Claude Code Workspace workflow
A personal Claude Code setup: four loading layers, an advisor model mode, ten hooks and working commands
- 156 ★ on the repository
- Last commit
Extended access
Why this level
- The configuration overwrites the global `CLAUDE.md` and attaches hooks that run commands on every prompt
- Automatic hooks run scripts after file edits in your projects
Install
- Works in
Manual install
Save a copy of `~/.claude/CLAUDE.md`. Then clone `https://github.com/zeuikli/claude-code-workspace.git` into `~/claude-code-workspace` and write the line `@$HOME/claude-code-workspace/CLAUDE.md` into `~/.claude/CLAUDE.md`. Then run `/load-plan` and show the output. Enable hooks only with the user consent.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The configuration overwrites the global `CLAUDE.md` and attaches hooks that run commands on every prompt
- Automatic hooks run scripts after file edits in your projects
This is third-party code. Review the repository files before installing.
What it does
Claude Code Workspace is a personal Claude Code working configuration in Traditional Chinese with an English part in the README. At its center is a four-layer loading scheme backed by a `/load-plan` command with token cost estimates: at session start `CLAUDE.md` and four rules load (about 4 thousand tokens), while skills load on trigger words and spend no context before that. Models are split by role ("Advisor" mode): a Sonnet executor for daily tasks, Haiku for search across dozens of files, Opus as an advisor only for architecture, security and boundary decisions. Ten hook events cover session start, a reminder to delegate to subagents, a `/deep-review` reminder before commits, automatic syntax checks of `.sh`, `.json` and `.py` after edits and memory saving on context compaction. Commands cover development (`/debug`, `/perf`, `/map`, `/deep-review`, `/frontend-design`), marketing (`/marketing`), writing (`/writing`), research (`/research`), project management (`/pm`, `/agent-team`) and workspace upkeep (`/prime`, `/retro`, `/context-report`, `/load-plan`). Setup is a repository clone plus a `@path/CLAUDE.md` line in `~/.claude/CLAUDE.md`.
Who it is for. Claude Code users who want a ready starter configuration with token savings, models split by role and automatic checks.
Good fit when
- You want a starter Claude Code configuration where the token cost of each session is known in advance
- You want to fix which model does routine work, which searches and which decides tough architecture questions
- A commit needs a review reminder and edits need an automatic syntax check of scripts
- One working configuration must serve different projects without copying
Not a fit when
- You need a licensed product for a team: the repository has no license and the terms must be clarified with the author
- You need an English-language setup: rules and docs are mostly in Traditional Chinese
- You need a minimal configuration without hooks: the set includes ten hook events and writes a global `CLAUDE.md`
Example request
Run /load-plan and show how many tokens each configuration layer spends at session startLimitations
The README install overwrites `~/.claude/CLAUDE.md` with a reference line to the repository, and the quick start downloads the repository and recreates its folder on update, so save your own global `CLAUDE.md` first. Hooks run on every prompt and after edits, which shows in session behavior.
Commands and rules fit the author scenarios: model versions in advisor mode are hardcoded in the configuration and will go stale. Docs and prompts are in Traditional Chinese and token figures were measured by the author and depend on caching. The repository has no license file.
How to disable. Remove the `@.../CLAUDE.md` line from `~/.claude/CLAUDE.md` (restore your saved copy), remove the added hooks from `settings.json` and delete the repository clone.
FAQ
How do I install the configuration locally?
Clone the repository and write the line `@$HOME/claude-code-workspace/CLAUDE.md` into `~/.claude/CLAUDE.md`. Hooks for projects are added in `.claude/settings.json`.
What does the /load-plan command do?
Shows all available configuration tools across the four loading layers with a token cost estimate for each.
What is Advisor mode?
A split of models by role: Sonnet executes tasks, Haiku searches and Opus steps in only for architecture, security and boundary decisions.
Which hooks are enabled?
Ten event types including session start, prompt submit, before a Bash command, after file edits and context compaction. The list is in the README.
Is a license needed to use it?
The repository has no license file, so for team or commercial use clarify the terms with the author.
Discussion
No questions or reviews yet.
Related
WorkflowsA personal ~/.claude for Claude Code: twelve subagents, review commands, worktree and goal skills and a Go status line.
A Claude Code workspace folder: hooks with a local Ollama detect the project type, pick skills and remind you to update the docs
A template for Claude Code projects: parallel agent swarms, workflow skills, safety hooks and rules in one install.
Claudette runs several Claude Code sessions in parallel, each in its own git worktree, from a window or the console
Workflows for Claude Code: /do, /omo, /bmad-pilot and more commands, with Codex, Gemini or OpenCode writing the code through codeagent-wrapper
An environment for Claude Code that keeps context across sessions via hooks, ledgers and handoffs, orchestrates agents and saves tokens
README in short
The repository is set up as a personal Claude Code workspace with a bilingual README, a loading scheme, rules and a command set. The author gives an efficiency table (request caching, sparse checkout, double-load protection) and a Claude Code best-practices guide.
README badge
Original SKILL.md
---
name: add-skill
description: 建立符合官方規格的 Claude Code skill：蒐集需求、產出正確格式的 SKILL.md、驗證命名與描述規則、套用 Progressive Disclosure 結構。適合在新增 skill 或 slash command 時使用。
disable-model-invocation: true
context: fork
---
# Add Skill — 建立新 Skill
## 執行流程
### Step 1：取得最新官方規格
優先抓取官方文件，確保符合目前規格：
```
WebFetch: https://docs.anthropic.com/en/claude-code/skills
```
若失敗，用 WebSearch：`Claude Code skills SKILL.md format site:docs.anthropic.com`
### Step 2：確認基本資訊
從使用者取得（或直接從 prompt 解析）：
1. **範圍**：全域 skill 或 repo-local skill？
- **全域** → `~/.claude/skills/{skill-name}/`（所有專案可用）
- **Repo-local** → `.claude/skills/{skill-name}/`（僅此 repo，git 版控）
2. **名稱**：小寫、僅用連字號（優先動名詞形式：`processing-pdfs`、`testing-code`）
3. **用途**：做什麼？（1-2 句）
4. **觸發條件**：什麼時候用？（關鍵字、場景）
5. **複雜度**：純指令型，還是需要 scripts / references？
6. **自由度**：嚴格步驟型，還是原則指引型？
### Step 3：建立目錄結構
```
{base-path}/skills/{skill-name}/
├── SKILL.md # 必要：主指令（<500 行）
├── config.json # 可選：使用者設定（首次啟動時填寫）
├── scripts/ # 可選：給 Claude 組合用的腳本與函式庫
│ └── main.py
└── references/ # 可選：詳細文件（只能一層）
└── examples.md
```