The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions involve interpolating user-provided strings (such as app names, context names, and feature names) directly into shell commands (e.g., pnpm dlx @nestjs/cli@latest new <app-name>). This is a standard indirect prompt injection surface where a malicious user could attempt command injection through crafted naming. This risk is inherent to the primary purpose of a CLI-driving skill.
Ingestion points: User-supplied parameters for context, feature, and application names within SKILL.md and references/new-project.md.
Boundary markers: No specific boundary markers or input validation instructions are provided for these variables.
Capability inventory: The skill uses pnpm exec nest and pnpm dlx to execute commands in the shell.
Sanitization: No explicit sanitization or escaping instructions for user input are present.
[EXTERNAL_DOWNLOADS]: The skill fetches and executes the official NestJS CLI and related schematics using pnpm. These are well-known packages from a reputable source.
Evidence: Instructions include pnpm dlx @nestjs/cli@latest new ... and pnpm add -D @nestjs/schematics.
[DATA_EXPOSURE]: The skill mentions the use of .env files for configuration. It correctly advises creating .env.example files and ensuring that local credentials are ignored (e.g., via .gitignore), which aligns with security best practices for credential management.
Evidence: references/new-project.md instructs to create a local .env with "ignored local credentials" and use it with nest start --env-file .env.

nestjs-cli