Markdown files that explain an AI Agents, how to use a skill. For example, you can explain how to run GoHugo, what the features are, and explain everything well in a Markdown file. If you bundle these, you have a repo containing multiple “skills”.
They are generally stored at ~/.claude/skills/gohugo/skill.md. But essentially, they are just explainer Markdown files.
Originally Claude Skills
Originated from Claude Code, which extents its knowledge with skills. Claude explains it as “Create, manage, and share skills to extend Claude’s capabilities in Claude Code. Includes custom commands and bundled skills”.
# First Impressions
Written when heard for the first time in 2025-10-17.
If you work with Markdown or an open text format (like code or a local presentation), this is what I use, Claude Code, all along. I haven’t checked too much, but I find it counterproductive to wrap it in a zip and then unwrap it again, which adds multiple serialization/deserialization steps, no?
But yeah, I get it, for non-local work, maybe. Getting the Claude Code experience to everyone.
i’m letting the smarter folks figure all this out and just picking the tools i like every now and then. i like just using claude code with vscode and still doing some things manually HN
Yeah, avoiding all the serialization and deserialization, as I’m already working in Markdown and open text for almost all my stuff. The Claude Skill only makes sense for people who don’t have their data in multiple proprietary formats; then it makes sense to package them into another format. But this can get messy pretty quickly!
# A Little Later
Today (2026-03-17), they are just open GitHub repos with how-to guides, see List of Skills below—essentially docs for a tool, nothing more, nothing less.
# Just Markdown Files?
Love this statement by Kepano:
if you started writing evergreen notes in 2022, those Markdown files are now software libraries Tweet
What came to mind for me: Before, we called them Markdown files. Today, they’re called Skills.
# Power Of Note Taking
If you add too many, e.g., import a ton, they become ineffective because the token size and context are just too large. Additionally, you don’t know what’s in them, meaning they may contain vulnerabilities or incorrect statements, as most are created with AI.
That’s why it’s better to have only a few you know and trust, and the rest are just notes in your note-taking app. What Obsidian users are doing all along, each note is essentially a skill, if you capture what you learn. The power of note taking never disappoints me.
And to me, that’s the only sustainable version of cumulating skills over time, which is with manual Smart Note Taking.
# Don’t Generate Skills
In the same way as I avoid AI-generated notes or copy-pasting from a website as much as possible in my Second Brain due to not being my own thinking and my own notes anymore, the same goes for skills. If you just copy random notes or collect as much as possible, they lose their power.
Additionally, the more notes you have, the more specific your learning is, the more important stuff just gets lost in the context you give the model.
The best is to create your own skills, basically what I do with my Obsidian vault, all of my Markdown notes are skills I can hand over whenever I need or want to an agent. It’s basically note-taking and writing down things you learned, that later can be taught or shown to an agent to upload that knowledge (are we in the Matrix now? 😀).
Again, don’t produce skills with LLMs, see outcome in Agent Skills Are Spreading Hallucinated npx Commands.
# Maintenance of Skills?
Some skills for the tools you use often are helpful, though skills you can verify or trust. E.g., I use skills from Obsidian or Dagster, but again, this could just be a link to the docs, because skills need to be maintained too. If the docs get updated but the skills do not, I’d rather point my Agents to the docs.
# Lists of Skills and Prompts
- List on Github: anthropics/skills: Public repository for Skills: Skills are folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks. Skills teach Claude how to complete specific tasks in a repeatable way, whether that’s creating documents in line with your company’s brand guidelines, analyzing data using your organization’s workflows, or automating personal tasks.
- Superpowers: Claude Code superpowers: core skills library
# Find Skills
# Danger of Skills: Vulnerabilities and More
Be Careful w/ Skills, they can contain hidden comments that are not seen on GitHub, but for Agents. Don’t just install or use random skills you haven’t checked or don’t trust.
E.g., a hidden comment that is commented out below: 
(
source)
# Further Reads
Read on at Vibe Coding or Will AI Replace Human Thinking.
Also see:
- Model Context Protocol (MVC)
Origin:
@christiannolan.bsky.social on Bluesky, Anthropic
References: Markdown, Claude Code, Claude