Claude Codex Memory Sync — Development skill for Claude Code

Claude Code → Codex Memory Sync

Convert Claude Code project auto-memory into local Codex `ad_hoc` staging notes with one PowerShell command.

[](https://github.com/DaizeDong/claude-codex-memory-sync/actions/workflows/test.yml) [](sync-claude-memory-to-codex.ps1) [](LICENSE) [](#languages)

[English](README.md) | [中文版](README_CN.md)

This is an independent community project. It is not affiliated with or endorsed by Anthropic or OpenAI.

⭐ Read this first: the design philosophy

**Stage verifiable memory updates; never pretend the two agents share a brain.**

This tool is intentionally small. Its design follows three principles:

1. Fill the seam, do not add another agent surface. Claude Code already stores project memory, and Codex already owns its memory consolidation. This tool only converts one representation into the format accepted by the detected local ingress contract. It adds no agent terminal, daemon, MCP server, database, vector store, or model call.
2. Stage through the detected local contract, do not impersonate Codex internals. The script appends self-contained notes under extensions\ad_hoc\notes\. It never rewrites Codex memory summaries, rollout evidence, or SQLite state. A successful sync means “safely staged,” not “already consolidated or guaranteed to be recalled.”
3. Memory quality matters more than memory volume. Dry-run first, select conservatively, fail closed on likely credentials, bound every input, and deduplicate incrementally. Copying every log and stale decision would make the pool larger while potentially making recall worse.

The design target is the smallest audit