GH-600 · GitHub Agentic AI Developer
Certification: GitHub Certified: Agentic AI Developer · Taken: May 2026 (beta) Official: GH-600 study guide
GH-600 tests whether you can run AI agents inside GitHub safely: Copilot cloud agent, custom agents, MCP servers, hooks and Actions, together with the controls around them. Almost every question is a control-plane question. Who is acting, within what boundary, with which tool, producing what evidence, gated where, and undone how?
Domains
| Domain | Weight |
|---|---|
| Tool use and environment interaction | 20–25% |
| Agent architecture and SDLC processes | 15–20% |
| Evaluation, error analysis and tuning | 15–20% |
| Multi-agent coordination | 15–20% |
| Guardrails and accountability | 10–15% |
| Memory, state and execution continuity | 10–15% |
Pages in this section
- Study guide: the scenario parser and decision tables for every domain.
- Gap notes: the six boundaries where plausible answers go wrong, and "choose two" traps.
- Hands-on checklist: things to try yourself before the exam.
- Resources: courses, docs, talks and eval frameworks.
The scenario parser
Fill this in before you look closely at the answer choices:
| Slot | Question to answer |
|---|---|
| Actor | Who acts: user, cloud agent, custom agent, workflow, MCP server, external system, reviewer? |
| Scope | What boundary matters: org, repo, branch, PR, issue, workflow, environment, secret, token, network? |
| Tool | Which mechanism: built-in Copilot tool, MCP tool, hook, GitHub Actions, scan, review, custom agent? |
| Artifact | What proves progress: plan, diff, PR, check, workflow log, trace, session log, audit event? |
| Gate | Where do humans decide: before mutation, merge, deploy, secret access, external impact? |
| Rollback | How do you undo or stop it: close the PR, revert, rerun, disable the tool, revoke the token, reset memory, escalate? |
If an option doesn't name an artifact or a control boundary, it's probably too vague.