The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a delegate-based communication system that ingests data from external wearable devices, which could serve as a vector for indirect prompt injection if the incoming data contains malicious instructions intended for the agent.\n
Ingestion points: The IWearableDelegate interface and its methods OnMessageReceived, OnContextReceived, OnTransferReceived, and OnFileReceived (as described in SKILL.md and reference/api-reference.md) process data payloads originating from paired watches.\n
Boundary markers: The provided code templates do not include specific delimiters or system instructions to distinguish between data and potential commands within the wearable payloads.\n
Capability inventory: The skill provides capabilities to write received files to the local file system (OnFileReceived moves files to AppData/Shiny.Wearables) and to communicate with external nodes via SendMessage and Transfer commands.\n
Sanitization: Although the skill encourages structured data processing via JSON element access, it lacks guidance on sanitizing or escaping natural language content received from the wearable device before it is processed by the agent.

shiny-wearables