The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external Excel (.xlsx) files, which serves as an ingestion point for potential indirect prompt injection attacks. If a spreadsheet contains malicious instructions in its cells, the agent might interpret them as commands if they are processed into the conversation context.
Ingestion points: Functions such as read_workbook, read_sheet, and read_to_dataframe in scripts/xlsx_utils.py extract data from external file paths provided at runtime.
Boundary markers: There are no explicit markers or developer instructions provided to the agent to treat spreadsheet content as untrusted data or to ignore embedded instructions.
Capability inventory: The skill allows file system writing (wb.save in scripts/xlsx_utils.py), enabling an attacker who successfully injects instructions to potentially overwrite or create local files.
Sanitization: The skill performs no sanitization or filtering of the cell content before returning it to the agent.

xlsx