By Amotion AIGH-700 · GitHub Certified: Agentic AI Business Professional
GH-700 study guide: every exam skill explained
This GH-700 study guide goes through every skill in Microsoft's official study guide for the GitHub Certified: Agentic AI Business Professional exam and says in plain English what each one means. Each of the four skill areas has a free lesson that teaches one topic in depth, with two sample questions and a short exercise in a test repository (you need a Copilot plan for the Copilot steps).
The exam has four skill areas. The percentages show the share of the exam each one takes, from the official study guide, last updated on 13 October 2026. Most questions cover generally available features, but the guide says preview features can appear when they are commonly used. GitHub Agentic Workflows are in public preview.
Now you can get Claude Certified and earn OpenAI badges and certifications too Here's how
Free GH-700 lessons
| Skill area | Weight | Free lesson |
|---|---|---|
| 1. Use GitHub platform features | 20–25% | Copilot budgets, AI credits and policies |
| 2. Enhance productivity by using GitHub Copilot | 20–25% | Copilot custom agents and guardrails |
| 3. Design and implement agentic workflows and automations | 25–30% | GitHub agentic workflows |
| 4. Perform human-in-the-loop tasks for agentic workflows | 20–25% | Human-in-the-loop for agents |
The GH-700 prep material has the full module for each skill area, with a guided lab for each one, in the Timo portal for Microsoft certifications.
1. Use GitHub platform features (20–25%)
This area tests whether you can set up and run work on GitHub the way an organization expects: within budget and policy, with the right model and instructions, and with work tracked in projects, issues and pull requests.
| Study guide line | What it means in practice |
|---|---|
| Evaluate budget guardrails | Know which Copilot budget stops spending and when. Since June 2026 Copilot is billed in AI credits. User-level budgets apply all month, while organization and enterprise budgets apply only after the included credits run out. |
| Evaluate token usage | Understand that cost depends on the model and the number of tokens, and know where to see usage: the AI usage page, the Usage this cycle card in Copilot settings, and cloud agent session logs. |
| Adhere to organizational standards | Work within the Copilot policies your enterprise or organization sets, such as which features, agents and models people may use. |
| Understand considerations for secrets management | Keep passwords and API keys out of prompts and files, and store them as secrets. Copilot cloud agent has its own Agents secrets, and only secrets named with a COPILOT_MCP_ prefix go to MCP servers (connectors to other systems). |
| Evaluate governance, privacy, usage, and risk controls | Know what policies, content exclusion and the audit log each control. For example, content exclusion keeps chosen files out of Copilot's context. |
| Apply Responsible AI principles | A person reviews what Copilot produces before relying on it. Instructions guide the model, but they do not guarantee it follows every rule. |
| Recommend branch strategies and policies | Use branch protection rules and rulesets so changes, including those from Copilot, reach the main branch only after review. |
| Identify how different models affect model output and explainability | Different models trade off speed, cost and quality. A larger model is not always the right one for a simple task. |
| Recognize considerations for choosing between models, including unique strengths of each mode | Choose a model for the task, within what your plan and your administrator allow, or use auto model selection. |
| Identify the characteristics of skills, prompts, and instructions | Custom instructions apply to almost every request, a skill loads only when its task comes up, and a prompt file is run by hand for one interaction. |
| Recommend when to use Model Context Protocol (MCP) servers | Use an MCP server when Copilot needs data or actions from another system. Skip it when built-in tools already do the job. |
| Create projects, including selecting an appropriate template | Start a GitHub project from a blank table, board or roadmap, or from a template. |
| Manage and customize views, including columns, automation, and custom fields | Give each team a view: table, board or roadmap, with its own columns, grouping, custom fields and built-in automations. |
| Configure access to GitHub projects | Set who can read, write or administer a project. Only project admins can add collaborators. |
| Manage issues, including creating, assigning, linking, and tagging | Create, assign, label and link issues, use sub-issues for large pieces of work, and set issue types. |
| Utilize the GitHub platform for collaboration | Use mentions, comments, reviews and discussions to work with others in the repository. |
| Create pull requests | Open a pull request to propose changes for review. |
| Assign pull requests | Assign the pull request to the people responsible and request reviewers. |
| Link pull requests | Link the pull request to the issues it resolves. Closing keywords such as "Closes #42" work only when the pull request targets the default branch. |
| Link project items | Add issues and pull requests to a project from their sidebar so the project shows their status. |
| Create and update wikis | Keep team documentation in the repository wiki. By default, only people with write access can edit it. |
Free lesson: Copilot budgets, AI credits and policies.
2. Enhance productivity by using GitHub Copilot (20–25%)
This area tests whether you can get reliable work out of Copilot: choosing where and how to chat, writing prompts that give consistent results, and setting up custom agents that stay within limits.
| Study guide line | What it means in practice |
|---|---|
| Choose an appropriate model for business purpose or task | Pick a faster model for simple work and a stronger one for complex analysis, within what your organization allows. |
| Manage sessions, including starting, resuming, renaming, pinning, and archiving | Start, resume, rename, pin and archive Copilot conversations and sessions. The controls differ between Copilot Chat on github.com, the GitHub Copilot app and Copilot CLI. |
| Select an agent | Choose which agent does the work, for example a custom agent from the agents panel instead of the default cloud agent. |
| Use @ mentions to attach content including tools, files, and issues | Attach the exact issues, files, pull requests or tools Copilot should use, so it works on the right data. |
| Use slash commands | Know what /explain, /fix, /tests, /doc, /clear, /delete and /chronicle do, and where each one works. For example, /chronicle summarises past sessions in Copilot CLI and the Copilot app. |
| Define information to be included in the response, such as sentiment, coherence, and creativity | Say what the answer must contain, such as tone, points to cover or level of detail. |
| Define the data to be acted on and input variables | Name the exact input data and any values that change each time, such as a date range or a project name. |
| Specify structure of the response, such as table format or maximum length | Ask for a fixed format, such as a table with named columns or a maximum length, so results can be compared. |
| Revert changes by using checkpoints | Go back to an earlier point when Copilot's changes are wrong. VS Code calls these checkpoints, and Copilot CLI uses /rewind. |
| Create a custom agent by adding web search to instructions or prompts | Write an .agent.md profile in .github/agents with a description, instructions and an optional tool list. If the agent needs web search, give it a web tool where the product has one and say in the prompt when to use it. Cloud agent does not currently support the web tool. |
| Save an agent to a GitHub project | The study guide lists this line, but GitHub's documentation does not describe a feature by that name. Custom agents are saved in a repository or organization, and agent sessions can be shown in project views (GitHub changelog). |
| Implement agent guardrails | Limit what an agent can do, mainly through its tools list, which is an allow-list. |
| Review custom agent activities | Check session logs, the pull request (which names the agent used) and the audit log. |
Free lesson: Copilot custom agents and guardrails.
3. Design and implement agentic workflows and automations (25–30%)
The largest area. It tests whether you can choose work worth automating and set up an AI agent to do it safely. An agentic workflow is a Markdown file of instructions that an agent follows in GitHub Actions on a schedule or when something happens. A Copilot automation is a separate feature that starts Copilot cloud agent with a saved prompt when a trigger fires.
| Study guide line | What it means in practice |
|---|---|
| Identify use cases that can be optimized by agentic workflows or automations | Pick repeated, well-defined work whose result a person can check, such as issue triage or a weekly report. |
| Decompose use cases into processes and tasks | Break the work into steps, and decide which steps the agent does and which a person does. |
| Specify decision-making criteria | Write down the rules the agent applies, such as which label goes on which kind of issue. |
| Align agentic workflows with common design patterns | Reuse known shapes, such as issue triage, scheduled reports and commands people type in an issue comment, instead of designing from scratch. |
| Create agentic workflow files | Write a Markdown file in .github/workflows with settings at the top, then compile it with gh aw compile (a command-line tool) and commit both files. |
| Configure workflow, schedule, permissions, and output | Set the trigger or schedule, keep permissions read-only, and allow only the writes the workflow needs as safe outputs. |
| Configure folder structure for an agentic workflow | Know where workflow files, shared pieces of instructions, skills and custom agents live in the repository. |
| Create agentic workflow instructions | Write the instructions as you would for a new team member: specific, with context, and with what to do in unusual cases. |
| Configure self-critique loops | Have the agent's plan or output reviewed before it is accepted, for example by the rubber duck agent (a different model that critiques the work) in Copilot CLI and the Copilot app, or by tests that must pass. |
| Select and configure triggers | Choose what starts the automation: a schedule, a new issue, or a pull request opened or updated. |
| Configure prompts and models | Write the automation's prompt and, optionally, choose its model. |
| Select and configure tools within automations | Choose the tools the automation may use. GitHub calls this the main way to control what an automation can do. |
| Install external skills | Add a skill from another repository, after checking what it contains. |
| Create custom skills | Write your own SKILL.md file with a name, a description and instructions. |
| Call skills | Let Copilot load a skill when the task matches it, or call one by name. |
| Connect agentic workflows to MCP servers to improve context | Give the workflow data from another system through an MCP server, with only the tools it needs allowed. |
Free lesson: GitHub agentic workflows.
4. Perform human-in-the-loop tasks for agentic workflows (20–25%)
This area tests whether you stay in control of agents: where a person must decide, how much freedom the agent gets, and what you do when it needs approval, goes wrong or must be stopped.
| Study guide line | What it means in practice |
|---|---|
| Identify critical decision points where human judgment is required | Mark the steps with legal, financial or customer impact, and keep a person in charge of them. |
| Choose how the agent will perform work, including interactive, plan, and autopilot modes | In Copilot CLI and the Copilot app, interactive mode works step by step with you, plan mode shows a plan first, and autopilot carries on by itself. |
| Design and configure escalation paths based on tool usage | Decide which tool use needs approval, and block risky actions outright. In Copilot CLI, deny rules beat allow rules. |
| Respond to requests for interaction from agentic workflows, including approvals, rejections, or revisions | Approve or reject a waiting run, or ask Copilot for changes on its pull request. |
| Interact with an agentic workflow while it is running | Send steering messages to a running Copilot session to change direction. |
| Interact with agentic workflows to resolve failures | Find why a run failed, fix the cause and run it again. |
| Stop an agentic workflow that is running | Stop a cloud agent session or cancel the workflow run. Work already pushed is kept. |
| Ensure that agentic workflows meet requirements | Check the results against what the workflow was meant to do before relying on it. |
| Review agentic workflow outputs | Read the issues, comments and pull requests the agent produced, and its logs. |
| Monitor agentic workflow runs | Watch run status, duration, token use and cost over time. |
Free lesson: Human-in-the-loop for agents.
Test yourself
The free GH-700 practice test has 20 questions across all four skill areas in 40 minutes, with an explanation and a source link for every answer. For full tests of 200 questions and prep material, click here. For exam facts, the beta discount and the GH-300 comparison, read the GH-700 exam guide.
Now you can get Claude Certified and earn OpenAI badges and certifications too
- Register for all four Claude certification exams through Amotion AI, which is Certified in the Claude Partner Network. Claude Certified Associate covers business and knowledge work, not coding.
- Take 11 OpenAI PartnerU learning programs with assessments and course-completion certificates, and work towards the expert badges for ChatGPT, Codex and APIs.
- Practise with 20 full-length Claude mock exams.
Independent Timo practice material. Not affiliated with or endorsed by GitHub or Microsoft.