The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local system utilities for non-malicious tasks. Evidence: scripts/check_video.py calls ffprobe to validate video technical specifications. Evidence: scripts/notify_user.py uses osascript and say on macOS to provide user notifications and audio feedback.
[INDIRECT_PROMPT_INJECTION]: The skill automates interactions with external web platforms, creating a potential surface for indirect prompt injection from platform-controlled data. Ingestion points: Browser DOM snapshots and video metadata. Boundary markers: The skill instructions require the agent to generate and display a field verification table. Capability inventory: Browser automation (form filling and submission) and local command execution. Sanitization: A mandatory awaiting_confirmation human review step is implemented to verify the final publication state, effectively mitigating risks of automated obedience to malicious external instructions.

lov-media-publisher