The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The scripts/package-loader.mjs utility is used to dynamically bootstrap and install Node.js dependencies (such as @hyperframes/producer and sharp) from the NPM registry at runtime. While it enforces pinned versions and uses the --ignore-scripts flag to mitigate some risks, the ability to download and execute code during skill execution is a significant capability that could be abused if package names or versions were manipulated.
[COMMAND_EXECUTION]: Multiple scripts within the skill (package-loader.mjs, animation-map.mjs, contrast-report.mjs) utilize node:child_process (spawnSync) to programmatically execute system commands like npm and node. Additionally, the references/design-picker.md guide instructs the agent to launch a local HTTP server using python3 -m http.server to host a design selection interface for the user.
[INDIRECT_PROMPT_INJECTION]: The skill's primary workflows involve ingesting data from external, potentially untrusted sources, including crawling websites (/website-to-video, /product-launch-video) and fetching GitHub Pull Request data (/pr-to-video).
Ingestion points: External URLs and GitHub PR descriptions/diffs are processed via headless Chrome and the gh CLI.
Boundary markers: The instructions do not explicitly detail robust boundary markers or safety instructions to separate untrusted external content from the agent's core instructions during the prompt expansion phase.
Capability inventory: The skill has broad capabilities across its scripts, including file system access, network requests (via the crawler), and shell command execution.
Sanitization: There is no documentation of rigorous sanitization of the external content before it is used to influence the generation of the video composition plan.
[EXTERNAL_DOWNLOADS]: The skill instructions and scripts refer to downloading various resources, including Google Fonts metadata and NPM packages, which introduces a dependency on external services and registries. These are documented as standard operations for the skill's functionality.

hyperframes