sweetclaude

SweetClaude

TL;DR

Start Claude Code and run:

/plugin marketplace add carson-sweet/sweetclaude@main
/plugin install sweetclaude@sweetclaude-stable
/sweetclaude:help

What It Is

A software development partner for the full project lifecycle — from the first idea through design, implementation, testing, and ship. SweetClaude teaches itself about your project and preferences through conversation, then applies those learnings to deal with the less-fun parts of the project.

SweetClaude works well when structure, traceability, and product thinking is valued over raw coding speed. It's the wrong tool if you just want to go faster — we suggest GStack if speed is your primary goal. It was built for software development, but has also been used successfully for academic research, product marketing strategy, and other knowledge-intensive work.

EXPERIMENTAL: If you want fast and disciplined, there is the experimental (and therefore potentially unpredictable) John Wick mode.

Requires Claude Code and an Anthropic subscription. We recommend using a model designed for day-to-day work (e.g. Sonnet 4.6 at the time of writing) although you can take it up to top-tier frontier models — just be mindful that models like Opus burn through your Anthropic usage way faster. Our honest experience is that Opus is overkill for most anything except perhaps initial product discovery and brainstorming, or if Sonnet gets stuck on a tough problem. You can always switch models mid-stream with the Claude Code command /model .

Install SweetClaude (below) then run /sweetclaude:help to get a guided walk-through of what it is, how it works, and whether it's right for your project before touching a single file.

Key Design Principles

There's a massive number of Claude Code plugins out there. To help you decide if SweetClaude might be one that's helpful, here are the top principles used when designing SweetClaude:

• Successful projects take more than code. If you're here you probably already know this, but really thinking through what you're building, for who, why, and how they'll use it results in a good product that delivers value to users. Unfortunately, that part is a lot less fun than vibe-coding — it's real work. SweetClaude is primarily about taking care of that work — and a bunch of other lessfun work — in a structured, reliable way.
• Hide the complexity. SweetClaude has over 100 skills built natively on Claude Code's Skills framework and Anthropic's multi-agent architecture. These skills compose into dynamic, situation-driven workflows across four operating modes. Few people want to learn all that. To hide the complexity, SweetClaude uses a conversational interface with just a few commands as entry points, and an orchestrator skill does the rest (you can also manually invoke most skills if you want).
• Adapt to user preferences as they evolve. AI assisted vibe-coding is fantastic for rapid experimentation and iteration to see if an idea is viable. An assistant framework should be able to capture things passively without killing the vibe and then be able to use what it captured to dial up the structure and discipline as needed, based on what it learned in the vibe phase. The framework adjusts to the user's preferences — not the other way around — and is designed to dial up or dial back the level of structure mid-project as needed.

Quick Start

Start Claude Code in your project folder as you normally would, then inside Claude Code run these commands:

/plugin marketplace add carson-sweet/sweetclaude@main
/plugin install sweetclaude@sweetclaude-stable
/sweetclaude:help

Basic Usage

Inside Claude Code, the primary command is /sweetclaude:go which keeps things simple. Use natural language after the command to drive things. The /sweetclaude:go command acts as a primary interface (sometimes the only command needed) to SweetClaude's work state, skills, and workflow orchestration, so you don't have to know every skill command — which would be madness.