Organization
Where you store AI-related files affects how well tools can find and use them.
The Problem
Different AI tools look for configuration in different places. Claude Code looks for CLAUDE.md. Cursor has its own config. Copilot expects certain patterns. If you scatter prompts and instructions across tool-specific locations, you end up duplicating content and maintaining multiple copies.
A Vendor-Agnostic Approach
Store reusable prompts and instructions in a vendor-agnostic .ai folder. Then create symlinks for each tool you actually use.
.ai/
prompts/
code-review.md
commit-message.md
context/
architecture.md
conventions.md
.claude/
CLAUDE.md -> ../.ai/context/conventions.mdThe .ai folder holds the source of truth. Tool-specific folders just point to it.
Plans and Context
For project plans and longer-term context, consider where files should live:
- In the repo: If the context is project-specific and should travel with the code
- Outside the repo: If plans span multiple projects or contain notes you don’t want committed
Some teams keep a top-level plans/ folder in their projects directory for cross-project planning, with subfolders per project.
Namespacing
When you accumulate many prompts, naming conventions help you find them. Prefix related prompts with a namespace:
skill.review,skill.commit,skill.refactorseed.architecture,seed.conventions,seed.codebase
When tools list prompts alphabetically—or when you’re browsing a folder—everything in the same namespace groups together. You immediately see all your context-seeding commands or all your skill-related prompts.
This matters especially when running tools that don’t auto-load skills, or when you want to explicitly invoke them. A consistent naming scheme turns a pile of prompts into an organized system.
The Pattern
Centralize AI instructions in a vendor-agnostic location. Use symlinks to expose them to specific tools. Namespace prompts so related commands cluster together. Keep plans where they make sense for your workflow.