Big picture in four diagrams
Claude Code is an AI that reads files, edits files and runs commands. This workflow is its manual, its tools and its seat belt.
1. What it is made of
2. How it runs
3. How you work with the AI (req / reply)
4. What it can do
The little words next to the spinner
While Claude Code is thinking or working, a random English word rotates next to the spinner: Thinking, Infusing (to pour something into something), Pondering, Cooking and so on. They are only decorative status words. They do not mean it is doing anything specific; to see what it really does, look at the tool calls shown below (reading files, editing files, running commands).
Hover for motion; the left menu has the details of every part.
AI Dev Workflow Guide
More than documentation: this is a starter workflow kit you can really install into your project: a CLAUDE.md template, 3 Rules, 3 commands, 5 Skills, a permission example and the req/reply communication method. Every piece is shown in full; you can copy it or install it in one step.
Three steps to use it
- Download and install: go to the One-step install page and copy the command (Windows PowerShell or Mac/Linux bash). The script never overwrites any file you already have.
- Fill in your project: open the generated
CLAUDE.mdtemplate and replace the [bracketed] parts with your project. - Try a command: open Claude Code and type
/status; then try/estimate your request.
What is in the kit
| Type | Guest sees (full content) | Extra after sign-in | Where |
|---|---|---|---|
| Rules | 3 | +5 | Rules |
| Commands | 3 | +4 | Commands |
| Skills | 5 | +2 | Skills |
| settings.json permission example | 1 | - | Permissions |
| req/reply communication | rule + template | - | req/reply |
| Install script + zip | lite kit | full kit zip | Install |
The exact numbers are on the full inventory page, counted from the real files by the generator.
First, understand these concepts
What this is
Claude Code is Anthropic's command-line AI coding assistant, and it also has a VS Code extension. By itself it can only "read files, edit files, run commands". What turns it into a dependable partner is a handful of things you give it:
| Thing | What it does | In one line |
|---|---|---|
| CLAUDE.md / Rule | Rules read at the start of every conversation | Tells it "the house rules of this project" |
| Memory | Experience it writes down itself | Remembers your preferences and past pitfalls across conversations |
| Skill | A set of steps loaded on demand | Read only when the matching task comes up, so it does not use everyday context |
| Command | A slash command you trigger by hand | Saves a routine you often run as /name |
| Permission | Which actions need no question | The balance between safety and speed |
| MCP | External tools (browser, database, ...) | Lets it "see" web pages and "click" buttons |
| Sub-agent / loop | Clones that work in parallel, scheduled repetition | Split big jobs, check periodically |