AI coding assistants keep getting more capable, but they still have a blind spot: they don't naturally remember how you like things done.
That's where agent skills come in. Think of them as plugins for your AI's brain.
In this guide I'll walk through how I use skills, how to manage them, and which ones I actually use.
1.) Use skills.sh
If you're experimenting with agent skills, the easiest way to manage them is skills.sh.
It's a registry of reusable AI skills you can install straight into your project, or globally.
I install most skills globally (which goes into .agents/skills).
Each skill contains:
- Instructions for the agent
- Example usage
- Structured prompts
- Reusable workflows
Instead of copy-pasting prompts between projects, you install the skill and tell your agent to use it.
For example:
Use the
/plannerskill to break this task down into steps.
Use the
/commitskill to generate a commit message for these changes.
Agents are pretty good at using skills on their own, but instead of praying they actually use the desired skill, just be explicit.
2.) Know Which Skills You Have
One mistake I see: installing loads of skills and then forgetting they exist.
So it's worth reviewing your installed skills every so often. Otherwise you forget what your agent can do.
Agents also struggle when you have too many skills. In those cases it will often miss a relevant skill when trying to do some work.
A small, well-curated set beats dozens you never use.
3.) Be Explicit
Agents are good, but they're not mind readers.
If you want consistent results, tell the agent which skill to use.
Bad:
Write a commit message for this change.
Better:
Use the
/commitskill to generate a commit message for these changes.
The second way the agent follows the workflow defined in the skill instead of improvising each time.
That predictability is a big part of why skills are worth using.
4.) General-Purpose Skills
A few have become go-tos for me.
Planning Skills
Planning skills get the agent to think through complex tasks while it's creating the plan of action.
Generate Agent Instructions
View Generate Agent Instructions
I created this one to generate structured instructions for AI agents.
Use it when you're:
- Have just added a feature
- Updating your projects current AGENTS.md
- Defining behaviour rules
- Building reusable system prompts
It saves you from manually writing long instruction docs.
Brainstorming
Good for idea generation.
I use it for:
- In conjunction with the Planner skill
- Product ideas
- Feature brainstorming
- Naming
- Marketing angles
It pushes the agent into divergent thinking mode, which usually beats a generic prompt.
Planner
I get a lot of use out of this one in development.
It gets the agent to:
- Understand the problem
- Break it into steps
- Execute tasks one at a time
If you've ever had an AI jump straight into coding the wrong thing, a planning skill fixes that.
5.) Coding Skills
These matter most if you use AI a lot for development.
Commit
Another one I built.
It generates clean commit messages from your changes.
You get:
- A clear commit title
- Short and sweet
- Stops the agent from writing a book
Your Git history stays readable instead of full of freestyle commit messages.
JavaScript Mastery
Focused on higher-quality JavaScript.
It pushes the agent toward:
- Modern patterns
- Performance awareness
- Clearer structure
- Modular code
Use it when you want the AI to behave more like an experienced JS developer.
UI Mockup Fidelity
I created this one for front-end work with Tailwind CSS.
It helps the agent turn design mockups into code without drifting from the design.
Use it for:
- Translating Figma designs
- Recreating UI screenshots
- Matching layout, spacing, and hierarchy
Without something like this, the output often drifts away from what you actually wanted.
Web Design Guidelines
A solid option for front-end in general.
It reinforces good practice around:
- Layout
- Spacing
- Typography
- Accessibility
Like giving your AI design instincts.
6.) Writing Skills
AI can help with writing. Structured writing skills make the output better.
Copywriting
Use it for:
- Landing pages
- Marketing content
- Product descriptions
- Blog intros
It steers the AI toward persuasive, structured copy instead of generic filler.
Humanizer
Improves the tone of AI-generated content and ultimately helps it sound a bit more human.
It strips out:
- Robotic phrasing
- Repetitive wording
- Overly formal tone
If you publish AI-assisted writing, this one's worth using.
Final Thoughts
Agent skills are still early, but they're quickly becoming one of the most useful parts of my AI workflow.
Skills give your assistant repeatable capabilities instead of fragile one-off prompts.
Once you have a small library of them, the AI gets more predictable and less frustrating to work with.
If you use AI heavily for development, they're worth trying.