Akhtar Jalbani

Posted on Jun 6

What Happens When a Teacher Learns Agentic AI — And Brings It Into the Classroom

#agentskills #aws #awsacademy #ai

I teach cloud computing using AWS Academy Learner Lab. Every semester, my students hit the same invisible walls:

They follow an AWS tutorial, ask an AI assistant for help, and get told to "create an IAM role" — which is blocked in Learner Lab
Their credentials expire mid-session and they don't know why their commands suddenly stopped working
Someone leaves an EC2 instance running overnight and burns through half the lab budget
A well-meaning AI suggests Route 53 or a NAT Gateway — neither of which behaves as expected in the lab environment

These aren't AWS problems. They're Learner Lab-specific constraints that no amount of official AWS documentation covers — and that AI coding assistants have absolutely no awareness of.

So I built something to fix that.

What Are Agent Skills?

Agent Skills are a relatively new open standard — SKILL.md files that give AI coding assistants structured, on-demand expertise. Think of them as a cheat sheet that loads automatically when relevant, without you having to re-explain the same context every session.

Microsoft published an Azure Agent Skills repo covering 193 Azure services. It works beautifully with Claude Code, GitHub Copilot, Cursor, and others.

I took that idea and built the AWS equivalent — specifically tuned for AWS Academy Learner Lab.

The Repo: aws-agentic-skill

👉 github.com/akjalbani/aws-agentic-skill

It contains 7 skills covering the core services students encounter in their first AWS courses:

Skill	What it teaches
🔬 `aws-learner-lab`	The most important one — LabRole constraints, credential refresh workflow, budget guardrails, unsupported services
⚡ `aws-lambda`	Serverless functions, triggers, packaging, CloudWatch logs
🪣 `aws-s3`	Buckets, uploads, permissions, static website hosting, boto3 usage
💻 `aws-ec2`	Launching instances, SSH, security groups, stopping to save budget
🔒 `aws-iam`	Why you can only use LabRole, passing roles to services, fixing credential errors
🏗️ `aws-cloudformation`	Infrastructure as Code, deploying and deleting stacks cleanly
🌐 `aws-vpc`	Default VPC usage, security groups, avoiding expensive NAT Gateways

What a Skill Actually Looks Like

Each skill is a SKILL.md file with YAML frontmatter and structured markdown. Here's the pattern:

---
name: aws-learner-lab
description: >
  Use this skill whenever working inside an AWS Academy Learner Lab environment.
  Provides critical constraints, workarounds, and best practices specific to
  Learner Lab sessions. Always load this skill before any other AWS skill when
  the user mentions "Learner Lab", "Academy", "vockey", or session-based credentials.
---

# AWS Academy Learner Lab

## ⚠️ Critical Constraints

### IAM Restrictions
- You CANNOT create IAM users — the lab blocks this
- You CANNOT create new IAM roles — use the pre-existing `LabRole`
- When a service asks "Choose or create a role", always select **LabRole**

### Credentials & Sessions
- Credentials expire when your lab session ends (~4 hours)
- To get credentials: click "AWS Details" → "Show" next to AWS CLI
- Always include aws_session_token — Learner Lab requires it

...and so on

The AI assistant reads the description field to decide when to load the skill, then uses the full content to answer correctly.

How It Works in Practice

Once the skills are installed, you just talk naturally to your AI assistant. No setup prompts, no repeating yourself every session.

I tested it myself — as a student would. I opened Claude Code in a fresh terminal and typed:

Using the aws-learner-lab and aws-s3 skills, create an S3 bucket and upload a test file to it

Here's what happened:

Step 1 — Claude Code starts and reads the installed skills

![Screenshot: Starting Claude Code in the terminal]
Claude Code launches and discovers the aws-learner-lab and aws-s3 skills automatically

Step 2 — Creates the S3 bucket using the correct Learner Lab approach

No IAM role creation attempts. No wrong region. Just the right command.

Step 3 — Uploads the test file

The upload command, correctly formed with region and credentials context

Step 4 — Lists all buckets to confirm

The bucket appears. It worked.

No AccessDenied errors. No suggestions to create IAM roles. No wrong region. The skill did its job — it gave the AI exactly the context it needed to operate correctly inside Learner Lab.

Installation (5 Minutes)

Prerequisites

Node.js (LTS)
Claude Code: npm install -g @anthropic-ai/claude-code
AWS CLI

Install the skills

# Clone the repo
git clone https://github.com/akjalbani/aws-agentic-skill.git

# Copy skills to Claude Code's global skills folder
# macOS/Linux:
cp -r aws-agentic-skill/skills/* ~/.claude/skills/

# Windows (Command Prompt):
xcopy /E /I "aws-agentic-skill\skills\*" "%USERPROFILE%\.claude\skills\"

For other AI assistants, change the destination:

Assistant	Path
Claude Code	`~/.claude/skills/`
GitHub Copilot	`~/.copilot/skills/`
Cursor	`{project}/.cursor/skills/`
Codex CLI	`~/.codex/skills/`

Connect to your Learner Lab

Every session, refresh your credentials:

Start your lab → wait for the green dot
Click AWS Details → Show next to AWS CLI
Copy the three lines into ~/.aws/credentials
Run aws configure set region us-east-1
Verify: aws sts get-caller-identity

You should see LabRole in the ARN — you're connected.

Why This Matters for Educators

The biggest time sink in a cloud lab session isn't the concepts — it's the environment friction. Students spending 20 minutes debugging an AccessDenied error aren't learning AWS. They're learning frustration.

Agent skills shift that. Instead of the AI giving generic advice that breaks in Learner Lab, it gives advice that's specific to the student's exact environment. The constraints are baked in. The guardrails are built in. The right commands appear first.

My plan for upcoming classes is to have students install these skills on day one — before they touch a single AWS service. The AI becomes their lab partner that already knows the rules of the lab.

What's Next

This is v1 — 7 skills, the core services. I'm planning to add:

aws-dynamodb — NoSQL tables, queries, DynamoDB from Lambda
aws-api-gateway — REST APIs connected to Lambda
aws-rds — managed databases with Learner Lab networking constraints
aws-sns-sqs — messaging patterns students encounter in later courses

The repo is open source and contributions are welcome. If you're an AWS Academy instructor and want to add a skill for a service your students use, the pattern is straightforward — follow the existing SKILL.md structure and open a PR.

Resources

Are you using AI tools in your cloud classes? I'd love to hear what's working — drop a comment below.

DEV Community