Coredence

Guide

Copilot instructions in VS Code: the files it reads, and the one that follows you

Copilot in VS Code starts every chat knowing nothing about you. Instruction files are how you change that — and one of them follows you into every project.

Ahmet Karaca ·

The files Copilot reads

  • .github/copilot-instructions.md in a repository — instructions for that project, committed with the code.
  • .github/instructions/*.instructions.md — instructions that apply only to matching files, chosen with an applyTo glob in the file's front matter.
  • AGENTS.md at the repository root — a format several agents share, Codex among them.
  • ~/.copilot/copilot-instructions.md in your home folder — your personal instructions, applied in every workspace.

They add up: Copilot reads every file that applies, and VS Code's documentation says not to rely on any order between them. If two files disagree, fix one of them.

They shape chat and agent work, not the suggestions that appear while you type.

The personal file

Project files describe the project. The personal file describes you: how you want changes made, what you never want done without asking, how much explanation you want. It lives in your home folder, so it is the same in every repository and never ends up in someone else's.

VS Code's own advice is to keep each instruction short and self-contained, and to say why when it is not obvious — a model that knows the reason applies the rule to cases you never wrote down.

- Read the existing file before proposing a new one.
- Ask before deleting files or running destructive commands.
- Plan in a few lines before touching more than three files.
- Prefer a formula or a named constant to a hard-coded number:
  the number will change and the reason will be forgotten.

Where good instructions come from

Not from a blank page. The instructions that change behaviour are the ones written after Copilot got something wrong: each time you said "no, not like that", the default was off in a way a sentence could fix.

The trouble is that those corrections are scattered across months of chats, and across every assistant you use. The personal file for Copilot, CLAUDE.md for Claude Code, AGENTS.md for Codex, a rule in Cursor — each holds its own copy of you, and they drift apart.

Where Coredence fits

Coredence measures that picture instead of asking you to write it. It reads your own sessions, commits and notes, consolidates them into memories about how you work, and keeps every memory tied to the traces it grew from. If you turn it on, it writes your mind into ~/.copilot/copilot-instructions.md and refreshes it every cycle — in its own marked section, leaving whatever you wrote there yourself untouched. GitHub Copilot CLI reads the same file; Claude Code, Codex and Gemini get the same mind in their own formats.

You can try the real app on a made-up person's mind, in your browser, without installing anything.

Try it in your browser