The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation guide in SKILL.md provides instructions to download and execute a setup script for the HeyGen CLI tool using 'curl -fsSL https://static.heygen.ai/cli/install.sh | bash'. This follows the standard installation pattern for this well-known AI video service provider.
[EXTERNAL_DOWNLOADS]: The skill dynamically fetches media files from HeyGen's servers based on search results. The download process in 'scripts/lib/freeze.mjs' includes safety checks, such as a 256MB file size limit to prevent potential denial-of-service through disk exhaustion.
[COMMAND_EXECUTION]: The scripts 'probe.mjs' and 'heygen-search.mjs' execute the external binaries 'ffprobe' and 'heygen' respectively. These operations use 'execFileSync', which passes arguments as a literal array rather than through a shell, effectively mitigating common command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill accepts natural language descriptions (intents) from the agent to search for media. It correctly implements boundary security by passing these intents as literal command-line arguments to the search tool, ensuring they cannot be misinterpreted as shell commands.
[DATA_EXPOSURE_AND_EXFILTRATION]: The brand provider script reads local project documentation (e.g., 'design.md') to extract visual tokens like hex colors and fonts. This behavior is restricted to the local project directory and is consistent with the skill's purpose of matching assets to a project's style.

media-use