The Agent Skills Directory

External Package Installation: The skill references several official Node.js packages and CLI tools from the Genkit ecosystem. These are standard dependencies for developing with the framework and are hosted on the public NPM registry. Evidence: Mentions npm install -g genkit-cli and various @genkit-ai/* packages across multiple files (e.g., SKILL.md, references/a2ui.md).
Indirect Prompt Injection Surface: As a framework for building AI agents, the documented patterns involve processing user-supplied data which could potentially contain malicious instructions. The skill includes specific guidance on maintaining security boundaries, particularly for generative UI components. (1) Ingestion points: Data enters the system via chat.send() calls in references/agents.md and user actions in references/a2ui.md. (2) Boundary markers: The framework utilizes Zod schemas for structural validation of inputs and outputs. (3) Capability inventory: The documentation covers middleware that grants agents access to the filesystem and shell execution for development tasks. (4) Sanitization: The references/a2ui.md file explicitly advises that renderers are responsible for sanitizing output before it reaches the DOM.
System Capability Documentation: The skill describes advanced features like filesystem access and shell tool integration. These are presented as opt-in middleware capabilities intended for creating powerful AI assistants, with recommended guardrails such as sandboxing and human-in-the-loop approval.

developing-genkit-js