The Agent Skills Directory

[SAFE]: The skill performs legitimate media processing tasks using industry-standard tools like Whisper and FFmpeg. The commands and processes defined in the SKILL.md file are consistent with its stated purpose and do not involve any suspicious file system or network operations.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data by transcribing video audio (SKILL.md). However, the workflow includes a manual review step where the user modifies the text before it is returned to the agent for subtitle generation. This provides a human-in-the-loop boundary that prevents automated execution of instructions that might be embedded in the audio transcription.

videocut-subtitle