The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external data through the scripts/latency_arbitrage_diagnostics.py script which processes CSV files provided by the user.
Ingestion points: The Python script reads untrusted data from a CSV file specified via command-line arguments.
Boundary markers: There are no explicit boundary markers or instructions to the agent regarding the separation of data and instructions in the input CSV.
Capability inventory: The script performs local file reading and writing. It does not perform network operations, subprocess execution, or arbitrary code evaluation.
Sanitization: The script uses pd.to_numeric and pd.to_datetime with errors='coerce', which ensures that any non-numeric or non-date injection attempts are discarded or replaced with NaN before the data is summarized and returned to the agent's context.

latency-arbitrage