Can I add a system prompt to GitHub Copilot?

GitHub Copilot reads a .github/copilot-instructions.md file in your repository as custom instructions. That is the equivalent of a system prompt for Copilot: you write it, commit it, and Copilot applies it to suggestions and chat. It is not an API system message, but it serves the same purpose.

Can I see Microsoft's Copilot system prompt?

No. Microsoft's own system instructions for Copilot products are not public. They set the base behavior, safety constraints and content policies. You work within them, not around them.

Do Copilot custom instructions persist between sessions?

It depends on the product. GitHub Copilot reads its instructions file from the repository, so they persist as long as the file is there. Microsoft 365 Copilot and the standalone assistant handle persistence differently and may require you to re-enter instructions for each session.

Can I use the same instructions for GitHub Copilot and other AI tools?

Not automatically. GitHub Copilot reads its repository instructions file; other tools may use different formats. A shared record outside any single tool gives you one place to maintain the rules, which you can then reference or copy into each tool's instruction layer.

Why does Copilot ignore my instructions sometimes?

Either the instructions are in a layer that does not apply to the current feature (for example, instructions in one product do not carry to another), or Microsoft's product-level constraints are overriding them. Check which layer your instructions are in and whether the behavior you want is permitted by Microsoft's usage policies.

Copilot System Prompt: What Can You Actually Change?

The answer depends on which Copilot product you are using. Some let you add standing instructions; others do not. Here is what each product allows and what Microsoft controls.

June 24, 2026Updated October 21, 20264 min read

"Copilot" covers a family of Microsoft products that each handle user-editable instructions differently. GitHub Copilot, Microsoft 365 Copilot, and the standalone Microsoft Copilot assistant all have system-level instructions you cannot read or change, but some of them also have layers where you can add your own standing instructions.

Knowing which layer you are working in matters. Instructions you add at the wrong level either do not apply or do not persist. Tools like fenbs keep reusable work rules in a shared record outside any single Copilot product, so they remain available when you switch between Microsoft tools or when a session ends.

GitHub Copilot: what you can configure

GitHub Copilot reads a .github/copilot-instructions.md file in your repository as custom instructions. Copilot applies those instructions to suggestions and the chat feature in supported editors. This is the closest equivalent to a system prompt for GitHub Copilot: you write the file, commit it to the repository, and Copilot picks it up.

The instructions file is the right place for project-specific rules: the coding language and version, style conventions, architecture patterns to follow, and constraints on what Copilot should not change. Keep it short and specific; a long file risks diluting the most important rules.

Beyond that file, you can add per-request context through the chat interface. That context applies to one session, not the whole project. GitHub Copilot's cloud agent (formerly the coding agent) has its own behavior rules that Microsoft sets; those are not user-configurable.

Microsoft 365 Copilot: prompts and custom instructions

Microsoft 365 Copilot is the AI layer in Word, Excel, Teams, Outlook and other Microsoft 365 products. It does not use a system prompt in the API sense. Your instructions come through the prompts you send in each session, or through custom prompts you can save in the Copilot interface.

Some Microsoft 365 Copilot configurations allow administrators to set organization-level instructions that apply to all users. Those are set by your organization's IT team, not by individual users. Individual users can save prompts for their own reuse within the product, but those are session prompts, not standing instructions that apply automatically.

For building custom Microsoft Copilot agents with their own system prompts and instructions, see Copilot Studio and Microsoft Copilot agents. Those products give you explicit control over the instructions each agent follows.

Microsoft Copilot (the standalone assistant)

Microsoft Copilot, the standalone AI assistant, accepts instructions through its chat interface. You can include standing instructions at the start of a conversation or use a persistent memory feature if your account has it enabled. There is no user-accessible system prompt field in the API sense.

Microsoft's own system instructions for this product are not public. They set the assistant's behavior, safety constraints and capabilities. You cannot override them through user-level instructions, but you can work within them by being specific about the role, format and constraints you want for each conversation.

What Microsoft controls

Across all Copilot products, Microsoft sets the base behavior through training and product-level system instructions. Those cover safety, content policies, and the way the assistant handles certain types of requests. They are not visible to users and cannot be overridden through user-level instructions or the repository instructions file.

If a Copilot product refuses a request your instructions seem to permit, the refusal is usually coming from a Microsoft-controlled constraint, not from something you have set incorrectly. Check Microsoft's usage policies for the relevant product before assuming the instructions file needs changing.

See Copilot vs ChatGPT for a broader comparison of how the two products handle user-provided instructions.

How fenbs helps

Keep your work rules outside any one Copilot session

fenbs stores reusable work rules and prompts that any connected AI app can read. Add the rules you want GitHub Copilot, Claude, or another tool to follow as skills or Decisions. When a session ends or you switch products, the rules are still there for the next session.

Create a free account

Free for one person · No credit card · Works with Claude, ChatGPT, Cursor and more

Keeping instructions reusable across tools

One practical problem with Copilot instructions is that they are product-specific. The .github/copilot-instructions.md file only applies to GitHub Copilot. Instructions you save in Microsoft 365 Copilot only apply there. If you work across multiple AI tools, you end up maintaining separate instruction sets for each one.

A shared project record solves this. Write the rules that should apply across all your tools once, in a central place, and reference that record from each tool's instruction layer. That way a change to a project rule only needs to happen once.

How to do this with fenbs

Create a free account and add a project for the codebase or work you are configuring Copilot for.
Write a .github/copilot-instructions.md for project-specific GitHub Copilot rules.
Add the most important rules as Decisions in fenbs so the reasoning is recorded and the rules are available to other connected tools.
Add reusable task-type instructions as skills in fenbs.
Connect your other AI tools over MCP from Integrations.
At the start of each session, ask your AI to read the project context so it starts from the current approved rules.

fenbs does not integrate directly with Microsoft 365 Copilot or the standalone Microsoft Copilot. It connects to Claude, Claude Code, ChatGPT, Cursor, Codex CLI, Gemini CLI, GitHub Copilot and Devin Desktop over MCP. For products not on that list, the shared record is still useful as the single place where rules are written, even if you paste them manually.