Source: Dev.to

How to Use Markdown Files to Orchestrate Multi‑File Changes in GitHub Copilot

---

AI‑assisted coding is everywhere. Tools are fast, accessible, and can turn a simple prompt into working code in seconds.

But anyone using AI seriously for development knows the pattern:

1. You ask for code.
2. You clarify intent.
3. You fix structure.

Back and forth. Each iteration consumes time, and time is a finite resource.

If AI usage becomes more metered in the future—per prompt, per token, per request—clarity becomes critical. Every vague instruction has a cost.

This is where structured prompting inside the codebase, especially with GitHub Copilot, becomes a serious productivity multiplier.

---

The Problem: Repetitive Multi‑File Changes

In real‑world projects, features rarely touch just one file. Adding something small like a new filter or API integration often requires:

• Updating types
• Modifying API services
• Updating hooks
• Changing state management
• Adjusting UI components
• Possibly updating pages
• Maybe touching tests

That is easily 6–8 files for a “minor” feature. Doing this manually is repetitive; doing it via casual chat is fragmented. The issue isn’t AI capability; it’s a lack of structure and context.

---

The Solution: Structured Markdown Prompting

Instead of typing ad‑hoc instructions into the chat sidebar, define your requirements inside the repository using Markdown files. This creates a blueprint that Copilot can use to orchestrate changes across the workspace.

---

Step 1 — Define the Project Source of Truth

Create a Project DNA file to serve as the source of truth for your codebase. By explicitly defining your tech stack, folder hierarchy, and architectural boundaries, you eliminate “AI guesswork” and ensure Copilot respects your specific design patterns from the very first line.

Location: docs/PROJECT_CONFIGS.md

Example:

---

Step 2 — Create Workflow Blueprints

For specific, repetitive tasks, create targeted instruction files.

Example: docs/workflows/API_SERVICE.md

Describe exactly how a service should be built: where it lives, how it handles errors, and what the import order should be.

---

Step 3 — Orchestrating with the Coding Agent

This is where the magic happens. Instead of using the standard chat, switch to the GitHub Copilot Coding Agent and select a high‑reasoning model like Claude 3.5 Sonnet or Claude 4.5 (if available in your tier).

1. Attach Context – Add your project-config.md and any relevant source files to the chat context.

2. Issue the Command – Use a concise prompt that references your workflow file:

   /new service LoginService - Follow rules in docs/workflows/service.md

---

Step 4 — Controlled, Multi‑File Output

While standard chat provides snippets you have to copy‑paste manually, the Agentic Workflow handles the heavy lifting by natively generating and placing files directly into your codebase.

The agent can create:

• A new types file
• A new service file
• Correct imports
• Strict typing
• Proper folder placement
• No UI‑logic leakage

Everything aligns with the existing project conventions defined in Markdown.

---

Conclusion

GitHub Copilot performs significantly better when:

• Context is explicit
• Conventions are documented
• Scope is clearly defined
• Architecture rules are enforced

Markdown becomes a lightweight contract between the developer and the AI.

My GitHub Repo for the Project

External Links: (add any additional URLs here)

• GitHub Copilot Prompt Files Guide

• AI Prompting Techniques – MindStudio University