The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The mandatory 'Agent Workflow' in SKILL.md requires agents to analyze untrusted codebase files (e.g., vite.config.js, package.json) via explore-codebase. Given the agent's capability to implement configurations and execute build scripts, this facilitates a high-severity indirect prompt injection vulnerability where malicious data in the repository can subvert the agent's logic. Evidence: Ingestion point (SKILL.md workflow), missing boundary markers, and write/execute capabilities across multiple templates.
[External Downloads] (LOW): Templates in setup.md and InertiaSetup.md recommend standard npm packages (e.g., vite, laravel-vite-plugin) from the official registry. Following the [TRUST-SCOPE-RULE], these are considered low risk as they originate from whitelisted/common sources.
[Command Execution] (MEDIUM): templates provide commands for building production assets (npm run build) and configuring server-side rendering (SSR) via process managers like Supervisor and PM2, which involve shell execution for persistence.
[Data Exposure] (LOW): references/environment.md provides guidance on using VITE_ prefixes and includes explicit warnings against exposing sensitive data like API keys or passwords, aligned with security best practices.

laravel-vite