Why

You run more than one AI CLI: Claude Code in the terminal, Cursor in the editor, whichever model leads next month. Each wants the same conventions in its own format: CLAUDE.md, AGENTS.md, .cursor/rules, GEMINI.md.

agnostic-ai keeps one source of truth in Markdown plus YAML frontmatter, aligned with AGENTS.md. sync writes what each tool expects, where it expects it. Switching tools is one command. Running four is the same command.

Install

brew install --cask Chemaclass/tap/agnostic-ai                                                      # macOS / Linux
irm https://raw.githubusercontent.com/Chemaclass/agnostic-ai/main/scripts/install.ps1 | iex         # Windows
curl -fsSL https://raw.githubusercontent.com/Chemaclass/agnostic-ai/main/scripts/install.sh | bash  # no package manager

The scripts need no Go toolchain: each fetches the archive for your OS and CPU, verifies it against checksums.txt, and puts the binary on PATH.

Inside Claude Code, skills that install and drive the CLI for you:

/plugin marketplace add Chemaclass/agnostic-ai
/plugin install agnostic-ai@chemaclass

Other routes (go install, npx, winget, Scoop, prebuilt archives) and agnostic-ai upgrade, which detects the route you used: install reference.

Quickstart

agnostic-ai init --demo   # scaffold specs, one example per kind
agnostic-ai sync          # emit native config for every target

Edit anything under .agnostic-ai/, run sync again. Already have CLAUDE.md or .cursor/rules? agnostic-ai import claude (also cursor, codex, gemini, cline, ...) turns them into specs first.

Watch mode, shell completion, CI drift gate, commit-vs-ignore: Getting started.

How it works

One rule in .agnostic-ai/rules/conventional-commits.md:

---
name: conventional-commits
description: Always use Conventional Commits format.
globs: "**/*"
alwaysApply: true
---

Use Conventional Commits. Subject under 72 chars. Body explains why, not what.

agnostic-ai sync turns it into .cursor/rules/conventional-commits.mdc:

---
description: Always use Conventional Commits format.
globs: "**/*"
alwaysApply: true
---
<!-- Generated by agnostic-ai -->

Use Conventional Commits. Subject under 72 chars. Body explains why, not what.

The same file lands in every target's own rules directory: .claude/rules/, .kiro/steering/, .trae/rules/, and the rest. Tools without one (Codex, Gemini, Aider, Junie) get the body inlined in their entry-point instead. Every root file (CLAUDE.md, AGENTS.md, GEMINI.md, CONVENTIONS.md, .github/copilot-instructions.md) carries the same short pointer back to the specs, so no tool reads a stale copy.

Supported targets

25 targets, all first-class: find your row, read what works out of the box. sync enables 20 by default. Amp, Warp, Jules, Goose, and Augment are opt-in: add them to targets: or pass -t amp,warp,jules,goose,augment.

• ✅ native — the tool's own format, at the path it auto-loads.
• ◐ bundled — folded into the target's entry-point file, no per-spec file.
• ○ opt-in — set the matching outputs.<target>.* key to materialize a file.
• - not supported — skipped with a warning (silence it with on-unsupported: silent).

Five kinds here. command, settings, review, environment, and ignore are in the full matrix, which also carries a Verify with the real CLI checklist per target. Golden-snapshot tests lock every cell.

Documentation

Get started

• Getting started: first rule, first sync, in 2 minutes
• Spec format: frontmatter reference for every kind
• Examples: drop-in templates
• Why not symlinks: versus symlinks, copies, shared files

Reference

• Targets: what each adapter emits and where
• Configuration: agnostic-ai.yaml reference
• CLI reference: every flag, every command
• Errors: error code lookup

Workflows

• CI gate: drift checks on every PR
• Git hooks: pre-commit, lefthook, husky recipes
• Packs: share spec bundles across repos

Tools

• Claude Code plugin: install, init, sync, and import as skills
• Landing page: the visual why / what / how
• Playground: paste a spec, see what every adapter emits (WASM, runs offline)
• Editor extensions: VS Code and JetBrains, both shipped

Contributing

• Architecture & roadmap: adapter pattern, adding targets, release process

Share user configuration across Claude Code and Cursor

Put user-wide instructions in ~/.agnostic-ai/AGNOSTIC_AI.md, with optional rules/, hooks/, and skills/ folders beside it. Then run this from any directory:

agnostic-ai sync --global

This mode is opt-in and separate from project sync. It writes native user configuration for 22 of the 25 targets. See global configuration.