Free Training Course

Get Instant Access

Enter your name and email to unlock all 5 modules, the practice builder, and your completion certificate.

All 5 course modules — free, no credit card
Practice prompt builder + completion certificate
Future free tools and AI governance updates from E-Choice Solutions

Please enter a valid name and email address.

No spam. Unsubscribe anytime. Your information is never sold or shared.

You're in!

Welcome to Build a Claude Agent from Scratch. Your seat is confirmed.

Claude Agent Training
Free Training Course

Build a Claude Agent
from Scratch

Five focused modules. Hands-on practice. A completion certificate. No prior AI experience required.

5 Modules Knowledge Checks Practice Builder Completion Certificate
Module 1 of 5

What Is a Claude Agent?

Understand what an agent actually is, how it differs from a regular Claude conversation, and why it matters for your work.

You Provide
Instructions, rules,
voice, tasks
Claude Builds
A system prompt
with your specs locked in
You Get
A working agent
that runs every time
1
The core idea
Concept

A Claude agent is a specialized version of Claude that has a defined role, a voice, rules it follows, and specific tasks it performs. Think of it as hiring a focused specialist instead of asking a generalist the same question every day.

Without an agent: you explain your brand, your tone, your audience, your rules — every single time you open a chat. With an agent: Claude already knows all of that. You just pick a task and generate.

Key point: The agent lives inside a Claude conversation. Claude embeds your instructions into a "system prompt" — an invisible set of rules that runs automatically every time you use it.
2
Agent vs. regular Claude chat
Comparison
Regular Claude chat

You re-explain your context every session. Claude has no memory of your voice, brand, or rules. Results vary widely.

Claude agent

Your rules, voice, and tasks are locked in once. Every generation follows the same logic. Results are consistent and on-brand.

Important: An agent lives in the specific conversation where it was built. If you open a new Claude chat, the agent will not be there. Always return to the original conversation — bookmark it in your browser.
3
What can an agent do?
Use cases
  • Generate consistent marketing content in your brand voice
  • Write outreach emails that follow your exact structure every time
  • Produce proposals, summaries, or reports in your preferred format
  • Handle multiple content types from a single agent interface
  • Enforce rules automatically — banned words, tone, length, structure

An agent is not magic. It performs exactly as well as the instructions you give it. The better your input, the more useful the agent.

Knowledge Check

Two quick questions to confirm you have the core idea.

1. What makes a Claude agent different from a regular Claude conversation?
Claude agents use a completely different platform
An agent locks in your rules, voice, and tasks so you don't repeat them each session
Agents produce image or video output automatically
2. Where does a Claude agent live?
In the cloud, accessible from any Claude conversation
In a separate Claude account you have to create
Inside the specific Claude conversation where it was built
Module 1 complete
Module 2 of 5

The 5 Building Blocks

Every effective Claude agent is built from the same five inputs. Understand what each one does — and what happens when you skip one.

1
Purpose
What does this agent do?

The purpose is the agent's one job. It defines what the agent focuses on and — just as importantly — what it does not do. A focused agent outperforms a general-purpose one every time.

Too vague

"Help me with marketing."

Specific & effective

"Write LinkedIn posts, Gumroad sales copy, and outreach emails in my brand voice for E-Choice Solutions."

Ask yourself: What is the one job this agent does? What should it never do?
2
Brand & Voice
How does it sound?

This is where most people underinvest — and where consistency is won or lost. Your voice includes your tone, your personality, and the words you always use or never use.

  • Your name, business name, and positioning statement
  • Tone descriptors: professional, direct, warm, conversational, authoritative
  • Words or phrases you always use
  • Words or phrases you never use (banned words list)
  • Your target reader — who sees the output?
3
Audience
Who reads the output?

The same content written for a Fortune 500 procurement team reads completely differently from content written for a solopreneur just starting out. Define your audience precisely so the agent calibrates its language, depth, and tone automatically.

Too broad

"Business professionals."

Precise

"Federal agency contracting officers evaluating AI governance vendors."

4
Tasks
What does it produce?

List every content type or format the agent needs to handle. Each task becomes a selectable option in the agent interface Claude builds for you.

  • LinkedIn thought leadership post
  • Cold outreach email (recruiter / hiring manager)
  • eBook sales copy (Gumroad)
  • Instagram caption
  • Executive summary or capability brief
You can always add tasks later by telling Claude in the same conversation: "Add Instagram captions as a content type."
5
Rules & Constraints
What must it always or never do?

Rules prevent the agent from drifting. Without explicit rules, Claude defaults to generic behavior. With rules, the agent behaves consistently on every use.

Always rules

Use active voice. End every post with a direct call to action. Keep LinkedIn posts under 250 words.

Never rules

Never use the words "leverage," "elevate," or "game-changer." Never write in passive voice. Never open with "I."

Rules are the most powerful part of your agent. Invest real time here. A good rules list is the difference between an agent you use once and one you use every day.

Knowledge Check

Test the building blocks before moving on.

1. Which building block prevents the agent from using language you dislike?
Purpose
Audience
Rules & Constraints
2. Why is being specific about your audience important?
It tells the agent how to calibrate language, tone, and depth for that reader
It makes the agent run faster and cost less
It unlocks more content types in the agent
Module 2 complete
Module 3 of 5

Write Your Prompt

How to take your five building blocks and turn them into a prompt that Claude uses to build your agent — including how to refine it after the first build.

1
The starter prompt template
Use this to begin

Open Claude at claude.ai and send a message in this structure. The more detail you include from your five building blocks, the stronger the agent Claude builds.

Starter Prompt — Copy and Customize
I want to build a Claude agent that [describe the main purpose]. The agent should always [key rule 1] and never [key rule 2]. Here is information about my brand and voice: [paste your brand notes, tone descriptors, banned words list] My target audience is: [describe specifically] The agent needs to handle these content types: [list each task/content type] Please build this as a working interactive agent I can use right here in this chat.
Tip: You do not need perfect notes. Paste rough bullet points, a link to your website, or even a paragraph describing your work. Claude will organize it and ask follow-up questions if it needs more.
2
Ask Claude to guide you
If you're not sure what to include

If you are not ready to fill in every section, try this instead. Let Claude ask the questions — then answer them one by one.

Discovery Prompt
I want to build a marketing content agent. Ask me all the questions you need before building it so the agent matches my brand accurately.

Claude will ask about your business, voice, audience, banned words, and content types. Answer each question in your own words. Once Claude has everything, it will build the agent.

3
How to refine after the first build
Iteration is expected

Your first agent build is a starting point — not a final product. Test it immediately by generating a sample. Then use plain-language corrections in the same chat.

Refinement Examples — Say These in the Chat
"The tone is too formal. Make it sound more direct and confident." "Add Instagram captions as a content type." "It keeps using the word 'leverage' — add that to the banned words list." "The LinkedIn posts are too long. Cap them at 200 words." "The output reads too generic. Here is an example of my best post — match this style exactly."
You can refine the agent as many times as needed in the same conversation. Claude updates it on the spot. Only bookmark it when you are satisfied with the results.
4
Optional: Make it permanent
Advanced — Custom Skill upload

Once your agent performs exactly as you want, you can install it as a Custom Skill — making it available in every Claude conversation automatically, not just the one where it was built.

  • 1
    Tell Claude: "Build me the SKILL.md file and zip it so I can upload it as a Custom Skill."
  • 2
    Download the zip file Claude creates
  • 3
    In Claude.ai, go to Settings → Features
  • 4
    Upload the zip file — your skill is now active across all conversations
Only do this after thorough testing. Do not upload a skill that still needs refinement — you would have to rebuild it from scratch.

Knowledge Check

Three questions on the prompting process.

1. You want Claude to guide the process with questions rather than writing the full prompt yourself. Which approach should you use?
Start a blank chat and wait for Claude to ask questions
Use the discovery prompt: "I want to build an agent. Ask me what you need before building it."
Fill in the starter template and submit it immediately
2. After building your agent, you notice it writes in a tone that is too formal. What should you do?
Start a new Claude chat and rebuild the agent from scratch
In the same chat, tell Claude: "The tone is too formal. Make it more direct."
Accept the formal tone — that's how the agent was designed
3. When should you upload your agent as a Custom Skill?
Immediately after the first build
Every agent must be uploaded as a Custom Skill to function
After testing thoroughly and confirming the agent performs correctly
Module 3 complete
Module 4 of 5

Practice: Build Your Prompt

Apply the five building blocks to your own situation. Fill in each section below — your completed agent prompt will generate automatically.

How this works: Fill in all five sections. When you click "Generate My Prompt," the app assembles your inputs into a ready-to-use Claude agent prompt. Copy it and paste it into claude.ai to build your agent.
1
Purpose
What is this agent's one job?
Describe the main purpose in one or two sentences
Be specific. What does the agent do and — if relevant — what does it not do?
2
Brand & Voice
How does your content sound?
Tone descriptors (select all that apply)
Professional Direct Authoritative Conversational Warm Bold Empathetic Educational
Your name and business name
Banned words or phrases (things the agent must never write)
Brand colors
Used when the agent references visual content, Canva directions, or social media graphics. Hex codes are standard — your designer or brand guide will have these.
3
Audience
Who reads the output?
Describe your target reader precisely
4
Tasks
What content types does the agent produce?
Quick-add common content types
LinkedIn post Cold outreach email eBook sales copy Instagram caption Executive summary Proposal paragraph Twitter/X post Newsletter section
Add any others not listed above
5
Rules & Constraints
What must the agent always or never do?
Your "always" rules
Your "never" rules
Your Agent Prompt
After copying this prompt, go to claude.ai, start a new conversation, paste the prompt, and send it. Claude will build your agent as an interactive tool right inside the chat.
Module 5 of 5

Course Complete

You now know how to build a Claude agent — from concept to a working prompt. Here is your completion certificate.

Certificate
of Completion
This certificate is proudly presented to
Course Participant
For successfully completing Build a Claude Agent from Scratch — a training course covering agent architecture, prompt engineering, brand voice design, and iterative refinement. Presented by E-Choice Solutions, LLC.

Type your name above — it updates on the certificate instantly.

Opens your browser's print dialog — select "Save as PDF." Make sure Landscape orientation is selected.

What's Next

Put It to Work

🚀

Build your first real agent

Take the prompt you built in Module 4 to claude.ai and let Claude assemble your agent. Test it with five different tasks before considering it ready.

🔁

Refine until it sounds like you

Run three to five generation tests. Pay attention to tone, banned word slips, and format consistency. Correct each issue in the same chat.

📌

Bookmark the conversation

Your agent lives in the Claude chat where it was built. Bookmark it in your browser so you can return to it anytime without searching.

📚

Explore more from E-Choice

Visit echoicesolutions.com for AI governance frameworks, consulting services, and the full eBook library at balthazardeetta.gumroad.com.

Course complete 🎉