The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches media assets, brand icons, and configuration data from several domains including HeyGen, SVGL, JSDelivr, GitHub, and DuckDuckGo. These are trusted vendor domains or well-known services used for their intended functionality, such as retrieving official brand logos or downloading color lookup tables (LUTs).
[COMMAND_EXECUTION]: The skill orchestrates complex media workflows by spawning local tools such as ffmpeg, ffprobe, heygen-cli, python3, and npx. Security review confirms that these tools are invoked using array-based argument passing (via execFileSync and spawn), which prevents shell injection vulnerabilities by avoiding the use of the system shell for command parsing.
[DATA_EXFILTRATION]: Usage metrics and account identifiers (HeyGen account email or username) are transmitted to PostHog for analytics purposes. This behavior is documented in the skill's metadata and maintainer-facing references. Users can opt out of this tracking using the HYPERFRAMES_NO_TELEMETRY or DO_NOT_TRACK environment variables.
[PROMPT_INJECTION]: The OpenAI Codex image generation provider accepts natural language intent strings directly into a prompt template for the codex exec CLI. While this creates an indirect prompt injection surface where a malicious intent could attempt to manipulate the downstream tool's behavior, it is a low-risk finding intrinsic to the primary purpose of the media generation feature.
[REMOTE_CODE_EXECUTION]: The skill automates the setup of local AI capabilities, including building the whisper.cpp transcription tool from source and installing various Python packages (e.g., google-genai, mflux). These actions are performed to enable high-performance, on-device media processing and are triggered through explicit setup or resolution commands.

media-use