What if your AI coding assistant forgot your project's conventions every time you started a new session? That is the default experience, and a growing number of developers are fixing it the same way they solved shell configuration decades ago: dotfiles.
The concept is straightforward. Just as .bashrc configures your terminal and .gitconfig sets your Git preferences, files like CLAUDE.md, .cursorrules, and .github/copilot-instructions.md tell AI assistants how to behave in your codebase. Check them into version control, and every team member's AI assistant follows the same rules.
What Goes in These Files
The contents are surprisingly practical. Build commands ("use pnpm, not npm"), coding standards ("2-space indentation, types everywhere"), architecture context ("API handlers live in src/api/handlers/"), and guardrails ("never commit .env files"). The best entries are specific and verifiable. "Format code nicely" does nothing. "Run npm test before committing" gives the AI something concrete to follow.
More advanced setups include path-scoped rules that only activate for certain directories, reusable "skills" that capture solutions to recurring problems, and even hooks that prompt you to save a fix as a skill when you solve the same problem twice.
Which Tools Support What
| Assistant | Config File | Scope |
|---|---|---|
| Claude Code | CLAUDE.md, .claude/rules/*.md |
Project, user, local |
| GitHub Copilot | .github/copilot-instructions.md |
Repository |
| Cursor | .cursorrules, .cursor/rules/ |
Project |
| Windsurf | .windsurf/rules/*.md |
Workspace, global |
| Aider | CONVENTIONS.md via config |
Project |
Claude Code and Windsurf have the most mature systems, with support for modular rule files, path-scoping, and hierarchical overrides (global, project, local). Cursor and Copilot are simpler but functional.
The practical advice: keep each file under 200 lines, use project-level files for team standards and user-level files for personal preferences, and review periodically to remove outdated instructions. Conflicting rules across files can cause unpredictable behavior since the AI may pick either one arbitrarily.
For teams already using AI coding assistants daily, spending 30 minutes writing a good config file is one of the highest-leverage improvements you can make. It turns a general-purpose assistant into one that actually knows your codebase.