Claude Code slash command
Deepdive
Deepdive: Deep iterative thinking slash command for Claude Code - enables multi-round exploratory reasoning and non-linear problem-solving
Install
git clone https://github.com/wanpengxie/deepdive.gitFrom the README
Deepdive 中文 | English 让 Claude Code 真正思考的 slash command (而不是"看起来在思考") A Slash Command Fix 👇 快速开始 工作原理 为什么这个方法有效? 几个简单但强大的想法: 1. thinking过程转化为显式的todo list,而不是一次思考就结束,好处是 * 尽量让claude自己梳理问题背景和边界,而不是消耗用户的心智 * 在梳理清晰的基础上,输出todo或者ask user,可以细粒度的跟用户对齐方向和预期,并且精准的控制claude的行为 * 单次output的长度是有限的,限制了claude单次thinking的长度(即使是ultrathink),而转化为agent+todo list执行理论上可以允许模型无限长度、无限深度的思考(即使context消耗完了,也可以compact结论到文档里,继续执行todolist) 3. 用元策略显式的指导claude如何进行非线性思考和迭代 4. 避免过于强调目标导向带来的偏差:允许模型坦诚的承认自己的置信度,对用户更透明 1. 复用TODO tools到思考过程 传统TODO(任务清单): Deep-iterate TODO(思考步骤): 为什么有效? - 强制分步(不能一次性写完) - 复用Claude Code内置的多步推理能力 2. 非线性思考 真实的思考不是: 而是: Deep-iterate允许困惑、允许推翻、允许不完美。 因为真实的创造性思考本来就是混乱的。 适合场景 - ✅ 架构设计(没有标准答案) - ✅ 技术选型(需要权衡) - ✅ 产品方向(需要理解本质) - ✅ 深度分析 - ✅ 持续讨论问题 - ❌ 直接的技术执行:bug修复、写mvp等等 - ❌ 纯信息查询(不需要设计决策) 背景信息 command的动机就是上图,自从6月份claude code降职之后,我每天有60%以上的时间在跟claude code对骂 经过仔细观察,我发现 问题不是AI变笨了 Claude没变笨,是response的模式改变了。 典型场景: 看起来专业?但可能这个项目的用户只有100人。 Claude没问: - 业务目标是什么? - 有多少用户?多少内容? - 有历史数据吗? 它只是在表演回答,而非真实思考。 这就是"QA模式"的陷阱: - 接受问题框架(不质疑前提) - 过早收敛(为了看起来专业) - 被query带偏(不探索本质) Deepdive:让Claude学会慢思考 这是一个slash command。但它改变的不是功能,是思考模式。 Claude会: 第零步:理解问题 探索A:从业务阶段理解 探索B:从用户行为理解 反例验证 诚实的结论 对比可见:后者在理解问题而非快速回答。 贡献 Issues和PRs欢迎。 这是个实验性框架,远不完美。 License MIT Author: wanpeng.xie@gmail.com Created using /deepdive _理解问题 → 真实探索 → 深入本质 → 诚实总结_ _Real thinking, not performative iteration._
How custom slash commands work
- A slash command is a Markdown file. Put it in ~/.claude/commands (for you) or .claude/commands (for a project), and it becomes /file-name inside Claude Code. Claude Code now treats commands and skills the same way, so .claude/skills/name/SKILL.md also works.
- Arguments you type after the command are passed in as $ARGUMENTS.
- If the repo is a plugin marketplace, installing it from /plugin adds the commands for you.
Details
- Repository
- wanpengxie/deepdive
- Type
- Slash commands
- License
- GPL-3.0
- Last updated
- 9 November 2025
- Topics
- claudeclaude-codeclaude-code-commandsdeep-thinkingiterative-reasoning
FAQ
What is Deepdive?
Deepdive: Deep iterative thinking slash command for Claude Code - enables multi-round exploratory reasoning and non-linear problem-solving It is a slash command for Claude Code, from wanpengxie/deepdive on GitHub.
How do I install Deepdive?
Clone wanpengxie/deepdive from GitHub and follow the setup steps in its README.
Is Deepdive popular?
wanpengxie/deepdive has 51 GitHub stars and 3 forks. It was last updated on 9 November 2025.
More slash commands
claude-code-commands
30rlajous/claude-code-commands
Git workflow commands, hooks and agents for reviews, releases and notifications.
My Claude Code Workflow
44yaukwan/my-claude-code-workflow
Contains a complete workflow, sub-agent and command line
Claude Code Linux Desktop Slash Commands
30danielrosehill/Claude-Code-Linux-Desktop-Slash-Commands
Slash commands for Linux desktop sysadmin with Claude Code (or any other agent)
SourceAtlas
48lis186/SourceAtlas
A set of AI-powered slash commands for Claude Code and OpenSkills (support Cursor, Windsurf and Gemini CLI) that help you understand any codebase quickly
Claude Code Seo
46mxfschr/claude-code-seo
An executable SEO system for Claude Code: 26-section knowledge base + 9 slash commands running live audits against GSC and DataForSEO
BHIL HELM Sci Fi Visualizer
41PolymathWizard/BHIL-HELM-Sci-Fi-Visualizer
Sci-fi dashboard framework for Claude Code. Profile data, match a design language, build a single-file HTML mockup with real bindings, QA it against WCAG AA and an IP checklist, gate it SHIP/HOLD/REWORK. 15 token sets, 4 subagents, 6 slash commands, stdlib-only validators. MIT + CC BY 4.0.
Browse the most popular Claude Code mods, all slash commands, or the skills and MCP server directories.