Developing One Product with Claude Code and Codex

Two AIs do not automatically share the same context merely because they use the same GitHub repository.

The key is to share code through Git and share long-term context and policies through repository files.

Recommended Structure

MyProduct/
├─ CLAUDE.md
├─ AGENTS.md
├─ .ai/
│  ├─ PRODUCT.md
│  ├─ ARCHITECTURE.md
│  ├─ DEVELOPMENT_POLICY.md
│  ├─ GIT_POLICY.md
│  ├─ UX_POLICY.md
│  ├─ TEST_POLICY.md
│  ├─ DECISIONS.md
│  └─ CURRENT_STATE.md
└─ Source/

flowchart TD
    C[Claude Code] --> CM[CLAUDE.md]
    X[Codex] --> AM[AGENTS.md]
    CM --> A[.ai/ Shared Context]
    AM --> A
    A --> R[Repository Source]

Why This Structure Is Needed

Claude and Codex do not automatically share their conversation histories.

For example, suppose a Claude session makes the following decision.

Place the Blend Shape creation UI in the Avatar > Blend Shape tab.

If this decision remains only in the chat, Codex cannot know it. Externalize lasting decisions into a file such as .ai/DECISIONS.md.

Do Not Develop in Parallel in the Same Working Tree

Poor structure:

The same folder
├─ Claude edits Camera.cpp
└─ Codex also edits Camera.cpp

Recommended:

flowchart TD
    G[Git Repository] --> S[stable]
    S --> A[Worktree A<br/>feature/camera]
    S --> B[Worktree B<br/>feature/avatar]
    A --> C[Claude Session]
    B --> X[Codex Session]

Each AI works in its own branch and worktree, with cross-verification at the PR stage.

Cross-model Review

Claude implements
→ PR
→ Codex Review
→ Claude revises
→ Human / CI checks
→ Merge

Codex implements
→ PR
→ Claude Review
→ Codex revises
→ Human / CI checks
→ Merge

The goal is not to make the models identical. Instead:

The same source of truth + different reasoning paths + mutual verification

is more valuable.

The Role of CURRENT_STATE.md

This is an index for quickly reconstructing the current work.

Owner	Branch	Task	State
Claude	`feature/camera`	Camera animation	In Progress
Codex	`feature/library`	Library browser	Review

Even if a session is interrupted, a new agent can read the repository and recover the current state.

Tools That Can Be Added Later

Keeping shared policies in .ai/ means that adding a new coding agent requires only an entry file for that agent.

Claude → CLAUDE.md ─┐
Codex  → AGENTS.md ─┼→ .ai/*
Other  → ENTRY.md  ─┘

This creates a development operating structure that is not tied to a particular AI product.