The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to run specific Python scripts (ai_training_data_audit.py, data_product_strategy_picker.py, and data_asset_valuator.py) to process user-provided data files. These scripts are located within the local file system using relative paths.
[SAFE]: No signs of prompt injection, data exfiltration, or obfuscation were detected. All network-related discussions in the instructions refer to legitimate data infrastructure services (Snowflake, Databricks) within a business context.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data inputs (sources.json, profile.json, corpus.json) which serves as an ingestion surface for untrusted data. This is documented as a functional requirement for the skill's purpose.
Ingestion points: Data is ingested through the <plan> argument and several JSON files (sources.json, profile.json, corpus.json).
Boundary markers: None explicitly defined in the instructions for the input data files.
Capability inventory: The skill has the capability to execute Python scripts in the local environment.
Sanitization: No specific sanitization or validation logic is defined within the prompt instructions for the external JSON content.

cdo-review