The Agent Skills Directory

[COMMAND_EXECUTION]: The build script (scripts/build.py) executes binaries found within the user-provided project directory (--fe). Specifically, it searches for and runs node_modules/.bin/tailwindcss. If the provided project contains a malicious binary at that path, it would be executed by the agent. Additionally, it executes the sips utility on macOS for image processing.
[DYNAMIC_EXECUTION]: The scripts/build.py script generates a temporary TypeScript configuration file (tw.config.ts) that dynamically imports the project's existing Tailwind configuration (tailwind.config.ts, etc.). This configuration code is then executed by the Tailwind CLI. This represents a risk if the project's configuration file contains malicious code.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and process the entire frontend codebase (JSX, TSX, Vue, etc.) to port components to vanilla JavaScript. This ingestion of untrusted codebase data creates a surface for indirect prompt injection, where malicious instructions embedded in source code comments or strings could influence the agent's behavior during the porting process.
Ingestion points: Project source files (JSX, TSX, Vue, routes) located in the directory specified by --fe.
Boundary markers: No specific boundary markers or sanitization logic for the source code content are present; the agent is instructed to "read HẾT" (read all) files.
Capability inventory: The skill can execute shell commands via subprocess.run (Python) and control a browser via Playwright (Node.js).
Sanitization: The skill lacks sanitization for the codebase content it processes before interpreting it.
[EXTERNAL_DOWNLOADS]: The build script fetches icon codepoints and font CSS from external sources.
Evidence: Fetches configuration from Google's official Material Design Icons repository on GitHub and font data from Google Fonts APIs.

ui-snapshot