AI Skills Every Software Engineer Needs

📝 Note[!note]
Video coming soon.

A skill is just a saved prompt. At the simplest it's a file (or a list of files) with instructions in it that you can hand to an agent whenever you need them.

Often you'll package other stuff with it too, like scripts or reference docs, so a skill can end up being a whole folder. The big difference from a subagent is that a skill gets brought in to your CURRENT context, where a subagent has its own context and only returns the result. You can invoke a skill manually, or the agent can pull it in when it sees it needs it.

I'm still not sure how much standardized skills matter, or if it's fine for everyone to just write their own in whatever shape works for them.

Do You Need a Skill?

Think of skills as functions for your workflow. Any time you're repeating yourself, there's a good chance it could be skillified.

Anytime you're doing something specialized is another one. If I wanted to design 3D models, that's out of scope of my expertise, so I would go find a skill for it.

Also any time you're following opinionated requirements, since a skill can tell the agent to follow certain specs or rules so you get consistent results. If something is scriptable, use a script for that part though, and let the skill point at it.

You can create your own skills too. Some people think they have to go find the perfect skill, but you can just start writing stuff down as a skill, or have the agent craft the skill for you.

There are also sites that list top skills, such as The Agent Skills Directory. I recommend skimming any skill before you install it, so you know you trust what it's telling the agent to do.

ℹ️ Info[!info]
This video gives you an introduction to the most important skills, but there is a lot more to learn about AI than what fits on YouTube. Watching random videos is great for specific topics like this one, but I recommend more structured learning to help you actually apply AI concepts.

For that I've partnered with Coursera, which has a wide range of AI courses and tracks. A few examples:

  • Generative AI for Everyone has over 800K students enrolled and is taught by Andrew Ng, who's considered a leader in the AI space. It's a good overview of AI and how to build with it, and a great place to start if you want the foundation.
  • Generative AI with LLMs goes deeper on how models work and gets in to more advanced topics like fine tuning. Good option if you want to go beyond the fundamentals.
  • IBM RAG and Agentic AI is a 10 course track that covers things end to end, including vector databases and AI search.

These are designed to give you practical, job-relevant skills, and you can get certificates to add to your resume or LinkedIn. Coursera is currently offering 40% off Coursera Plus Monthly for your first 3 months.

The Skills

Exact skills are subject to change, this is the overview outline.

Organization

A skill for braindumps and task making, or an issue writer. You dump out whatever's in your head and it turns it in to tasks.

Specialized Skills

Design skills are first, like frontend-design or skills that build style guides and design systems. I'm giving design its own category since it's the biggest visible thing in your apps.

Next up is mobile development.

Then there are skills provided by companies, like Vercel, Hostinger, etc. If you're releasing apps, it's also worth creating a skill pack of your own for them.

Subagent Orchestration

Skills for coordinating subagents. My custom launch skill is the example I use here.

Review

Anything for reviewing or auditing, like prepping for deployment, an app store audit, or security. This also covers verification that actually does something (runs the app, clicks around) rather than just reading the code.

Skill Building

A skill that builds skills. This may already be built in to your harness, but others exist, like Skill Creator.

Reflection

Skills that are built to improve your systems and workflows over time. One example is continuous-learning-agent.

Token Optimization

Skills for context management, token optimization, or local delegation.

Research

For research, there's Deep Research, which builds multi-source web reports.

Complex Workflows

Guides for workflows where a mistake is expensive, like Database Migrations for database safety.

Voice and Style

A skill for writing in a voice, or according to some standard. AI is pretty bad in the writing department, it either sounds like a robot or writes an entire novel without saying anything at all. This applies to content, but also to requirements like PR structures.