DenisSergeevitch/agents-best-practices

The gap in what exists

While many agent frameworks exist, clear, provider-neutral best practices for designing individual 'skills' or components remain fragmented, often buried in framework-specific docs.

Wedge to win

Developers building complex agentic systems who are frustrated by vendor lock-in or lack of clear, actionable, and neutral design patterns will appreciate well-documented, portable examples.

Reputation value (worth doing free for proof) — 85/100

A merged PR with a high-quality, provider-neutral agent skill example to a 1.5k-star repo builds significant credibility as an AI agent developer and contributor.

Likely monetization

unclear

Incumbents to beat

LangChainLlamaIndexCrewAIAutoGen

You are an expert AI agent developer. I need you to create a new agent skill example following the 'agents-best-practices' GitHub repository's guidelines (DenisSergeevitch/agents-best-practices). This skill should be a 'unit test generator' that takes a user-provided JavaScript function and generates appropriate unit tests for it. The core agent logic should be provider-neutral and demonstrate how to integrate with a generic LLM interface, specifically using AI SDK v6 with Gemini. The front-end should be a simple Next.js 16 App Router application (React 19, Tailwind v4) where users can input their function and view the generated tests. Host this on Vercel with Neon Postgres (though the database won't be heavily used for this MVP, set up for future scalability). 

Your task is to:
1.  Set up a new Next.js project with the specified stack.
2.  Implement the AI SDK v6 integration for Gemini to act as the test-generating agent.
3.  Design a `provider-neutral` skill interface for the unit test generation within the agent logic.
4.  Create a user interface with input fields for the JavaScript function and a display area for the generated tests.
5.  Include basic error handling and loading states.
6.  Provide clear `README.md` documentation for setting up and running the example, emphasizing how it adheres to the 'agents-best-practices' principles.

Focus on the MVP: just generating tests for one function at a time. Do not implement complex state management or user authentication for this first iteration. Assume the user provides valid JavaScript. Verify by running `npm run dev` and ensuring the UI appears and test generation works for a simple function like `function add(a,b){return a+b;}`.