The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading and installing external software and packages from remote repositories.
It instructs the user to clone the voicebox repository from GitHub (jamiepine/voicebox.git).
It uses npx create-video@latest to download and execute the Remotion project bootstrapper from the npm registry.
[COMMAND_EXECUTION]: The skill incorporates a dynamic code execution workflow through the use of Remotion. This involves creating React-based video templates that are rendered into final assets by injecting data from external feeds (JSON arrays). This pattern of mixing executable code with external input is a known security risk if the input data is not strictly validated.
[COMMAND_EXECUTION]: The skill provides example shell commands for executing local CLI tools and system utilities.
Example: node tools/clis/google-ads.js reports get ... to fetch performance data.
Example: ffmpeg commands for combining audio and video tracks.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted performance data from various sources and uses it to influence the generation of new ad variations.
Ingestion points: The skill reads data from CSV files, pasted text, and API outputs (SKILL.md).
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore embedded commands within the ingested performance data.
Capability inventory: The skill has the capability to execute shell commands (node, ffmpeg, npx) and interact with ad platform APIs.
Sanitization: The skill documentation does not specify any sanitization, escaping, or validation steps for the external content before it is processed by the agent.

ad-creative