Claude Code vs Copilot vs Codex repo structure
A common question is:
What belongs in Claude files, what belongs in GitHub Copilot files, and what belongs in Codex files?
Claude side
Claude-specific repo files usually include:
CLAUDE.md.claude/rules/.claude/agents/.claude/skills/.claude/settings.json
Use these for Claude-specific instructions, specialist roles, reusable workflows, and hooks.
GitHub Copilot side
GitHub Copilot repo files usually include:
.github/copilot-instructions.md.github/instructions/.github/agents/.github/skills/.github/hooks/
Use these for Copilot-specific instructions, scoped guidance, agents, skills, and optional Copilot hooks.
The .github/hooks/ folder is specifically for GitHub Copilot hook configs, which run in Copilot CLI and Copilot cloud agent. It is not a generic GitHub repository feature and it is separate from GitHub Actions.
OpenAI Codex side
OpenAI Codex repo files usually include:
.codex/agents/.codex/hooks.json
Codex has no tool-specific instruction file: it reads AGENTS.md directly. It discovers skills in .agents/skills/, a cross-agent folder that GitHub Copilot also reads.
Shared layer
Some things should stay outside the vendor-specific layers:
README.mdAGENTS.md- external/shared
skills/
These help keep the repo understandable across tools.
The simplest mapping
Think of it like this:
CLAUDE.md↔.github/copilot-instructions.md.claude/rules/↔.github/instructions/.claude/agents/↔.github/agents/↔.codex/agents/.claude/skills/↔.github/skills/↔.agents/skills/(Codex and Copilot).claude/settings.jsonhooks ↔.github/hooks/↔.codex/hooks.json
They are not always identical features, but they play similar structural roles.
Why combining them is useful
Keeping each tool’s files on its own side in one repo helps teams:
- keep tool-specific behavior organized
- avoid mixing unrelated instructions together
- make the repo easier to maintain
- teach others with a real example
Next page
Read Getting started.