claude-starter-kit

Turn Claude into an AI assistant that actually remembers your work.

The Problem · Quick Start · What You Get · Skills · FAQ

---

The Problem

If you use Claude or ChatGPT in your browser, you know this feeling:

• The Amnesia Loop — Every new chat starts from scratch. You re-explain your project, your brand voice, your team, what you did yesterday. Every. Single. Time.
• The Context Wall — You hit "conversation too long" right when you're making progress. Everything disappears.
• The Copy-Paste Dance — The AI can't see your files. You manually feed it documents, screenshots, spreadsheets. Over and over.

This kit fixes all of that.

It gives Claude a permanent memory, file access, and reusable skills — so every conversation picks up exactly where the last one left off.

What Changes

	Regular Chat 🌐	With This Kit 🤖
Memory	Forgets everything when you close the tab	Remembers your project details across sessions
Your Files	You copy-paste documents into chat	Claude reads and edits your files directly
Consistency	Drifts off-topic, ignores past instructions	Follows your rules and preferences every time
Skills	Does whatever the base model can do	You can teach it new abilities (like asking Google AI for help)

---

3 Steps to Start

You don't need to know how to code. The whole setup takes about 5 minutes.

What you need first:

1. Claude Code installed (Anthropic's CLI tool — follow their install guide)
2. A Claude Max/Pro subscription, or an Anthropic API key

Then:

# 1. Download and rename this kit
git clone https://github.com/awrshift/claude-starter-kit.git my-project
cd my-project

# 2. Launch Claude Code
claude

# 3. That's it — Claude sets everything up automatically

New to Terminal? Terminal is a text-based app on every computer. On Mac: search "Terminal" in Spotlight. On Windows: search "Command Prompt". You just type commands and press Enter.

Tip: If you need to navigate to a folder, type cd (with a space), then drag the folder from Finder/Explorer into the Terminal window. Press Enter.

On first launch, Claude will:

• Ask about your project (name, description, language)
• Set up memory and context files
• Offer to install the Gemini skill (optional — needs a free Google API key)
• Clean up demo files and greet you

No manual configuration needed. Just talk to Claude.

---

What You Get

Think of this kit as a fully furnished office for your AI assistant:

🧠 Permanent Memory

Claude keeps a notebook (.claude/memory/) that it reads every session. Your project details, decisions, preferences — all remembered automatically. You never re-explain anything.

🛠️ Built-in Skills

Four pre-installed "superpowers" that extend what Claude can do. Get second opinions from Google AI, run structured brainstorms, make complex decisions with a framework. All included in .claude/skills/.

📋 Project Tracking

Each project gets a Journal (projects/your-project/JOURNAL.md) — a single file where Claude tracks tasks, decisions, and progress. You can have multiple projects running at once.

🔬 Experiment System

Trying different approaches? The experiments/ folder is built for structured research — define a question, explore options, compare results, make a decision.

💾 Auto-Save

Before Claude's conversation gets long and compresses, a built-in safety net reminds it to save all progress. Nothing gets lost.

---

Superpowers

Four skills are included right in the project (.claude/skills/). Each one gives Claude a new ability. You don't need all of them — start with what makes sense for your work.

Gemini — Get a Second Opinion

"Two AI brains are better than one — especially when they think differently."

Claude asks Google's Gemini AI to review its own work. Different AI = different blind spots caught.

Real scenario: You ask Claude to write a marketing strategy. Before showing it to you, Claude sends it to Gemini for critique. You get the best of both worlds.

How to use: Say "ask Gemini", "second opinion", or "check with Gemini"

Setup: pip install google-genai + a free Google API key

---

Brainstorm — Two AIs Debate Your Idea

"When Claude and Gemini disagree, that's where the best insights hide."

A structured 3-round dialogue where Claude and Gemini challenge each other. Diverge → Deepen → Converge. Ends with one clear recommendation.

Real scenario: You're choosing between three campaign approaches. Claude and Gemini debate the pros and cons in three rounds, then present you with a winner.

How to use: Say "brainstorm", "let's think through options", or "explore this with Gemini"

---

AWRSHIFT — Think Before You Build

"The framework that stops you from building the wrong thing."

A step-by-step decision process: define the problem → research options → set success metrics → test → decide. Perfect for complex choices with unknowns.

Real scenario: Your team can't agree on which tool to use for project management. AWRSHIFT guides you through structured evaluation — not gut feeling, but evidence.

How to use: Say "awrshift", "let's research this", or "think this through"

---

Skill Creator — Teach Claude New Tricks

"Turn any repeatable workflow into a reusable command."

If you find yourself giving Claude the same instructions every week (e.g., "write my Friday report in this format"), turn it into a permanent skill.

How to use: Say "create a skill" or "turn this into a skill"

---

Skills at a Glance

Skill	What it does	Needs setup?
Gemini	Second opinion from Google AI	Free API key
Brainstorm	Two AIs debate, one answer	Needs Gemini
AWRSHIFT	Step-by-step decision framework	No
Skill Creator	Build your own skills	No

---

What People Build With This

Because Claude can read, write, and remember — non-technical professionals are building things like:

1. Content Strategist — Claude remembers a 20-page brand guideline. It drafts blog posts, saves them to a "Drafts" folder, and updates a content calendar — all automatically.

2. HR Specialist — Claude acts as an interview prep assistant. It knows the company values, generates custom questions based on a resume, and saves evaluation notes.

3. Project Manager — Claude maintains risk logs and status reports. Every Friday, the PM talks through the week and Claude updates all project documents.

4. Marketer — Claude tracks A/B test experiments. It remembers which ad copy worked last month and uses that data to write next month's campaigns.

5. Course Creator — Claude organizes entire curriculums. When you change Module 2, it automatically knows to update references in Module 5.

---

How It Actually Works

Your Agent's Anatomy

Every Claude agent built with this kit has 6 core components working together:

Component	What it is	Simple analogy
Brain	CLAUDE.md	Your agent's job description
Memory	.claude/memory/	A notebook it reads every morning
Rules	.claude/rules/	House rules posted on the wall
Skills	.claude/skills/	Special abilities in a toolbox
Journal	projects/X/JOURNAL.md	A to-do list for each project
Context Hub	context/next-session-prompt.md	A sticky note: "pick up here tomorrow"

What Loads When

Not everything loads every time. The kit uses layers — always-on at the bottom, on-demand at the top:

Layer	What loads	When
Always	Brain + Rules + Memory index	Every session
Start	"What to do next" prompt	Session start
Project	Journal for that project	When you work on it
Topics	Detailed knowledge files	Only when relevant
Reference	Experiments, docs	Only when needed

Session Lifecycle

Every session follows the same cycle — start, work, save. Context is preserved automatically:

A built-in safety net (pre-compact hook) makes sure progress is saved even when a long conversation gets compressed.

---

FAQ

What is "Terminal"? Is it safe?

Terminal is a text-based way to talk to your computer. Every Mac and Windows PC has one built in. It's completely safe — you're just typing commands instead of clicking buttons. Think of it as texting your computer.

Do I need to know how to code?

No. After the 5-minute setup, you communicate entirely in plain English. You can say things like: "Read the marketing plan and write three emails based on it, then save them in my emails folder."

Is my data private?

Yes. Claude Code runs on your computer and talks to Anthropic's API, which by default does not train models on your data. This is significantly more private than free web chat tools.

Does this cost money?

Claude Code requires either a Claude Max/Pro subscription or Anthropic API credits. The Gemini skill needs a free Google API key. The kit itself is completely free and open source.

What's the difference between this and regular Claude?

Regular Claude lives in a browser, resets every chat, and can't see your files. Claude Code (with this kit) lives on your computer, remembers everything across sessions, creates and edits files, and follows your custom rules.

Can I use this without the Gemini skill?

Absolutely. All skills are optional. The core system (memory, context, journals, rules) works perfectly on its own.

Can I add my own skills?

Yes. Create a folder in .claude/skills/your-skill/ with a SKILL.md file. Or just say "create a skill" and Claude will build one for you using the Skill Creator.

I messed something up. How do I start over?

Close Terminal. Delete the project folder. Download the kit again and start fresh. Your data is just files on your computer — nothing is locked.

---

Credits

Built by Serhii Kravchenko (@pmserhii) — based on 1000+ sessions of iterative refinement with Claude Code, building AI content pipelines, multi-agent systems, and GEO optimization tools.

License

MIT — use it, modify it, share it.