edit-video

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes ffmpeg and ffprobe through various shell command recipes and a local Python script (scripts/check_blended_cuts.py). The script is used to detect crossfades or dissolves in finished videos. It executes external tools using subprocess.run with a list of arguments, which is the recommended secure way to avoid shell-related vulnerabilities when handling user-provided file paths.\n- [EXTERNAL_DOWNLOADS]: The documentation contains references to numerous external tools and assets, including open-source video editors (Kdenlive, Shotcut), developer resources on GitHub, and community-developed plugins (Reactor, Video Copilot). These are provided for the user to enhance their creative workflow and point to established, legitimate sites.\n- [DYNAMIC_EXECUTION]: The skill provides extensive guidance on automating DaVinci Resolve using its Python and Lua APIs. The instructions include robust security advice, such as working within throwaway projects, using local Python interpreters, and verifying API return values to handle potential failures gracefully. It explicitly discourages unsafe practices like passing code via stdin heredocs.\n- [INDIRECT_PROMPT_INJECTION]: The skill involves processing external video files and subtitle data. While the ingestion of untrusted file metadata through tools like ffprobe represents a potential attack surface for indirect prompt injection, the skill does not contain instructions that would cause the agent to blindly execute commands embedded in such data. The workflow maintains clear boundaries between the data being processed and the agent's instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 02:19 AM