ARES — AI Repository Standard (ARS)

Portable repository knowledge for AI coding tools. One .ai/ source → Cursor, Copilot, Claude Code, Codex, Antigravity.

Topics: ai · ai-coding · repository-standard · cursor · github-copilot · claude-code · codex · cli · developer-tools · standards · portability · golang · dot-ai · open-source · specification

Latest release: Download latest — binaries for macOS, Linux, and Windows

ARES is the reference implementation of ARS, the AI Repository Standard.

It lets a repository define durable AI coding knowledge once in .ai/, then generate provider-specific files for Cursor, GitHub Copilot, Claude Code, OpenAI Codex, and Antigravity.

The golden rule: delete generated provider files, run ars compose, and everything comes back from .ai/.

When to use ARES

Write repository knowledge once in .ai/. Let ars compose and ars import keep every AI tool in sync.

flowchart LR
  subgraph sources["Where knowledge lives today"]
    GH[".github/copilot-instructions.md"]
    CU[".cursor/rules/"]
    CL["CLAUDE.md"]
    CX["AGENTS.md"]
  end

  AI[".ai/ canonical source"]

  AG[".antigravity/"]

  subgraph outputs["Generated provider artifacts"]
    GH2["Copilot"]
    CU2["Cursor"]
    CL2["Claude Code"]
    CX2["Codex"]
    AG2["Antigravity"]
  end

  GH -->|ars import github| AI
  CU -->|ars import cursor| AI
  CL -->|ars import claude| AI
  CX -->|ars import codex| AI
  AG -->|ars import antigravity| AI

  AI -->|ars compose --target copilot| GH2
  AI -->|ars compose --target cursor| CU2
  AI -->|ars compose --target claude| CL2
  AI -->|ars compose --target codex| CX2
  AI -->|ars compose --target antigravity| AG2

Loading

Scenarios

1. You started with GitHub Copilot and want to move to Cursor

Your team already invested in .github/copilot-instructions.md. Import it into .ai/, edit there, then compose Cursor artifacts. No rewrite from scratch.

ars import github
# refine .ai/instructions/, .ai/skills/, etc.
ars validate
ars compose --target cursor

2. Your team uses all four providers

Different engineers use different tools. Maintain one .ai/ tree and compose each target when knowledge changes.

ars validate
ars compose --target cursor
ars compose --target copilot
ars compose --target claude
ars compose --target codex

3. You are starting a new repository

Scaffold .ai/ on day one so provider files are always generated, never hand-maintained.

ars init
# add instructions, agents, skills, prompts under .ai/
ars validate
ars compose --target cursor   # or whichever tool you use first

4. Knowledge is scattered across provider-owned files

Rules live in .cursor/, instructions in .github/, and CLAUDE.md drifted months ago. Import each source into .ai/, reconcile conflicts, then compose everything from one place.

ars import cursor
ars import github --overwrite   # only when you intend to replace existing .ai/ files
ars import claude
ars validate

5. You want CI to catch drift before merge

Validate structure and references in CI. Optionally compose artifacts so generated files stay in sync with .ai/.

ars validate --json
ars compose --target cursor

6. Someone deleted generated provider files

That is expected. Generated artifacts are disposable; .ai/ is the source of truth.

rm -rf .cursor/rules .github/copilot-instructions.md CLAUDE.md AGENTS.md
ars compose --target cursor
ars compose --target copilot
ars compose --target claude
ars compose --target codex

Installation

macOS and Linux (one-line installer)

No Go required:

curl -fsSL https://github.com/okfriansyah-moh/ares/raw/main/install.sh | bash

Then add to PATH if prompted:

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc && source ~/.zshrc

Windows

The one-line installer is bash-only. On native Windows, use one of the options below.

Download a release binary (no Go required):

1. Open GitHub Releases.
2. Download ars-windows-amd64.exe.
3. Put it on your PATH (for example %USERPROFILE%\.local\bin\ars.exe).

PowerShell example:

$installDir = "$env:USERPROFILE\.local\bin"
New-Item -ItemType Directory -Force -Path $installDir | Out-Null
Invoke-WebRequest `
  -Uri "https://github.com/okfriansyah-moh/ares/releases/latest/download/ars-windows-amd64.exe" `
  -OutFile "$installDir\ars.exe"

Then add %USERPROFILE%\.local\bin to your user PATH and verify:

ars --version

WSL: if you use Windows Subsystem for Linux, run the macOS/Linux one-line installer inside your WSL shell.

Go Install

Works on macOS, Linux, and Windows when Go is installed:

go install github.com/okfriansyah-moh/ares/cmd/ars@latest

Docker

docker run --rm -v "$(pwd):/repo" ghcr.io/okfriansyah-moh/ares:latest compose --target cursor --root /repo

On Windows, use Docker Desktop and mount your repo path in the same way (for example -v "%cd%:/repo" in PowerShell).

Homebrew

Coming soon:

brew install ars-standard/tap/ars

Quick Start

ars init

# Edit canonical repository knowledge:
# .ai/manifest.yaml
# .ai/instructions/
# .ai/agents/
# .ai/skills/
# .ai/prompts/

ars validate
ars compose --target cursor

Command Reference

Command	Description
ars init [--root <path>] [--force]	Scaffold .ai/.
ars validate [--root <path>] [--json]	Validate .ai/ structure and references.
ars compose --target <target> [--root <path>]	Generate provider artifacts from .ai/.
ars import <source> [--root <path>] [--overwrite]	Import provider artifacts into .ai/.

ars compose --target codex creates AGENTS.md only when it is missing. If AGENTS.md already exists at the repository root, ARES validates the path and preserves the file.

Provider Support

Provider	Compose target	Output
Cursor	cursor	.cursor/rules/, .cursor/prompts/
GitHub Copilot	copilot	.github/copilot-instructions.md
Claude Code	claude	CLAUDE.md
OpenAI Codex	codex	AGENTS.md
Antigravity	antigravity	.antigravity/

Provider artifact	Import source
.github/copilot-instructions.md	github
.cursor/rules/*.mdc	cursor
CLAUDE.md	claude
AGENTS.md	codex
.antigravity/	antigravity

Architecture

.ai/
  manifest.yaml
  instructions/
  agents/
  skills/
  prompts/
      |
      v
   ars compose
      |
      +--> .cursor/
      +--> .github/copilot-instructions.md
      +--> CLAUDE.md
      +--> AGENTS.md
      +--> .antigravity/

flowchart TB
  subgraph ai[".ai/ — canonical source"]
    M[manifest.yaml]
    I[instructions/]
    A[agents/]
    S[skills/]
    P[prompts/]
  end

  CLI[ars compose]

  subgraph providers["Provider artifacts — generated, disposable"]
    CUR[.cursor/]
    COP[.github/copilot-instructions.md]
    CLAUDE[CLAUDE.md]
    AGENTS[AGENTS.md]
    ANTI[.antigravity/]
  end

  M --> CLI
  I --> CLI
  A --> CLI
  S --> CLI
  P --> CLI

  CLI --> CUR
  CLI --> COP
  CLI --> CLAUDE
  CLI --> AGENTS
  CLI --> ANTI

Loading

ARES is a local, file-based CLI. It is not an agent runtime, provider router, workflow engine, memory system, database-backed app, web app, or marketplace.

Repository Format

.ai/
  manifest.yaml                 project metadata
  instructions/<name>.md         repository-wide instructions
  agents/<name>/AGENT.md         agent role, responsibilities, uses, boundaries
  skills/<name>/SKILL.md         reusable knowledge
  prompts/<name>.md              reusable prompt templates

See SPEC.md for the full ARS v1 specification.

Contributing

Read SPEC.md, docs/architecture.md, and docs/PLAN.md before changing behavior. Keep .ai/ as the canonical source of repository knowledge and provider files as generated artifacts.