Almost every developer (and not only) is now using coding agents such as Claude Code, Codex, Cursor, or GitHub Copilot, but most don’t trust the quality of the generated code.
One reason is that even the latest models lack the most recent product knowledge. They may have been trained on data missing newer io.Intelligence capabilities like io.Assist or AI Server. And they may also have outdated information about our MCP packages and suggest an approach that is no longer correct.
Agents can search the web to fill those gaps, but often miss the most relevant documentation, take more time, and consume unnecessary tokens. That’s what the io.Intelligence AI Dev Kit aims to improve.
What is the AI Dev Kit?
The V1 kit has two main parts:
- The
io-intelligenceskill gives compatible coding agents io.Intelligence-specific architecture knowledge, task routing, implementation workflows, and diagnostic playbooks. - The template materializer lets the agent retrieve one official application template from a catalog-pinned, immutable Git tag.
The kit isn’t an io.Intelligence runtime package, and it doesn’t replace the documentation or the developer. Think of it as a better specialist guide for your coding agent.
The skill assists the agent with the correct context about MCP, Working Context, AI Web, io.Assist, AI Mastra Bridge, and the Agent Protocol. It also tells it where to find the exact documentation relevant to the current task and which file to inspect before recommending or changing an implementation. Here are a few examples of what the agent is good at:
- Explain the io.Intelligence architecture and select a suitable integration path.
- Add io.Intelligence capabilities to an existing project while preserving its framework and conventions.
- Integrate MCP tools, resources, prompts, servers, and clients.
- Build a ready-made assistant with packaged io.Assist or a custom assistant frontend with AI Web.
- Diagnose issues involving application discovery, tools, threads, streaming, context, or protocol events.
Application templates
The kit also ships five official starter applications:
| Template | Fit |
|---|---|
io-assist-react-packaged io-assist-angular-packaged |
The ready-made assistant and its supported configuration and extension points are enough. The assistant remains a normal package dependency. Simplest to maintain. |
io-assist-react-editable io-assist-angular-editable |
io.Assist is the starting point, but you have the freedom to customize it. The generated source becomes a customer-owned, one-way fork. |
ai-web-vanilla |
You want a framework-free starting point for a custom UI built directly on AI Web. Full freedom. |
The installed skill contains a small catalog and a materializer. When you request a template, the materializer downloads only the selected starter from a pinned Git tag into an empty target directory and records its origin in .io-intelligence-template.json. Every V1 template also ships with an empty AGENT_SERVER_URL, which must be set to a valid Agent Protocol backend before running the assistant.
The skill gives coding agents better context and a better route through io.Intelligence, but it does not make them infallible. Make sure you check the io.Intelligence documentation, review the agent’s recommendations, and understand their impact before applying or shipping them. The AI won’t be accountable for what reaches production - you will.
Quick Start
Install the skill using npx as shown here: Getting Started | io.Intelligence Framework
npx skills add InteropIO/iointel-dev-kit --skill io-intelligence
Check for available updates with:
npx skills check
Update the skill with:
npx skills update io-intelligence
Fork the repository in GitHub if you want to experiment with your own version.
Open an issue if you find missing guidance, a workflow that could be clearer, or a case where the agent consistently takes the wrong route, etc. For useful reports, please include the coding agent you used, prompt, relevant project context, what you expected and what the agent actually did.
Try it and tell us what you think
This is V1, and we’d really like to hear your feedback.
Try the skill with your preferred coding agent and tell us what worked, where it struggled, and which workflows, templates or diagnostics you would like us to add next. Share your experience here in the community as a comment or open an issue in the repository.