The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements an interactive workflow that ingests user-defined values (folder names, project names, entity fields) to construct and execute shell commands and generate code files. It lacks explicit instructions for the agent to sanitize or validate these inputs before interpolation into shell commands (e.g., mkdir -p [폴더명]), which creates a surface for indirect prompt injection or command injection.
Ingestion points: User-provided folder names, project names, and custom entity field definitions in SKILL.md.
Boundary markers: Absent. No explicit instructions to the agent to treat user inputs as untrusted strings.
Capability inventory: npx, npm install, mkdir, and file-writing capabilities via the agent's environment.
Sanitization: Absent. The skill does not provide patterns for escaping or validating user inputs.
[COMMAND_EXECUTION]: The skill's primary function is to execute shell commands to scaffold a development environment. It uses npx create-next-app@latest, npm install, and npx shadcn@latest. While these are legitimate developer tools, the execution of arbitrary commands provided in user-specified paths is a high-capability behavior.
[SAFE]: All referenced libraries (Zustand, Drizzle ORM, Tanstack Query, Better Auth) and initialization tools (Next.js, ShadCN) are industry-standard, well-known services and packages from trusted ecosystem providers.

nextjs15-init