Agent Sync

agsy (agent-sync)

[](https://github.com/IngSquared99/agent-sync/releases) [](LICENSE)

Merge & mount tool for multi-source AI instruction files: one set of sources, synced to Claude Code, Codex, Antigravity, and more.

**[📘 Full documentation](https://ingsquared99.github.io/agent-sync/#/en/)** ｜ [繁體中文 README](README.zh-TW.md)

---

A. Core Concepts: What is agsy?

The problem it solves

When you use several AI development tools at once (Claude Code, OpenAI Codex, Google Antigravity, Cursor…), each tool reads its own instruction locations, in its own shapes:

• Claude Code reads rules from .claude/rules/ and skills from .claude/skills/
• Codex reads the root AGENTS.md and skills from .agents/skills/
• Antigravity reads the root AGENTS.md, skills from .agents/skills/, and /name workflows from .agents/workflows/
• Cursor reads the root AGENTS.md and skills from .agents/skills/

The same coding conventions, skills, and procedures end up copied several times, in several formats, and every change has to be synced to every copy. On top of that you often want a personal shared library layered with a per-project one.

**agsy (agent-sync)** solves exactly this:

It **merges** instruction files from multiple sources into a single build output directory (`.agsy/` by default), **converts** them into each tool's native shape, then **mounts** the output into each tool's read location via links.

Edit only the sources, run `agsy apply` once, and every tool updates at the same time.

Know these terms first

The rest of the documentation uses these terms throughout.

Term	Meaning
source	An original instruction library you maintain (the sources array); there can be several
output