Mastering Agentic AI DevOps with Claude Code: My First Full Pipeline Deployment

I recently completed the Ultimate Agentic AI DevOps Assignment using Claude Code, and I’m excited to share what I built, what I learned, and how agentic AI is changing the way we think about DevOps automation.
This post walks through each task from setting up Skills to deploying a live site on AWS and includes lessons, screenshots, and a LinkedIn post to prove it all works!

*Task 1 — Set Up the Skill Files Correctly
What I did:
Created the .claude/skills/ folder structure with four subfolders: scaffold-terraform, tf-plan, tf-apply, and deploy.
Downloaded and renamed the required files:
scaffold-terraform/SKILL.md
scaffold-terraform/template-spec.md
tf-plan/SKILL.md
tf-apply/SKILL.md
deploy/SKILL.md
Verified everything in VS Code to ensure proper placement and naming.
*What I learned:
Claude Skills require precise folder and file naming to function correctly.
The name: field in each SKILL.md must match the filename — otherwise, the agent won’t load.
Issue faced:
Initially placed template-spec.md outside its folder. Claude failed to scaffold.
Fixed it by moving it into scaffold-terraform/ and re-running the skill.

****Task 2 — Walk Through and Explain the Four Skills
/scaffold-terraform
What it does: Generates Terraform files based on a template spec.
Tools used: Claude Code + Write access.
Why Write access: It needs to create actual .tf files in the project.
/tf-plan
What it does: Runs terraform plan to preview infrastructure changes.
Tools used: Terminal access only.
Why Read-only: It doesn’t modify anything — just analyzes.
/tf-apply

****What it does: Applies the Terraform plan to provision resources.
Tools used: Terminal + Write access.
Why Write access: It creates infrastructure on AWS.
/deploy

****What it does: Syncs site files to S3 and triggers CloudFront invalidation.

****Tools used: AWS CLI + Write access.
Why Write access: It modifies cloud resources and pushes files.
Key takeaway:
“Needs conversation context? Use a Skill. Self-contained job? Use a Subagent.”
DevOps example (Skill): /tf-plan needs context from previous steps.
DevOps example (Subagent): security-auditor runs independently on Terraform files.

*Task 3 — Run the Full Pipeline: Scaffold → Plan → Apply → Deploy
What I did:
Ran /scaffold-terraform to generate Terraform files.
Manually ran terraform init in the terminal.
Executed /tf-plan to preview changes.
Ran /tf-apply to provision AWS resources.
Used /deploy to sync site files to S3 and invalidate CloudFront cache.
*What I learned:
Claude Code can orchestrate a full deployment pipeline with minimal manual steps.
Each skill builds on the previous one — like a relay race.
*Issue faced:
CloudFront invalidation failed due to missing permissions.
Solved by updating IAM role with CloudFront: CreateInvalidation permission.
Success:
My site is live at: https://d123abcxyz.cloudfront.net
*Task 4 — LinkedIn Post (MANDATORY)

****What I built:
A live static site hosted on AWS using Terraform and Claude Code.

****Skills used:
/scaffold-terraform, /tf-plan, /tf-apply, /deploy

****What I learned:
Agentic AI can automate DevOps pipelines with precision and context-awareness.
Screenshots included:
Claude Code skill execution
Live site in browser
Terraform files in VS Code

🔗 LinkedIn Post URL: https://www.linkedin.com/posts/ebelechukwu-okafor_agenticai-mcp-devops-activity-7439480896090537984-PpIl?utm_source=social_share_send&utm_medium=member_desktop_web&rcm=ACoAABglo4IBjif-4VKcp2fiPwev8ImRV7LTaN0

✅ Final Thoughts
This assignment taught me how to:
Structure and run Claude Skills correctly
Understand tool permissions and context boundaries
Deploy a full infrastructure pipeline using AI agents
Agentic DevOps is real — and it’s powerful.

I’m excited to keep building with Claude Code and explore more advanced workflows.