Claude Skills / Guide
What are Claude skills? A plain-language explanation
Claude skills are installable procedures: folders of instructions Claude loads automatically when a task matches. Here is how they work, why they beat pasting prompts, and how the pieces fit together, written for founders rather than developers.
// Last updated July 3, 2026. Written by the two founders of HuntingHydra.
The problem skills solve
Everyone who uses AI seriously ends up with a graveyard of great prompts. A document somewhere with the good version of your outreach instructions. A chat you scroll back through to copy the audit format that worked. The output quality of your AI depends on your discipline about re-pasting your own best work.
Skills fix that. You package the instructions once, install them, and Claude applies them whenever the task calls for it. The prompt stops being something you type and becomes something you own.
The anatomy of a skill
A skill is a folder. The only required file is SKILL.md, which has two parts:
---
name: seo-page-audit
description: Audits a web page for SEO. Use when asked to
review a page's titles, headings, intent match, or rankings.
---
# SEO Page Audit
When given a URL or page copy:
1. Identify the query the page is trying to win.
2. Check the title, H1, and intro against that query.
3. ...The frontmatter (name and description) is the trigger. The markdown body is the playbook. Skills can also include scripts Claude executes, templates it fills in, and reference files it consults. But most useful skills are just well-organized text.
Progressive disclosure, the part that matters
Claude does not hold every installed skill in memory. It holds only the one-line descriptions. When your request matches one, Claude loads that skill fully. This is why you can install thirty skills without slowing anything down, and why the description field is the most important line you will write.
Rule of thumb: if you cannot tell from the description exactly when the skill should fire, neither can Claude.
Skills vs prompts vs MCP vs agents
The terms get mixed together. The distinctions are simple:
- Prompt. A one-off instruction. Lives in a chat, dies with the chat.
- Skill. A packaged, reusable procedure with an activation trigger. Lives in your setup, works in every chat.
- MCP server. A connection to an external tool or data source. Gives Claude capabilities, not judgment.
- Agent. Claude running a longer task with tools. Skills are what make an agent behave like a specialist instead of a generalist.
If you remember one framing: MCP gives Claude hands, skills give it a playbook. Companies that get value out of AI operationalize both.
What this looks like in practice
Take a concrete example from our own work. Our ICP research skill encodes the exact sequence we run for clients: pull the signal sources, score segments against six criteria, interrogate the weakest assumption, output a one-page profile with disqualifiers. Before it was a skill, that process lived in one founder's head. Now it runs identically every time, for us and for the founders who install it.
That is the real shift. Skills turn expertise into an artifact. The person with the best process wins, not the person with the best typing.
Where to go next
- See the full picture and a directory of 31 skills: the Claude skills hub.
- Using Claude Code? Install and activation mechanics: Claude Code skills guide.
- Want to encode your own process: how to build a Claude skill.
Questions, answered
A Claude skill is a folder with a SKILL.md file that teaches Claude a repeatable procedure, which Claude loads automatically whenever your request matches the skill description.
Install expertise instead of writing prompts.
31 Claude skills for positioning, SEO, GEO, GTM, copy, and UX. Built in real client sprints, tested in Claude, ChatGPT, and Gemini. $29/month.