The Agent Skills Directory

[COMMAND_EXECUTION]: The RendererOptions interface exposes the ffmpegPath and browserConfig.executablePath parameters, which allow the user (or the agent) to specify the file path for the FFmpeg and browser binaries to be executed. If the agent is tricked into using a malicious script or binary at these paths, it could lead to arbitrary code execution during rendering or diagnostics.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from an external source, creating a vulnerability where instructions embedded in the target page could influence agent behavior.
Ingestion points: The compositionUrl parameter in the render method (found in SKILL.md).
Boundary markers: None are present to delimit untrusted web content from agent instructions.
Capability inventory: The skill can execute subprocesses (via ffmpegPath and executablePath) and write files to the local system (via outputPath).
Sanitization: No sanitization or validation of the content at the compositionUrl is described.

helios-renderer