$ cat building-my-first-claude-skill.md
Building My First Claude Skill
Claude launched Skills this week, and I decided to experiment and ship my first Skill. I created an anti-ai-writing skill, a pattern-based guide that catches formulaic language before it lands on the page. I've been gathering these rules from both my own experience and various sources where people share tips about AI and writing. I have these as instructions in projects or prompts ready to copy and paste, or even mid-conversation with Claude, I realize I need to tell it about this, and have to stop my flow to give it instructions. The point of Skills is that you have these ready to go across all conversations and don't need to do the heavy lifting over and over again.
I wanted to start small and specific. Skills are set up differently depending on where you use them. I've added this Skill to Claude desktop and Claude Code. I haven't touched the API yet. My goal was to understand Skills. Writing is something I do daily. Having a reusable skill that works across all my projects without project-specific instructions made sense as a starting point.
What I Learned
Anthropic's Skill best practices gave me three applicable principles:
Only add context Claude doesn't already have. When creating the Skill, keep asking, "Does Claude really need this explanation?" I think you'll find most explanations get cut. If Claude already knows something, telling it again wastes tokens.
Match freedom to fragility. For anti-AI writing, I needed "low freedom" as there are specific patterns to avoid and concrete alternatives to use.
Test with all models. Skills augment models, so what works with Sonnet might not work with Opus.
To create a Skill, you can follow the instructions and build. You can also go to your settings, click into the capabilities section, and there you will see Skills. You can toggle off and on the pre-loaded Skills created by them. One of these Skills is a Skill template. Definitely an inception moment and also a smart move, Anthropic!
The Skill at Work
I dictated what I wanted this post to be about and asked Claude to create a draft without the Skill. I then took the same dictation and invoked the Skill and had Claude create another draft. There were a lot of AI tells in the first draft versus the second.
I asked Claude to compare the versions; it caught every pattern the skill targets. Here are a couple of examples and explanations:
Before: "Anthropic's skill best practices were genuinely helpful."
After: "Anthropic's skill best practices gave me three useful principles:"
Why: "Genuinely helpful" is promotional phrasing. "Gave me three useful principles" is factual and sets up what follows.
Before: "Last week I shipped my first Claude skill, and I'm pretty excited about how it turned out."
After: "Last week I shipped my first Claude skill. It stops Claude from writing like an LLM."
Why: Cut the emotional qualifier ("pretty excited") and the vague closer ("how it turned out"). The second sentence now does actual work—it tells you what the skill does. More direct, more informative.
Before: "After a few rounds of testing and refinement, I had something useful."
After: "After testing and refinement, I had something useful."
Why: "A few rounds of" adds unnecessary padding. The meaning is the same without it.
What's Next
I'm looking at other repetitive tasks where a Skill could save time. My goal with AI is to build tools that amplify my own work, not replace it. My recommendation when creating your first Skill is to start with something small and specific. Something that makes you say, "I wish I didn't have to explain this every time."
Now I've got one working. Time to find the next one.
$ _