The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (MEDIUM): The file rules/transcribe-captions.md contains code instructions to programmatically install the whisper.cpp native binary and download specific models. Executing native binaries from non-trusted organizations (remotion-dev) at runtime is a significant security boundary crossing.
[COMMAND_EXECUTION] (MEDIUM): Multiple files (e.g., rules/3d.md, rules/ffmpeg.md, rules/fonts.md, rules/parameters.md) instruct the agent to execute shell commands using npx, bunx, and ffmpeg. These instructions are used for dependency management and media processing but involve running arbitrary external code.
[CREDENTIALS_UNSAFE] (LOW): The file rules/voiceover.md instructs the agent to request an ELEVENLABS_API_KEY from the user and store it in a .env file. While required for the primary purpose of the skill, handling and storing API keys via the agent context requires caution.
[EXTERNAL_DOWNLOADS] (MEDIUM): The skill frequently encourages downloading packages from the @remotion/* and mediabunny namespaces. While these are part of the Remotion ecosystem, they are not listed as trusted sources and represent unverifiable external dependencies.
[INDIRECT_PROMPT_INJECTION] (LOW): The file rules/tailwind.md explicitly instructs the agent to fetch instructions from an external URL (remotion.dev/docs/tailwind) via WebFetch. If the content of that page is compromised or altered, it could inject malicious instructions into the agent's session.

remotion-best-practices