checks for non-negative integer counts, though it does not explicitly sanitize categorical metadata for potential prompt-injection content.
Boundary markers: No explicit prompt boundary markers or "ignore instructions" warnings are applied to the external data processing steps.
[DYNAMIC_EXECUTION]: The documentation demonstrates the use of Python's pickle module for maintaining analysis state.
Evidence: The references/workflow_guide.md file contains examples of pickle.load being used to restore counts and metadata from local files.
Context: While pickle.load can execute arbitrary code if used on untrusted files, its implementation here is intended for persisting complex scientific objects within the user's local environment, aligning with the primary purpose of the skill.

pydeseq2