The Agent Skills Directory

External Resource References: The documentation describes installing the Genkit CLI via a script from a vendor-controlled domain (cli.genkit.dev). This is a standard installation method for the development environment. In the context of Model Context Protocol (MCP) integration, the skill also provides examples of using common protocol servers through standard package runners like npx.
Command Execution Considerations: The skill instructs the user on how to run, trace, and debug applications using the Genkit CLI (genkit start, genkit flow:run) and Dart's development ecosystem. These tools are used for their intended purpose of local development, testing, and telemetry capture.
Indirect Prompt Injection Surface: As an SDK for building AI agents, the skill facilitates the creation of systems that ingest and process external data.
Ingestion points: Agents defined using this skill may ingest data from user messages, local filesystems, or external MCP servers (as seen in SKILL.md and references/genkit_mcp.md).
Boundary markers: The documentation for Generative UI (references/a2ui.md) mentions the use of catalog-based validation to ensure model outputs match expected schemas.
Capability inventory: Built agents may have capabilities such as filesystem access (references/genkit_middleware.md), tool calling, and network communication.
Sanitization: The documentation highlights security best practices, advising developers to treat model-generated surfaces as untrusted input and suggesting the use of host-level security policies (such as CSP) to mitigate potential risks.

developing-genkit-dart