This course contains the use of artificial intelligence.

Everyone is teaching you to build one agent. Nobody is teaching you to run a hundred capabilities safely.

The market is full of agentic-AI courses that walk you through wiring up a single agent with LangGraph, CrewAI, or MCP — and then stop. But the moment your organization has 40 teams shipping their own agent capabilities, the hard problem isn't building; it's governing and securing capability sprawl at scale. This is the only course built for that reality.

Agent Skills are modular, reusable capability packages — a SKILL Markdown file file, frontmatter, and bundled scripts — that an agent loads just-in-time through progressive disclosure. They're powerful precisely because anyone can write one and drop it in. That's also exactly why they're dangerous: an untrusted skill is untrusted code and untrusted prompts running inside your agent. This course teaches you to build them well, catalog them, put governance gates in front of them, and defend the supply chain that delivers them. What you'll learn

What an Agent Skill actually is — and how progressive disclosure separates knowledge from execution.
How to author a well-bounded, single-responsibility, correctly-triggering skill that never bloats your context window.
Semantic versioning, deprecation, changelogs, and eval harnesses so a skill can evolve without breaking every caller.
How to stand up a skill registry with discovery, tagging, ownership, and catalog hygiene.
A real governance model: RACI, vetting pipelines, least-privilege permissions, and audit trails.
Supply-chain security — the lethal trifecta, the OWASP LLM Top 10, and how to spot and block a malicious skill.

Why this course is different

Framework-centric courses optimize for a demo: one agent, one happy path. Enterprises fail somewhere else entirely — in the sprawl, the provenance, the review gates, and the incident response. We follow Northwind Cloud, a 1,500-engineer B2B SaaS company, from prompt sprawl to a fully governed internal Skills platform, so every concept lands on a concrete organizational problem instead of a toy. 100% hands-on, 100% free to run

Seven hands-on labs run entirely in Docker containers with a local Ollama + Qwen 2.5 model. No paid API keys, no cloud accounts, nothing that can start billing you. One docker compose up brings up the whole environment, and every lab also runs in a deterministic --no-llm mode so hardware is never a blocker. You'll build your first skill, version it v1→v2, publish it to a registry, wire a vetting gate, detect a malicious skill and harden deployment, and finish with a two-part capstone.

Who it's for

Enterprise software engineers, platform and AppSec engineers, and technical leads who are past "hello world" and now have to make agent capabilities safe, discoverable, and governable for many teams. If you've ever watched your org copy-paste the same agent prompt into ten repos with no catalog and no review — this course is for you.

By the end, you'll be able to design, version, catalog, govern, and secure an enterprise Agent Skills platform end-to-end — the difference between running one agent and safely running a hundred capabilities.