Skip to content
GH6 · the masterclass

The GH-600 Masterclass.

Eight video lessons for the GitHub Certified: Agentic AI Developer exam, recorded by a team that ships with these tools every day. Free to watch, no sign-up, no paywall.

Start with Module 0 Open the playlist on YouTube
lessons

Eight modules. Watch in order.

Each module maps to one exam domain and ends with a lab you build in your own sandbox repo. Open the chapters under any lesson to jump straight to the moment you need. Videos play here; nothing loads from YouTube until you press play.

Module 0 Orientation

Orientation & the GH-600 Landscape

Exam map, the plan-act-evaluate model, and how to study.

What the exam certifies, the six weighted domains, the plan, act, evaluate mental model, why GitHub is the control plane, and how to study without dumps.

Chapters
Module 1 Domain 1 · 15-20%

Agent Architecture & SDLC Integration

Task contracts, inspectable plans, and the seven anti-patterns.

The agent task contract, why planning is separated from execution, structured plan artifacts, gating a plan before write actions, the seven anti-patterns and their fixes, and risk-based autonomy. Ends with Lab 1: the governance kit in your sandbox repo.

Chapters
Module 2 Domain 2 · 20-25%

Tool Use, MCP & Execution Environments

Tools, MCP servers, allow lists, and safe execution scope.

The smallest sufficient tool set, per-tool permissions across org, repo, agent, tool, and run, the MCP deep dive and why it is the highest-risk grant, the seven-question execution context checklist, agents in CI, and safe execution paths. Ends with Lab 2: the GitHub remote MCP server, a tool allow list, and the evidence workflow.

Chapters
Module 3 Domain 3 · 10-15%

Memory, State & Execution

Short versus long-term memory, and stopping context drift.

The three memory types and their risks, scoping memory to the task, the expiration, pruning, and reset lifecycle, durable artifacts and decision logs, resuming long-running work, detecting context drift, and the memory surfaces of a GitHub-centered SDLC. Ends with Lab 3: the memory and state policy.

Chapters
Module 4 Domain 4 · 15-20%

Evaluation, Error Analysis & Tuning

Judge output from evidence, classify root causes, then tune.

Success criteria as outcomes plus constraints, qualitative and quantitative signals, the six signal types, automated signals from code scanning, secret scanning, and dependency review, failure analysis from evidence, and the three root-cause categories before you tune anything.

Chapters
Module 5 Domain 5 · 15-20%

Multi-Agent Orchestration

Coordinate agents with isolation, handoffs, and an arbiter.

The five coordination patterns and when to use each, isolation rules for parallel execution, the five common conflicts and their fixes, observability across agents, review and audit artifacts, and handling failed, partial, and stalled executions.

Chapters
Module 6 Domain 6 · 10-15%

Guardrails & Accountability

Risk-based autonomy levels and least-privilege guardrails.

The L0 to L5 autonomy ladder, classifying actions by risk weighted by reversibility and blast radius, where human judgment belongs, fail-closed controls, least privilege, explicit authorization for irreversible changes, and the thirteen-guardrail catalogue.

Chapters
Module 7 All six domains

Capstone & Exam Readiness

A governed agent workflow, timed mock, and booking checklist.

Assemble the capstone, a complete governed agent SDLC in your sandbox repo, graded against a 100-point exam-weighted rubric. Then the phrase bank, red-flag answer patterns, timed-mock strategy, gap analysis, and the booking decision.

Chapters

Independent study material. Not affiliated with, endorsed by, or sponsored by GitHub or Microsoft. No exam questions or dumps. Grounded in the official Microsoft Learn and GitHub documentation.

the certification

The exam behind the name.

GH-600 is the exam code for GitHub Certified: Agentic AI Developer, a role-based certification for developers who operate, integrate, supervise, and govern AI agents inside production software workflows, with GitHub as the control plane.

The exam is delivered through Microsoft Learn and maintained by GitHub. It runs 120 minutes, is proctored through Pearson Vue, includes scenario-based questions, and is currently offered in English. The certification launched in beta; check the official page for current availability and pricing in your region.

Official exam and prep

What the exam measures

  • Agent architecture and SDLC processes (15-20%)
  • Tool use and environment interaction (20-25%)
  • Memory, state, and execution (10-15%)
  • Evaluation, error analysis, and tuning (15-20%)
  • Multi-agent coordination (15-20%)
  • Guardrails and accountability (10-15%)

Modules 1 to 6 follow these six domains in order; Module 0 is the orientation and Module 7 the capstone.

Want the study book too?

Eight modules of study notes in one 68-page PDF.

Download the study guide (PDF) Back to the GH-600 course page