The Agent Skills Directory

Command Execution: The skill frequently executes system commands via ffmpeg, ffprobe, python, and the heygen CLI to process media. While these are controlled invocations, the use of subprocesses is a potential consideration for system integrity and requires that the host environment has these tools configured securely.
Dynamic Execution Patterns: For generating music via local models, the skill assembles and executes Python scripts. It incorporates user-provided prompts into these scripts using JSON serialization to ensure they are treated as data, which is a proactive measure against potential code injection.
Runtime Dependency Management: The skill can automatically install Python libraries from official registries if required for media generation. This behavior ensures functionality but means that external code is downloaded and installed on the user's system at runtime.
Indirect Prompt Injection Surface: User-supplied natural language intents are used to drive media search and generation. As these inputs influence the operation of external tools and APIs, they represent an area to review for potential indirect prompt influence.
Access to Sensitive Files: To manage authentication for the HeyGen service and maintain user settings, the skill reads and writes configuration files in the user's home directory (e.g., ~/.heygen/credentials). This is required for its intended functionality but involves the handling of sensitive credential data.
Network Operations and Asset Downloads: Media and LUT assets are retrieved from remote servers. The skill implements an SSRF protection utility that validates URLs to ensure they target public internet addresses rather than sensitive internal network resources.

media-use