claude-md-starter

作者 sumit-ai-ml已验证

A CLAUDE.md built on Karpathy's principles and gstack, adding a verification harness and autonomous-loop rules for Claude Code.

88
Stars
26
Forks
2026/8/24
添加时间

⚠️ 第三方软件声明

本 Skill 为第三方开源软件,独立托管于 GitHub。SkillTip 仅为信息目录,不控制或维护底层仓库。所显示的安全检查为自动化且范围有限,安装前请自行审查源码。

阅读服务条款

安装

添加到你的 Claude Code skills 目录:

# Add to your Claude Code skills
git clone https://github.com/sumit-ai-ml/claude-md-starter

快速入门

使用 claude-md-starter 等 Skills 的指南。

安全报告

已验证

上次扫描:—

{
  "status": "PASSED",
  "issues": []
}

README.md

claude-md-starter

An opinionated Claude Code setup, distributed as a reusable CLAUDE.md.

This repo is not an application. It is a single file worth copying: CLAUDE.md, a layered set of behavioral rules that make Claude Code plan before it codes, keep changes small, verify against a real definition of "done", and run autonomously only when the work is programmatically checkable. There is no build step and no runtime dependencies. It stands on two excellent open-source projects and adds project-specific harness and autonomy rules on top.

What's inside CLAUDE.md

CLAUDE.md lives at the repo root, so Claude Code reads it automatically when you start a session here. It is organized in layers, and each layer has a clear origin:

SectionWhat it doesOrigin
1. Think Before CodingSurface assumptions, ask, do not silently pickandrej-karpathy-skills
2. Simplicity FirstMinimum code, no speculative abstractionsandrej-karpathy-skills
3. Surgical ChangesTouch only what the task requiresandrej-karpathy-skills
4. Goal-Driven ExecutionTurn tasks into verifiable goalsandrej-karpathy-skills
5. HarnessThe exact commands that define "done"this repo (concept: harness engineering)
6. Loops & AutonomyWhen and how to run /goal, /loop, /batchthis repo (concept: Ralph loop)
TextHouse style for human-readable outputthis repo
gstack skillsRole-based slash commands (/office-hours, /ship, /review, ...)gstack

Sections 1 to 4 are based on the four principles from andrej-karpathy-skills, themselves derived from Andrej Karpathy's public observations on where LLMs go wrong when coding. The slash commands come from gstack. Sections 5, 6, and Text are original to this repo, though Harness and Loops & Autonomy build on the harness-engineering and Ralph-loop ideas credited below.

Requirements

Quick start

Three steps. The first two set it up; the third makes it yours.

Step 1: Install gstack. Open Claude Code and paste install gstack. That gives you the workflow slash commands (/office-hours, /ship, /review, and the rest). For the exact install command, see the gstack repo.

Step 2: Add the CLAUDE.md. Copy the CLAUDE.md from this repo into your project's root folder. Claude Code reads it automatically when you start a session there.

Step 3: Make it yours. Edit two sections in that CLAUDE.md:

  • Project: what you are working on, how to run it, where things live.
  • 5. Harness: what you want Claude Code to follow strictly, such as your test, lint, and build commands (or "renders in a browser and the HTML is valid" if there is no tooling yet).

Sections 1 to 4 (the Karpathy principles) and Section 6 (Loops & Autonomy) work as-is, with nothing to configure.

Using it

  • Day to day, just code. The principles steer Claude toward small, verified changes and away from silent assumptions.
  • For mechanical, programmatically verifiable work, prefer the built-in autonomy commands over typing "continue": /goal <end state>, /loop <interval, or until: condition>, and /batch <one change across many files>. See Section 6 for the rules.
  • For the role-based workflow (product interrogation, plan review, code review, shipping, QA), use the gstack slash commands. A good first move on any new idea is /office-hours.

Credits

This project is a remix and would not exist without the work below.

Embedded or required

  • andrej-karpathy-skills by multica-ai (repo) for the four-principle CLAUDE.md that Sections 1 to 4 are based on. Derived from Andrej Karpathy (@karpathy) and his observations on LLM coding pitfalls.
  • gstack by Garry Tan (@garrytan, repo) for the role-based workflow slash commands.

Concepts behind the custom sections

  • Section 5, Harness draws on harness engineering: the early-2026 framing that an agent is the model plus everything around it (Agent = Model + Harness), made reliable through deterministic checks and feedback loops rather than prompts. Attribution is contested. Mitchell Hashimoto is most often credited with naming and popularizing it, and Viv Trivedy ("Anatomy of an Agent Harness") with coining it and the Model + Harness formula. The underlying test-harness idea is older.
  • Section 6, Loops & Autonomy draws on the Ralph Wiggum loop by Geoffrey Huntley (mid-2025): a loop that re-runs an agent until it meets your exit conditions, with progress living in git rather than the context window. The general definition of an agent as something that "runs tools in a loop to achieve a goal" is Simon Willison's.

Please respect the licenses of the embedded or required projects when redistributing.

License

The custom sections (Harness, Loops & Autonomy, Text, Project) are original to this repo and released under the MIT License. The embedded principles and the referenced gstack commands belong to their respective authors above; check their licenses before relicensing any copied portions.


Built by Sumit Pandey (@sumit-ai-ml).

常见问题

What is claude-md-starter?

claude-md-starter is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by sumit-ai-ml. A CLAUDE.md built on Karpathy's principles and gstack, adding a verification harness and autonomous-loop rules for Claude Code. It has 88 GitHub stars.

Is claude-md-starter safe to use?

Yes. claude-md-starter passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.

How do I install claude-md-starter?

Clone the repository with "git clone https://github.com/sumit-ai-ml/claude-md-starter" and add it to your Claude Code skills directory (see the Installation section above).

Are there alternatives to claude-md-starter?

Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh claude-md-starter against similar tools.

评论 (0)

暂无评论,成为第一个分享想法的人!

ECC

by affaan-m

10

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

242,21936,702JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情
15

An agentic skills framework & software development methodology that works.

234,96620,863Shell
AI 智能体ai-agentsbrainstorming
查看详情

hermes-agent

by NousResearch

10

The agent that grows with you

234,43747,175Python
AI 智能体ai-agentsagent-orchestration
查看详情

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

185,94028,768JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情

cc-switch

by farion1231

3

A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io

128,8688,826Rust
AI 智能体claude-codeai-tools
查看详情

claude-code

by anthropics

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

120,03119,897Shell
AI 智能体
查看详情

开发者还喜欢

基于喜欢此 Skill 的开发者投票和收藏

ECC

by affaan-m

10

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

242,21936,702JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情
15

An agentic skills framework & software development methodology that works.

234,96620,863Shell
AI 智能体ai-agentsbrainstorming
查看详情

hermes-agent

by NousResearch

10

The agent that grows with you

234,43747,175Python
AI 智能体ai-agentsagent-orchestration
查看详情

n8n

by n8n-io

12

Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.

201,88160,308TypeScript
MCP 服务器apisai-tools
查看详情

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

185,94028,768JavaScript
AI 智能体ai-agentsanthropicclaude-code
查看详情

cc-switch

by farion1231

3

A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io

128,8688,826Rust
AI 智能体claude-codeai-tools
查看详情