Skip to content
Cheat sheets
Associate · GH-600Prep available

GitHub Certified: Agentic AI Developer (GH-600) practice exam and study path

A free, timed 50-question GH-600 practice exam in 120 minutes, scored by domain, with every option explained after you submit. Also 126 practice questions you can check one at a time, a page for every task in the exam guide, and a study path. No sign-up.

Start the free practice examGet the exam packPractise question by question →
Agentic AI Developer exam pack:

$6.99for 12 months

For developers who run AI agents inside real delivery work on GitHub: planning and acting, tools and MCP servers, memory and state, evaluation and tuning, coordinating several agents, and the guardrails and approvals that keep autonomous work safe.

The exam

Exam code
GH-600
Questions
Not yet published
Time
120 minutes
Passing score
700 of 1,000 (scaled)
Exam fee
$165 (USD)
Question types
multiple choice, multiple response

Exam GH-600 is a proctored exam of 120 minutes, offered in English; it may include interactive components, and 700 passes. Microsoft doesn't publish the number of questions (most of its exams have 40 to 60). The fee is set by the country or region (USD 165 in the United States). GitHub certifications are valid for two years. Our forms use 50 questions split 9 / 11 / 6 / 9 / 9 / 6, the midpoints of the guide's ranges.

From GitHub's certification page and exam guide, checked Oct 25, 2026.

Domains and weights

  • Domain 1: Prepare agent architecture and SDLC processes

    18%

  • Domain 2: Implement tool use and environment interaction

    22%

  • Domain 3: Manage memory, state, and execution

    12%

  • Domain 4: Perform evaluation, error analysis, and tuning

    18%

  • Domain 5: Orchestrate multi-agent coordination

    18%

  • Domain 6: Implement guardrails and accountability

    12%

On our roadmap this comes after GitHub Copilot. See every certification.

Practice

126 original practice questions, every option explained with the GitHub documentation page that proves it. Free, no sign-up; progress stays on this device.

Your results on this device →

More practice exams

The lessons, the sample questions on every task page and the first full practice exam are free and stay free. A pack adds the further timed GH-600 practice exams.

Agentic AI Developer exam pack

$6.99for 12 months

  • Every paid GH-600 practice exam, timed, with every option explained
  • The next version of this exam included when it changes within your 12 months
  • 12 months of access. One payment, nothing renews by itself.
  • 14-day refund, no conditions
  • 14-day refund, no conditions: ask within 14 days of buying and you get the whole payment back.
  • No pass guarantee. These are practice exams written by us from the exam guide and the GitHub documentation; a score here does not predict your result on the real exam.

Study path

One module per exam domain. Lessons that already exist on BytePatterns are linked; the rest are being written.

  1. Module 1: Prepare agent architecture and SDLC processes

    1. 1Which delivery steps to give an agent, and the tasks to keep for peopleTask 1.1

      Partly covered by: From Pilot to Production

    2. 2Agent anti-patterns, and an agent task's inputs, outputs and acceptance criteriaTask 1.1

      Partly covered by: Agents and Tools

    3. 3Plan mode, research and plan sessions, and structured plansTask 1.2

      Lesson coming

    4. 4Validating a plan and holding action until it is approvedTask 1.2

      Partly covered by: The Tool-Use Loop

    5. 5Autonomy settings, session logs, commits and pull requests as inspectable artifactsTask 1.3

      Partly covered by: Observability Basics

    6. 6Steering, comments and approvals that keep people in control without slowing deliveryTask 1.3

      Lesson coming

  2. Module 2: Implement tool use and environment interaction

    1. 7Choosing and configuring tools for custom agents, Copilot CLI and the Copilot SDKTask 2.1

      Partly covered by: Agents and Tools, The Tool-Use Loop

    2. 8Tool permissions: allow and deny rules, tool lists and pre-approved toolsTask 2.1

      Partly covered by: Guardrails

    3. 9MCP servers for repositories and custom agents, and the remote GitHub MCP serverTask 2.2

      Lesson coming

    4. 10MCP registries, allowlists and enterprise-managed settingsTask 2.2

      Lesson coming

    5. 11Execution context: copilot-setup-steps.yml, runners, the firewall and Agents secretsTask 2.3

      Lesson coming

    6. 12Repository and branch scope, CI invocation, and agent-opened branches and pull requestsTask 2.3

      Lesson coming

    7. 13Error handling, retries, rollbacks and escalation paths for agentsTask 2.4

      Partly covered by: Message Queues

    8. 14Traceability: agent commits, session logs and hooksTask 2.4

      Partly covered by: Tracing a Request

  3. Module 3: Manage memory, state, and execution

    1. 15Short-term, long-term and external memory: context windows, Copilot Memory and filesTask 3.1

      Partly covered by: Context Windows

    2. 16Memory scope, expiry, pruning and resetTask 3.1

      Lesson coming

    3. 17Durable progress: plans, checkpoints, session state and resuming workTask 3.2

      Lesson coming

    4. 18Detecting and correcting drift in long agent runsTask 3.2

      Partly covered by: Context Windows

    5. 19Sharing state across sessions, surfaces and agents without conflicting or stale contextTask 3.3

      Lesson coming

  4. Module 4: Perform evaluation, error analysis, and tuning

    1. 20Expected outcomes, constraints and acceptance criteria for agent tasksTask 4.1

      Lesson coming

    2. 21Qualitative and quantitative signals, and signals from automated scanningTask 4.1

      Partly covered by: Evaluating LLMs, LLM as a Judge

    3. 22Finding failures in logs, plans, traces, outputs and workflow artifactsTask 4.2

      Partly covered by: Observability Basics

    4. 23Classifying root causes: reasoning errors, tool misuse, context and environment issuesTask 4.2

      Lesson coming

    5. 24Tuning instructions, workflows, constraints, memory and tool accessTask 4.3

      Partly covered by: Evaluating LLMs

  5. Module 5: Orchestrate multi-agent coordination

    1. 25Orchestration patterns: orchestrator and subagents, /fleet, dynamic workflowsTask 5.1

      Partly covered by: Agents and Tools

    2. 26Isolating parallel agents and resolving overlapping or contradictory workTask 5.1

      Lesson coming

    3. 27Reviewable artifacts, handoffs and post-hoc analysis across agentsTask 5.2

      Partly covered by: Tracing a Request

    4. 28Failed, partial and stalled agent runs, and recovery with rollback and human reviewTask 5.3

      Lesson coming

    5. 29Adding, updating, replacing and retiring agents without breaking workflowsTask 5.4

      Lesson coming

  6. Module 6: Implement guardrails and accountability

    1. 30Classifying agent actions by risk and assigning autonomy levelsTask 6.1

      Partly covered by: Guardrails

    2. 31Human-in-the-loop: rulesets, code owners, approvals and blocked actionsTask 6.2

      Partly covered by: Guardrails

    3. 32Least privilege and controlled paths for irreversible changesTask 6.2

      Lesson coming

Every task in the exam guide

The exam guide splits each domain into task statements. Each one has a page with a short summary, the lessons that teach it and open sample questions.

Domain 1: Prepare agent architecture and SDLC processes

Domain 2: Implement tool use and environment interaction

Domain 3: Manage memory, state, and execution

Domain 4: Perform evaluation, error analysis, and tuning

Domain 5: Orchestrate multi-agent coordination

Domain 6: Implement guardrails and accountability

How these questions are made

  • Written by us from the exam guide's task statements and the GitHub documentation — never from real exam content.
  • Every option carries its own explanation and a link to the GitHub page that proves it.
  • A question enters a timed exam only after an independent check; until then it is a practice question.
  • Results are a plain percentage, per domain too — not an imitation of GitHub's scaled score.

Questions about the GH-600 practice exam

Is the GH-600 practice exam free?
Yes. The full practice exam (50 questions, 120 minutes) is free with no sign-up, and so are the 126 practice questions, the task pages and the study path. Your attempts are saved in this browser, on this device.
How many questions does it have, and is it timed?
50 questions in 120 minutes: the real GH-600 pace, scaled to this form. When the time runs out, the exam is submitted as it stands.
Are the answers explained?
After you submit, not during: like the real exam, the timed exam gives no feedback until the end. Then you get your score overall and per exam domain, and every question with each option explained and a link to the GitHub documentation page behind it. In practice mode an answer is explained as soon as you check it.
What does the Agentic AI Developer exam pack add?
3 more timed practice exams (Practice exam 1: 50 questions, 120 minutes; Practice exam 2: 50 questions, 120 minutes; Practice exam 3: 50 questions, 120 minutes), each scored and explained the same way, and the next version of this exam if it changes within your 12 months. 12 months of access. One payment, nothing renews by itself. Everything above stays free. See the pack.
Can I get a refund?
Yes, with no conditions: ask within 14 days of buying and you get the whole payment back. The details are in the terms.
Is this an official GitHub practice exam?
No. BytePatterns is not affiliated with, endorsed or sponsored by GitHub. Every question is written by us from the public exam guide and the GitHub documentation, never from real exam content, and a score here does not predict your result on the real exam. GitHub lists its own preparation resources on its certification page.