The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied natural language requests to determine project types, select technology stacks, and coordinate specialized agents to generate code and documentation. This creates a surface for indirect prompt injection where adversarial inputs could influence the generated project structure or code logic. • Ingestion points: User requests are processed in project-detection.md to extract keywords for template selection. • Boundary markers: There are no explicit boundary markers or instructions defined in the provided files to prevent the AI from following commands embedded within the user request during the planning phase. • Capability inventory: Orchestrated agents have the capability to perform file writes, database schema definitions, and package installations across all templates. • Sanitization: The skill performs keyword matching but lacks specific sanitization or validation logic for the natural language input.
[COMMAND_EXECUTION]: The setup steps in numerous template files (e.g., templates/nextjs-fullstack/TEMPLATE.md, templates/python-fastapi/TEMPLATE.md) involve the execution of shell commands such as npx create-next-app, npm install, and pip install. While these are standard for application scaffolding, they result in the execution of code from external registries based on the generated project configuration.

app-builder