The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructions in SKILL.md include commands that require elevated privileges or modify security trust settings.
Step 7 recommends sudo apt install ffmpeg for Debian-based systems to install necessary media processing tools.
Step 6 includes brew trust homebrew-ffmpeg/ffmpeg, which is an attempt to bypass security prompts when adding third-party software taps on macOS.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to download and integrate external code and other AI agent skills from third-party organizations.
Instructions include git clone https://github.com/browser-use/video-use to install a specific video editing engine.
Instructions also include npx skills add for heygen-com/hyperframes and remotion-dev/skills to expand motion graphics and captioning capabilities.
[COMMAND_EXECUTION]: The provided Python tools (mediakit.py, probe.py, deadspace.py, etc.) make extensive use of the subprocess module to execute system commands.
The tools call ffmpeg, ffprobe, and node to perform media processing and verify the environment.
The implementation uses list-based arguments for command construction, which mitigates direct shell injection, but the skill still grants the agent broad control over media processing pipelines.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain malicious instructions designed to influence agent behavior.
Ingestion points include user-provided video files (take.mov) and transcript files (take.words.json).
While the agent is instructed to read phrases from the transcript to understand the take, malicious content within these files could potentially influence agent behavior during the planning or review phases.

edit