Getting Started with Claude Skills
A 30-minute primer on Claude Skills for AU SME operators. What they are, why they matter, how to install from GitHub repos, how to write your first skill, and how to share across your team. Cross-references the anthropics/skills repo and community catalogs. Starting point for anyone evaluating whether skills are worth the time.
What a Claude Skill actually is, the 3 ways to use them (Claude.ai / Claude Code / API), how to install, how to write your first, the 5 rules that make skills work, and common beginner mistakes.
When to use
Triggers:
- "What are Claude Skills?" / "How do Claude Skills work?"
- "How do I install a skill?" / "Where do I find skills?"
- "Can I write my own skill?" / "How do I share a skill with my team?"
- "Claude Agent Skills" / "SKILL.md"
- Any AU SME operator trying to understand whether skills are worth investing time in
Don't fire for:
- ChatGPT Custom GPTs (use
getting-started-with-chatgpt-custom-gpts) - Claude Projects (different feature — scoped conversations with files, not reusable procedural knowledge)
- Programming the Claude Agent SDK (developer-level, out of scope)
What a Claude Skill actually is
A skill is a folder Claude reads when it thinks a task matches. Each skill contains:
- A
SKILL.mdfile with instructions in plain English - Optional reference files (
references/*.md) for longer docs - Optional scripts (
scripts/*.py) for executable helpers
When you give Claude a task, it scans available skills. If a skill's description matches, Claude opens it, reads the instructions, and follows them. You don't have to ask for the skill by name — the triggers are built into the skill's description.
Think of skills as procedural knowledge you share with Claude that sticks across conversations. Unlike a one-off prompt you type, a skill is always available and always applied when relevant.
Why skills matter for an AU SME
You're probably already prompting Claude the same way every week. "Write a professional reply to this customer email." "Turn these meeting notes into actions." "Review this code."
Each time you re-type the prompt, you lose:
- Consistency — today's reply isn't phrased like last week's
- Memory — tricks you learned last week don't persist
- Team alignment — your colleague prompts differently, gets different output
Skills fix all three. One skill = one well-tested workflow, applied the same way by everyone who uses it.
The payoff compounds. A practice with 5 skills for 5 repeat tasks saves each team member maybe an hour a week. Multiply across a team of 10 and that's 10 hours reclaimed without any head-count or software change.
The 3 ways to use skills (with Claude)
1. Claude.ai (Web / Desktop / Mobile)
Skills work on the Free, Pro, Max, Team and Enterprise plans, and they need code execution turned on. You manage them in Customize > Skills, where you can switch skills on or off and upload your own.
- Free, Pro and Max: turn on "Code execution and file creation" in Settings > Capabilities, then open Customize > Skills.
- Team: skills are on by default for the organisation.
- Enterprise: an owner turns on Skills in Organization settings > Plugins & skills.
Owners of Team and Enterprise organisations can also provision skills so they appear for everyone.
2. Claude Code (terminal / IDE)
For technical users. Skills live at:
~/.claude/skills/<skill-name>/— user-level (applies to all Claude Code sessions)<project>/.claude/skills/<skill-name>/— project-level (applies only when in that project)
Install a skill by cloning the folder to one of those paths. Claude Code auto-detects.
3. Claude API (for your own apps)
If you're building a tool that uses the Claude API, load skills as part of your system prompt or via the skill-loading API. This is developer-level work.
How to install a skill (no coding required)
Most AU SMEs use option 1 (Claude.ai) or have a technical team member set up option 2.
From a GitHub repo
Community skill repos (like anthropics/skills, ComposioHQ/awesome-claude-skills, travisvn/awesome-claude-skills) publish skills as folders.
For Claude.ai:
- Download the repo (use GitHub's "Download ZIP" button)
- Extract the specific
<skill-name>/folder - Zip that folder on its own, so the ZIP holds the
<skill-name>/folder withSKILL.mdinside - In Claude.ai, go to Customize > Skills, click "+", then "Upload a skill"
- Upload the ZIP. The skill appears in your list and you can switch it on or off
For Claude Code:
mkdir -p ~/.claude/skills cd ~/.claude/skills git clone https://github.com/anthropics/skills.git anthropics-skills # Or, for a specific skill only: curl -L https://github.com/owner/repo/archive/main.tar.gz | tar xz -C /tmp mv /tmp/repo-main/skills/<skill-name> ~/.claude/skills/
From the Tech Horizon Academy library
THA publishes its own library of skills (the one you're reading now is part of it). For members:
- Browse
/skillson academy.techhorizonlabs.com - Each skill page shows the full SKILL.md content
- Download the skill as a ZIP and upload it in Customize > Skills, or unzip it into your Claude Code skills folder
- Some skills are gated by tier (Free / Base / Pro / Partner)
How to write your first skill (30 minutes)
Start small. Pick one thing you prompt Claude to do every week.
The minimum skill
--- name: my-first-skill description: One paragraph describing when to use this skill. Include trigger phrases like "Use when the user says X, Y, or Z". The clearer the triggers, the more reliably Claude picks the skill. --- # My First Skill ## When to use - Trigger 1 - Trigger 2 ## Steps 1. Do this 2. Then this 3. Then this ## Output format What Claude should return.
The 5 rules that make skills work
- Clear triggers in the description. Claude won't pick up a skill it can't match. Include the phrases a user would actually say.
- Specific, not generic. "Help with marketing" is too broad; "Draft a 3-email cold outreach sequence in AU Spam Act-compliant format" is useful.
- Steps, not principles. A skill that lists principles without steps is an article, not a skill. If Claude can't execute it mechanically, it won't reliably perform it.
- One skill, one job. If your skill covers 5 different workflows, split it into 5.
- Test with real inputs. Write a skill, try it on 3 real examples, refine until it works. Don't ship skills you haven't tested.
Common beginner mistakes
- Writing an essay where a checklist would do. Prompts work better than prose in skills.
- Too many examples. 2–3 concrete examples is enough. 10 is overload.
- No trigger phrases in the description. Claude can't find a skill that doesn't say what it's for.
- Treating it as Python. Skills are instructions in English, not code. You don't need to tell Claude how to tokenise text.
Sharing skills across your team
Once you have a skill working, share it:
For Claude.ai Team and Enterprise
- Share the skill with colleagues from Customize > Skills (the "..." menu, then "Share"). People you share with can use it but can't edit it.
- Publish it to your organisation's library with "Publish to org", or ask an owner to provision it for everyone.
For Claude Code users
- Publish the skill folder to a shared Git repo (private is fine)
- Teammates clone it into their
~/.claude/skills/path - Updates propagate via
git pull
For the Tech Horizon Academy Partner Tier
- Partner-tier members can submit skills for inclusion in the Academy library
- Contact [support@techhorizonlabs.com] with the SKILL.md
How skills compose
Good skill libraries reference each other. A pricing skill mentions a packaging skill mentions a retention skill. Claude follows the chain as needed.
In your own SKILL.md, add a ## Related skills section listing sibling skills by slug. It makes your library a graph, not a pile.
Skills vs Projects vs prompts — when to use each
| Use case | Tool | |---|---| | One-off task, won't repeat | Plain prompt | | Recurring task with the same structure | Skill | | Ongoing body of work with files / conversation memory | Project | | Multi-step workflow with tool use and side effects | Agent SDK (developer) |
Skills and Projects are complementary. A Project loads the right skills for its scope; the same skill can serve many Projects.
Measurement
Use the ai-roi-measurement-for-sme skill to track whether your skills are paying off. Measure the same task before and after. If a skill doesn't measurably save time or raise quality, kill it.
What to do next (by role)
- Operators: Install 2–3 skills from the Academy library that fit your daily work. Use them for a month. Measure with
ai-roi-measurement-for-sme. Then consider writing your first skill. - Team leads: Pick one recurring team task. Write a skill that captures your best-version approach. Roll it out to the team. Measure the consistency improvement.
- Technical operators: Install Claude Code. Clone the anthropics/skills repo. Pick one skill that looks valuable. Run it against a real task.
What this skill does NOT do
- Write your skills for you. Use [superpowers:writing-skills] or similar tooling for that.
- Replace deep user manuals. Skills are procedural knowledge, not documentation. If your business needs a proper user manual, write a user manual.
- Work in ChatGPT. Different platform, different feature (Custom GPTs). See the sibling skill
getting-started-with-chatgpt-custom-gpts.
Tier access
Free. Foundational content, essential for any THA member starting with skills.
Related skills
getting-started-with-chatgpt-custom-gpts— the ChatGPT equivalent + how they differclaude-4-7-extended-thinking-for-sme— pair with extended thinking for higher-stakes skill-driven workai-roi-measurement-for-sme— measure whether your skills are actually helping
References
- →What's a Claude Skill?
- →How do I install skills from anthropics/skills?
- →Can I write my own skill?
- →How do I share a skill with my team?
Source
official
Author
Tech Horizon Academy
Version
1.0
Complexity
Compatible With
Prerequisites
- Access to Claude.ai (Free/Pro/Team/Max) OR Claude Code OR Claude API
- 30 minutes
Best For
Tags
