The Agent Skills Directory

[SAFE]: The code performs standard file reading and writing operations using Python's pathlib library. It reads from a path provided via command-line arguments and writes to another, which is expected behavior for a data processing utility.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process business signals from external JSON or CSV files. While this represents a data ingestion surface, the current implementation is a safe scaffold that does not interpolate this data into LLM prompts or use it in dangerous execution contexts.
Ingestion points: run.py reads data from the path specified in the --input argument.
Boundary markers: Not applicable in this scaffold implementation.
Capability inventory: File reading and writing via pathlib. No network or subprocess capabilities are present.
Sanitization: The skill uses standard json.loads for parsing input data.

analytics-budget-analyzer