The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external bank export files (CSV, OFX, QIF) where data fields like transaction descriptions or memos can be used as a vector for prompt injection.
Ingestion points: Bank export files loaded from user-specified paths as described in the Workflow section of SKILL.md.
Boundary markers: The instructions lack explicit delimiting or instructions to the agent to disregard natural language commands that may be present within the transaction data.
Capability inventory: The skill utilizes the transaction_search and categorize tools and performs direct database insertions, which could be influenced by malicious content in the parsed files.
Sanitization: There is no evidence of content sanitization, filtering, or validation for the transaction text fields before they are processed by the agent or secondary tools.

import-transactions