Upgrade your AI work process using Agent Skills for Mobile App Development

So you are already using AI in your development and have tried Agent mode, so what is next? It is time to try Skills and Superpowers, of course. In this blog, I will try to explain what they are, how to use them, and what they can and cannot do in your everyday work process.

What are Agent Skills?

Agent Skills is a thing added to AI Agents to extend and specify Agents for your set of things you are working on. There are sorts of scripts, but not really: a script is strict code that is not very flexible, a Skill is more like a set of instructions for AI.

If you are working with AI every day, you may experience tasks that require you to write the same part of your prompt every time. Skills are here to resolve this kind of issue – now you can add a Skill to a project and only ask AI what you actually want it to resolve, without repeating yourself.

Basically, Skills allow you to add custom routines to the AI process. Previously, we had only instructions in MD files, but those are common project guidelines and not specific enough, when Skills are like a path for AI to think of, and allow AI to do the thing with fewer token burns, more efficiently, and faster. 

What are Superpowers?

So, now understanding the nature of Skills, we can talk about superpowers. Superpowers are an independent library of Skills that allow you to add structure to your AI-based development process.

Before Superpowers, you just type your prompt, wait for AI to do the work (while hoping that the context window will not overflow), and then review the changes that AI made. It is fine for simple tasks, but when you try to do something a little bit complex, you experience one of these things:

• AI didn’t get you right, and the feature was implemented completely wrong
• AI doesn’t consider context and did the feature right, but nuanced 
• You forgot to ask for something in the prompt, and AI didn’t think about that
• The feature was done right to a description, code is functioning right too, but you do not like the architecture of the solution or other things that are not related to the “working state.”
• The context window of AI runs out in the middle of the task and now you are left with corrupted or unfinished code.

Superpowers add steps to development, using a set of custom Skills called The Basic Workflow. It consists of steps like this:

• Brainstorming  – generate a rough understanding of what we need to do, with detailed ideas, questions for a developer, etc. It is the earliest pre-code stage.
• Using-git-worktrees – after brainstorming, results were approved by the developer, create separate workspace was created for feature implementation.
• Writing-plans – this is the stage where brainstormed approved designs will be transformed into a detailed plan of implementation, which considers the project’s context, previously built codebase, AI instructions, architecture, etc.
• Subagent-driven-development or executing-plans – this will develop the feature by the plan, dividing plan steps into “1 step – 1 subagent” workflow. Those Skills are to resolve the issue of the context window of AI running out in the middle of the task.
• Test-driven-development – for implementing your feature with TDD
• Requesting-code-review – this should be activated between tasks, by user demand, or if the plan requires it. 
• Finishing-a-development-branch – activates the dev branch. Take a look at implemented changes, run tests, review code once more, and merge into dev on success and clean up the worktree.

This workflow is customizable, but mandatory in order to use Superpowers. Also, Superpowers are best for TDD, clean architecture, and standard Git flow. 

Why do you need to use Skills and Superpowers anyway? 

Skills are a must if you have routines with AI you want to automate more. Also, if there are tasks that are specific to your project and you need a more context-filled routine with AI, Skills are here to help.

Superpowers are a whole other thing. It changes your approach to an AI agent – by using Superpowers, the Basic Workflow, you will add more structure to how you interact with AI agents in general. It is a very good set of Skills for TDD projects or trying to be TDD. Also, it is a blessing with a large codebase, because Superpowers divides everything into small steps, and AI doesn’t die in the middle of a road or start to go in circles or hallucinate in the process. Also, if some step goes wrong, it’s a lot easier to get back on track with Superpowers than with regular AI development.

But if your process differs much from the Basic Workflow or you do not need to write tests or have a very complex, legacy project – Superpowers may not be for you.

In conclusion, I think that Skills are a must to try for every project that uses AI in development. But for Superpowers, it is not that obvious. I think you can consider a try if you can, but it may not be for your case. Anyway, Skills are an amazing thing to not only make things go more smoothly, but also to reduce token usage and relieve the pains of developers, so I strongly recommend them to all and anyone, who uses AI.

If you need the right technology to help you incorporate AI into your business or for other technology needs, schedule a free assessment.