The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data that typically originates from external sub-merchants, such as business names, descriptions, and metadata fields. This creates a surface for indirect prompt injection if the processed content contains instructions designed to manipulate the agent.
Ingestion points: The business_name, description, and metadata parameters in the POST /subaccount and PUT /subaccount endpoints defined in SKILL.md.
Boundary markers: The skill does not provide specific instructions or delimiters to the agent to ignore potentially malicious instructions embedded within these fields.
Capability inventory: The skill performs network operations via the paystackRequest helper to interact with the Paystack API.
Sanitization: No explicit sanitization or validation of the descriptive input fields is mentioned in the instructions.

paystack-subaccounts