The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestrator script scripts/run_pipeline.py uses subprocess.run to execute internal Python scripts (clean_data.py and visualize_data.py). This is a standard method for managing multi-step workflows but involves execution of local scripts.
Evidence: subprocess.run([sys.executable, str(clean_script)], check=False) in scripts/run_pipeline.py.
[PROMPT_INJECTION]: The skill processes external data files, which presents a surface for indirect prompt injection (Category 8) where malicious instructions embedded in data could influence future agent steps.
Ingestion points: The scripts/clean_data.py script reads files from problem_files/ and crawled_data/ using pandas (read_csv, read_excel).
Boundary markers: There are no explicit boundary markers or safety instructions to ignore embedded text instructions within the data being cleaned.
Capability inventory: The skill is limited to reading from input directories, performing data transformations, and writing results to paper_output/. It lacks network access or high-privilege system capabilities.
Sanitization: Standard data cleaning is performed (missing value imputation, type conversion), but there is no specific sanitization to remove potential prompt injection content from string columns.

data-cleaning-and-visualization