The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data from various high-sensitivity sources (e.g., speech-to-text transcriptions, keyboard activity, and device usage reports), which constitutes an indirect prompt injection surface. If an AI agent processes this content, an attacker could potentially influence its behavior through these data streams.
Ingestion points: Data is fetched via SRSensorReaderDelegate methods in references/sensorkit-patterns.md, specifically through SRFetchResult samples.
Boundary markers: The provided code examples lack boundary markers or instructions to treat sensor data as untrusted content distinct from system prompts.
Capability inventory: The skill code focuses on local data processing and logging; it does not implement risky capabilities such as unauthorized file writes or network exfiltration.
Sanitization: The skill contains no logic for sanitizing or escaping the fetched sensor data before it is handled by the agent.

sensorkit