The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The Outbox Pattern implementation in the OutboxProcessor class (SKILL.md) demonstrates a workflow that ingests potentially untrusted data (Payload) from an external repository and publishes it to a message bus. This represents a vulnerability surface where malicious instructions could be embedded in messages and later executed by downstream agents or systems.
Ingestion points: outbox.GetPendingAsync in SKILL.md reads messages from a database.
Boundary markers: The provided code does not implement delimiters or 'ignore' instructions for the payload data.
Capability inventory: The skill uses publisher.PublishAsync to send data to external systems.
Sanitization: The example code lacks explicit validation or sanitization of the Payload before publishing.

background-services