The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a workflow where the agent audits and reports on Amazon Ads data. This ingestion of external data from a user's Amazon account constitutes a potential surface for indirect prompt injection attacks.
Ingestion points: Sponsored Products data pulled from the user's Amazon Ads account.
Boundary markers: None specified in the provided instructions to differentiate data from instructions.
Capability inventory: The skill relies on the membrane CLI tool to perform audits and generate local reports.
Sanitization: No specific sanitization or validation methods are mentioned for the external data being processed.
[DATA_EXPOSURE]: The skill documentation identifies specific sensitive file paths used by the membrane tool, including ~/.membrane/connections/ which stores Amazon access and refresh tokens. While this is documented as legitimate local storage for the tool's operation, it explicitly identifies the location of sensitive credentials on the user's system.

membrane