The Agent Skills Directory

[COMMAND_EXECUTION]: The tests/test_scripts.py file uses subprocess.run to execute the skill's utility scripts for testing purposes. This is implemented securely using sys.executable and argument lists without invoking a shell.
[PROMPT_INJECTION]: The skill processes external data files (CSV, Parquet, Excel, JSON) which are potential vectors for indirect prompt injection.
Ingestion points: The load functions in scripts/profile_data.py and scripts/baseline_model.py read data using the pandas library.
Boundary markers: The skill requires a framing brief (project-brief.md) and a mirror step to confirm assumptions with the user, alongside a mandatory audit gate (scripts/profile_data.py).
Capability inventory: Capabilities include file system read/write access and local command execution (limited to tests).
Sanitization: The skill relies on standard, robust parsing libraries (pandas, pyarrow) for data ingestion.
This attack surface is mitigated by the skill's disciplined workflow, which includes a mandatory red-team pass (checklists/analysis-review.md) to challenge and verify all findings.
[SAFE]: No obfuscation, hardcoded credentials, or unauthorized network operations were detected. All dependencies are standard, well-maintained data science libraries.

data-scientist