The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/detect_all_issues.py orchestrates the profiling workflow by executing multiple internal profiling scripts (such as quality_score.py and distribution_analysis.py) using the subprocess module.
Evidence: The script uses subprocess.run() with list-based command construction and shell=False (default), ensuring safe execution of internal Python scripts.
[PROMPT_INJECTION]: The skill processes untrusted external data files and incorporates raw data samples into its generated JSON reports (e.g., sampled_values and top_value fields). This creates an indirect prompt injection surface if the resulting reports are subsequently processed by an agent without sanitization.
Ingestion points: Untrusted data enters the agent context through the input_path parameter in various profiling scripts, such as scripts/quality_score.py and scripts/deep_quality_analysis.py.
Boundary markers: There are no explicit delimiters or warnings used to isolate external data samples within the profiling reports.
Capability inventory: The skill possesses the capability to execute internal sub-processes via subprocess.run in scripts/detect_all_issues.py and perform file read/write operations.
Sanitization: The skill performs statistical analysis but does not sanitize or filter the content of the data samples (e.g., text values) included in the output reports.

magic-data-profiling