Stop Maintaining 20 Config Files for Your AI Coding Assistants

#javascript #productivity #ai #opensource

I use Claude Code. My teammate uses Cursor. Another swears by Copilot. Our contractor just started using Windsurf.

Every time we add a new library or establish a coding convention, someone has to update CLAUDE.md, .cursor/rules/, AGENTS.md, and whatever else everyone's using. It's a mess.

So I built Agent Hints — a registry that lets you write hints once and deploy them to 20+ AI coding assistants.

The Problem

AI coding assistants are great, but they need context about your project. Most tools have their own config format:

Claude Code reads CLAUDE.md
Cursor reads .cursor/rules/*.mdc
Copilot reads AGENTS.md
Windsurf reads .windsurf/rules/*.md
...and so on

If you're using multiple tools (or your team is), you're maintaining the same information in multiple places. And if you want to share useful hints with the community? Good luck getting everyone to manually copy your conventions into their specific format.

The Solution

Agent Hints is a CLI + registry that handles the translation for you:

npx agenthints add typescript-strict --agent claude
# Creates/updates CLAUDE.md with TypeScript conventions

npx agenthints add typescript-strict --agent cursor
# Creates .cursor/rules/typescript-strict.mdc with the same content

Same hint, different output formats. The registry handles the transformation.

Quick Start

# Interactive setup (recommended for new projects)
npx agenthints init

# Or add hints directly
npx agenthints add ultracite firebase --agent claude

The init command walks you through selecting your AI agent and browsing available hints:

$ npx agenthints init

◆ Select your AI agent
│ ○ claude - Anthropic Claude Code CLI
│ ○ cursor - Cursor AI IDE
│ ○ copilot - GitHub Copilot
│ ○ windsurf - Codeium's Windsurf IDE
│ ... (20+ options)

How It Works

Every hint in the registry has two parts:

meta.json — metadata like name, description, category, tags
hint.md — the actual content

When you run agenthints add, the CLI:

Fetches the hint from agenthints.tech
Transforms it for your target agent (adding frontmatter, headers, etc.)
Writes it to the correct location

For "append mode" agents like Claude, hints get wrapped in markers so they can be updated or removed later:

<!-- agenthints:typescript-strict:start -->
[hint content here]
<!-- agenthints:typescript-strict:end -->

For "per-resource" agents like Cursor, each hint becomes its own file with the appropriate frontmatter.

Supported Agents

The CLI currently supports 20+ AI coding assistants:

Single-file agents: Claude, Copilot, Codex, Gemini, Aider, AMP, Devin, Goose, Jules, Junie, Replit, Warp, Zed, and more.

Per-file agents: Cursor, Windsurf, Cline, Augment, Kilo Code, Kiro, OpenHands, Roo Code.

# See all supported agents
npx agenthints agents

Contributing Hints

The registry is open source. If you've got useful hints for a library or framework, you can contribute them:

# Clone the repo
git clone https://github.com/bensabic/agenthints.git
cd agenthints

# Use the interactive wizard
pnpm create-hint

The wizard prompts you for name, description, category, and tags, then creates the file structure. Write your hint in markdown, validate it with pnpm validate-hints your-hint, and open a PR.

Your hint becomes available to everyone using any supported AI tool.

Why I Built This

I got tired of copy-pasting the same conventions between config files. I also noticed the AI tooling ecosystem fragmenting — everyone's building their own instructions format, and there's no interoperability.

Agent Hints is my attempt at a portable, community-driven solution. Write hints once, use them everywhere.