The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/package-loader.mjs uses import() to load Node.js modules from paths resolved at runtime. This allows the skill to execute code from packages installed into temporary directories during the bootstrapping process.
[REMOTE_CODE_EXECUTION]: The helper utility scripts/package-loader.mjs is designed to download and install external packages from the NPM registry at runtime using npm install. While it incorporates security best practices such as version pinning and disabling lifecycle scripts (--ignore-scripts), the ability to fetch and execute third-party code at runtime remains a significant security surface.
[INDIRECT_PROMPT_INJECTION]: The design picker template (templates/design-picker.html) renders AI-generated content using innerHTML. This architecture creates a vulnerability surface where a malicious or subverted prompt expansion could inject scripts or event handlers, potentially leading to Cross-Site Scripting (XSS) when the user opens the picker in their browser.
[COMMAND_EXECUTION]: The skill includes scripts that spawn subprocesses to perform legitimate tasks, such as ffmpeg in scripts/extract-audio-data.py for audio decoding and npm in scripts/package-loader.mjs for environment setup. These commands are executed using safe patterns (argument arrays) but involve interacting with the host system's shell environment.

hyperframes-creative