The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The README.md file provides a quickstart installation command that fetches a script from https://install.iii.dev/iii/main/install.sh. This domain is part of the author's official infrastructure for the 'iii' engine and represents standard installation behavior for the project.
[REMOTE_CODE_EXECUTION]: The documentation includes a pattern where a remote shell script is downloaded and piped to a shell (curl | sh). This is a common installation method for developer tools; since the source is the vendor's verified domain, it is considered a legitimate installer rather than an untrusted execution risk.
[DYNAMIC_EXECUTION]: The framework allows worker processes to inject custom React components and stylesheets into the console at runtime via ESM import() and scoped <link> tags. This dynamic loading of executable content is the intended primary functionality of the skill. The system includes an ErrorBoundary for isolation and relies on the browser's origin-based security model.
[INDIRECT_PROMPT_INJECTION]: The skill implements a chat interface that renders data from potentially untrusted sources, such as AI model responses and tool outputs. The application addresses this attack surface by using react-markdown for sanitization, and the project includes specific end-to-end tests (web/e2e/adversarial-content.spec.ts) to verify that malicious scripts, unsafe protocols (e.g., javascript:), and event handlers (e.g., onerror) are successfully blocked.

console-injectable-ui