Claude for Beginners Workshop
Most people are using ten percent of what Claude can do. Everything from today's workshop, in one reproducible page — every step and every prompt below works standalone.
New to this? Start here
What's a "Skill"?
A Skill is you teaching Claude your process once, so you never have to explain it again. No code — just a name, a description of when it applies, and the exact steps, packaged so Claude can pick it up and follow it exactly, every time.
You don't need a GitHub account
Everything below works by copy-and-paste. You only need a claude.ai account (or Claude Desktop).
This workshop has three phases
Understand the power of Claude. Set up your environment.
Set up your connectors. Set up your skills.
Automate, scale, and expand. Schedule it, and watch it carry into Claude Code.
Step 1 — Connect your tools
If Claude can't reach the things you reach, and doesn't know the process you'd use, it can't do your job with you. Connectors let Claude access your apps and services.
- Click on Settings.
- Click Connectors.
- Find your preferred connector — toggle it on, sign in.
- Do the same for others (e.g. Google Drive).
- Confirm both are connected.
Outcome: Claude can now read and act on your real files, inbox, and calendar — instead of guessing.
Ask questions, once you're connected
Check my last week of supplier emails against my inventory spreadsheet —
what's actually low, and what's still waiting on a shipment?Draft an email to my supplier to renegotiate a new price for the [item].Step 2 — Build your CLAUDE.md
Think of CLAUDE.md as the manifest that tells Claude who you are, what you do, and what you want to achieve.
Interview me to build my CLAUDE.md. Ask me one question at a time: my role,
the one task eating the most of my time, how I'll know it's working, and the
smallest first step I want you to own. When you have all four answers, write
them into a CLAUDE.md file and confirm it back to me. If you need more
research, leave it for the refine steps.Step 3 — Install your first skill
A connector gives Claude your tools. A skill gives Claude your process. Install the skill factory.
Copy-paste (everyone)
- Open the public gist, click Raw, select all, and copy.
- Claude.ai / Claude Desktop / Cowork: Settings → Capabilities → Skills → upload (wrap the text as
SKILL.mdinside abuild-skill.zipif needed). - Claude Code: paste into
~/.claude/skills/build-skill/SKILL.md.
Use build-skill on: <describe your repetitive task in 1-2 sentences>.npx skills add iankiku/agent-skill-factory --skill build-skillI work at <where you work>, as a <what you do>, in the <your industry>
industry.
Use the build-skill pattern from https://gist.github.com/iankiku/0366d5701cf8268ee05c24cd30fa366b
(full repo: https://github.com/iankiku/agent-skill-factory) — either import it as a skill,
or just follow its pattern directly. Find or build me a skill for: <describe your
repetitive task in 1-2 sentences>.Go to https://claude.com/resources/use-cases including subpages, review and reproduce every
permitted use case and associated prompt, organize them with consistent metadata and
attribution in a new GitHub repository, and create a corresponding skill template for
each one. Then build a primary `build-skill` skill that helps users turn a narrow
domain-specific outcome into an executable skill by asking targeted questions,
selecting the best template, identifying required context, files, tools, APIs,
connectors, authentication, permissions, workflow steps, decision points, validation
criteria, failure modes, and fallbacks, then generating and iteratively refining a
first draft. If the user repeatedly says "I don't know," offer options, make clearly
labeled assumptions after three attempts, and proceed with a practical draft. The
generated skill must also decide when to use the primary agent versus
capability-based model or sub-agent delegation, define the context, outputs,
validation, and fallback for each delegated task, parallelize only independent work,
and retain final review and sensitive actions with the primary agent. Include setup
documentation and end-to-end validation examples, respect the source website's
licensing and access restrictions, preserve attribution, protect secrets, and review
the repository before publishing it to my GitHub.Step 4 — Schedule it
Once a skill has run correctly by hand at least once, put it on a schedule so it runs without you.
Scheduled Tasks — set a schedule, Claude runs it unattended.
/loop [interval] [prompt] — runs repeatedly while the session stays open.
Take the skill we just built [name the skill, or invoke it with /skill-name]
and schedule it to run every [enter time]. Confirm the schedule back to me
before you turn it on. I want the output in this format: [describe the
format you want].Got a workshop credit code?
Skip this if you weren't at the live session. Credits must attach to an Organization using the Org ID from console.anthropic.com.
1. Go to console.anthropic.com and sign in (or create an account).
2. Create or select an Organization. Credits attach to the org, so you will
need its Org ID (from the Console, not your Claude.ai profile).
3. Use the redemption link that comes with your credit code.
4. In the Console, create an API key for the session.
5. Check Billing then Credits for your balance. Done.Where the original material lives
agent-skill-factory (full repo)
94 catalogued use cases, 94 skill templates, and the build-skill meta-skill — full source.
build-skill — public gist
Single self-contained file. Open it, click Raw, select all, copy — no GitHub account needed.
The 94-use-case index
Every catalogued use case by category, linking the original Anthropic page, the catalog doc, and its template.
Anthropic's original use-cases page
The source material — reproduced with attribution in the catalog above.