Claude Starter Folder

A ready-to-use .claude folder structure for builders who want Claude Code to work with more context, better rules, agents, memory, and workflows.

By Arjav Jain · COO @ Hyperke, @lifeofarjav 9 min read Originally a reel on @lifeofarjav

By @lifeofarjav

A ready-to-use .claude folder structure for builders who want Claude Code to work with more context, better rules, reusable agents, project memory, workflows, and repeatable execution patterns.

This is not a random prompt pack.

This is a starter operating system for your Claude Code setup.

Download Link: https://drive.google.com/drive/folders/1yY4wdi0nS0QDf5DpgotgCNo_DI4sE0fI?usp=sharing

What this is

The .claude folder is where you organize the context Claude needs before it starts helping you build.

Most people use Claude Code like a chatbot. Then they complain that it forgets things, breaks patterns, repeats mistakes, or gives generic answers.

That is not a Claude problem.

That is a context system problem.

This starter folder helps you put your project rules, stack, agents, hooks, skills, workflows, memory, prompts, and gotchas in one place.

The goal is simple:

Make Claude act like a project-aware teammate instead of a random assistant.

What you get

This starter setup includes a clean .claude folder with files and subfolders for:

Project rules
Claude behavior instructions
Stack context
Style preferences
Known gotchas
Agents
Hooks
Skills
Commands
Workflows
Memory
Prompts
Snippets
Permissions
Scripts
Guides
Context files
Reusable examples

Use this as your starting point, then customize it for your own repo, business, product, or workflow.

Folder structure

.claude/
├── CLAUDE.md
├── rules.md
├── persona.md
├── settings.json
├── mcp.json
├── stack.md
├── style.md
├── env.md
├── gotchas.md
├── changelog.md
├── tasks.md
├── todo.md
├── api.md
├── ignore.md
├── ports.txt
├── deploy.sh
│
├── agents/
│   ├── code-reviewer.md
│   ├── debugger.md
│   └── researcher.md
│
├── hooks/
│   └── post_tool_use.sh
│
├── skills/
│   └── readme.md
│
├── commands/
│   └── readme.md
│
├── output-styles/
│   └── readme.md
│
├── memory/
│   └── readme.md
│
├── workflows/
│   └── readme.md
│
├── context/
├── guides/
├── prompts/
├── snippets/
├── scripts/
├── permissions/
├── routines/
├── plugins/
├── subagents/
├── examples/
├── teams/
└── tasks/

How to use it

Step 1: Drop the folder into your repo

Put the .claude folder at the root of your project.

your-project/
├── .claude/
├── src/
├── package.json
└── README.md

Claude can only use what it can understand.

If your repo has no clear rules, no clear stack context, and no clear workflows, Claude will guess.

And guessing is where bad AI output begins.

Step 2: Fill in the files Claude keeps needing

Start with these five files:

CLAUDE.md
rules.md
stack.md
style.md
gotchas.md

Do not try to make it perfect.

Just dump the context Claude keeps forgetting.

The first version should be messy but useful. You can clean it later.

Step 3: Add agents for repeated work

Use the agents/ folder for specialized Claude roles.

Start with:

code-reviewer.md
debugger.md
researcher.md

Each agent should define:

Name
Role
What it is for
Which tools it can use
When it should run
Rules it must follow
Expected output format

This is how you stop asking Claude to “act like a senior engineer” every single time.

You write it once.

Then reuse it.

Step 4: Add workflows for repeated execution

Use the workflows/ folder for tasks you do again and again.

Examples:

Fix a bug
Review a pull request
Launch a feature
Audit a campaign
Write a landing page
Clean a dataset
Generate content
Debug a failed build

If you do something more than twice, it should probably become a workflow.

That is the difference between using AI and building AI systems.

Step 5: Keep improving the folder

Your .claude folder should evolve with your project.

Every time Claude makes the same mistake twice, add a rule.

Every time you repeat the same instruction twice, turn it into a command, skill, or workflow.

Every time context matters, write it down.

The folder gets smarter as your project gets smarter.

File descriptions

`CLAUDE.md`

This is the main instruction file.

Use it to explain how Claude should behave inside your project.

Add:

Project purpose
Core instructions
High-level rules
How Claude should approach tasks
What Claude should avoid
How Claude should validate work
How Claude should communicate results

Think of this as the operating manual for your AI teammate.

`rules.md`

This file is for non-negotiable project rules.

Examples:

Do not change production config without approval
Do not introduce new dependencies unless necessary
Follow existing patterns before creating new ones
Keep changes small unless a refactor is requested
Always explain risky changes before making them

This file exists because vague instructions create expensive mistakes.

`persona.md`

This defines how Claude should behave.

Examples:

Senior engineer
Product-minded builder
Technical founder assistant
Brutally honest reviewer
Fast execution partner

This helps Claude adopt the right decision-making style for your project.

`stack.md`

This is where you document your actual tech stack.

Include:

Frontend framework
Backend framework
Database
Auth system
Hosting provider
APIs
Libraries
Testing framework
Design system
Deployment setup

The clearer this file is, the less time you waste explaining basic architecture.

`style.md`

This file defines your coding and communication style.

Include:

Naming conventions
Formatting preferences
Component patterns
API response style
Error handling style
Commenting rules
UI copy tone
Writing preferences

This stops Claude from producing work that technically functions but does not feel like your repo.

`env.md`

This file explains local setup and environment variables.

Include:

Required environment variables
Local setup steps
Development notes
Production notes
Service dependencies
Secret handling rules

Do not paste private keys, passwords, or sensitive tokens into this file.

`gotchas.md`

This is one of the highest ROI files.

Use it for the problems Claude would otherwise keep rediscovering.

Examples:

Broken package versions
Strange deployment behavior
Common test failures
Legacy code patterns
API quirks
Things that look wrong but are intentional

Claude cannot avoid traps it does not know exist.

Write them down.

`api.md`

Use this for API and integration notes.

Include:

Endpoint structure
Auth rules
Rate limits
Payload examples
Response formats
External service notes
Common errors

This makes Claude much better at integration work.

`tasks.md`

Use this for active work.

Examples:

Current sprint tasks
Bugs to fix
Features to build
Refactors to avoid
Open questions
Priorities

This gives Claude useful execution context.

`todo.md`

Use this for looser backlog items.

Keep it simple.

Claude should know what is active, what is later, and what should not be touched yet.

`changelog.md`

Use this to track important changes over time.

This helps Claude understand what recently changed and why.

`mcp.json`

Use this for MCP server configuration.

This is where tool servers and external context providers can be defined.

Keep this clean. A messy MCP setup creates more confusion than leverage.

`settings.json`

Use this for Claude-related configuration and preferences.

Examples:

Tool preferences
Repo-specific behavior
Execution settings
Project defaults

`ignore.md`

Use this to tell Claude what to avoid.

Examples:

Files not to edit
Folders not to touch
Generated files
Vendor files
Temporary files
Risky areas

Claude needs boundaries.

This file gives it boundaries.

`ports.txt`

Use this to document local ports and services.

Examples:

Frontend port
Backend port
Database port
Redis port
Webhook listener port
Local tunnel notes

This prevents small setup confusion from wasting time.

`deploy.sh`

Use this for deployment or deployment-related commands.

Keep it readable.

Claude should know what deploy commands exist, but it should not run risky deployment actions unless you explicitly ask.

Folder descriptions

`agents/`

This folder contains specialized agent definitions.

Example agents:

Code reviewer
Debugger
Researcher

Each agent should be specific.

Bad agent:

You are helpful.

Good agent:

You are a code reviewer. Inspect only the changed files. Prioritize correctness, security, regressions, and maintainability. Do not rewrite code unless asked. Return findings by severity.

The more precise the agent, the better the output.

`hooks/`

Hooks are for automation.

Use this for scripts that should run after certain actions, such as:

Formatting
Logging
Validation
Cleanup
Post-tool checks

Example:

post_tool_use.sh

Hooks can turn Claude from a chat assistant into part of your execution workflow.

`skills/`

Skills are reusable capabilities.

Examples:

SEO analysis
Landing page critique
Cold email writing
Debugging
UI review
Data cleanup
Competitor research

A skill should define how Claude performs a certain type of work.

`commands/`

Commands are repeatable shortcuts.

Examples:

/review
/audit
/debug
/ship
/summarize
/refactor
/research

If you repeatedly type the same instruction, make it a command.

`output-styles/`

Use this folder to define how Claude should format answers.

Examples:

Executive summary
Brutal audit
Client-ready explanation
Technical spec
Implementation plan
Bug report
Founder memo

Good output structure saves editing time.

`memory/`

This folder stores persistent project knowledge.

Examples:

Saved decisions
Client preferences
Product direction
Long-term constraints
Repeated business rules
Strategic context

This is how Claude stops forgetting what matters.

`workflows/`

Workflows are multi-step processes.

Examples:

PR review workflow
Feature build workflow
Bug fix workflow
Content creation workflow
Campaign audit workflow
Data cleanup workflow
Release checklist

A workflow tells Claude how to execute, not just what to do.

`context/`

This folder is for background information.

Examples:

Business overview
Product notes
ICP
User personas
Offer details
Market context
Internal docs

This gives Claude the “why” behind the work.

`guides/`

Use this for how-to documentation.

Examples:

How to deploy
How to test
How to add a feature
How to debug common issues
How to create a campaign
How to use a tool

`prompts/`

Use this for reusable prompts.

Examples:

Code review prompt
Research prompt
Cold email prompt
Debugging prompt
Landing page teardown prompt
Data cleanup prompt

`snippets/`

Use this for reusable blocks.

Examples:

Code snippets
Email snippets
SQL snippets
API payload examples
UI copy blocks
Response templates

`scripts/`

Use this for automation scripts.

Examples:

Setup scripts
Cleanup scripts
Deployment scripts
Validation scripts
Data transformation scripts

`permissions/`

Use this to define what Claude can and cannot touch.

Examples:

Safe files
Protected files
Risky files
Approval-needed actions
Allowed commands
Forbidden commands

This matters if you want speed without chaos.

`routines/`

Use this for recurring workflows.

Examples:

Daily repo health check
Weekly cleanup
Monthly audit
Release prep
Content planning
Campaign reporting

`plugins/`

Use this for plugin-related configuration and instructions.

Examples:

Installed tools
Plugin setup
Tool-specific rules
Integration notes

`subagents/`

Use this for helper agents under larger agent systems.

Examples:

Test writer
Security checker
Docs updater
Refactor planner
UI critic

`examples/`

This folder teaches Claude what good output looks like.

Examples:

Good PRs
Good components
Good reports
Good landing pages
Good emails
Good API patterns

Claude copies patterns well.

Show it the patterns you want.

`teams/`

Use this to organize groups of agents.

Examples:

Engineering team
Marketing team
Research team
Content team
Debugging team

This is useful once your setup becomes more advanced.

The simple rule

Do not overcomplicate this.

Start with the files that remove the most repeated explanation:

CLAUDE.md
rules.md
stack.md
style.md
gotchas.md

Then add agents, workflows, and commands only when they solve a real repeated problem.

Do not build a fake AI bureaucracy.

Build useful context.

Best use cases

This .claude starter folder is useful for:

Claude Code users
AI builders
Developers
Indie hackers
Technical founders
Growth operators
Automation builders
Agencies
Content creators
GTM workflows
Internal tools
Fast-moving projects

Why this matters

Most people do not have an AI problem.

They have a context problem.

Claude can only work with what you give it.

If your context is scattered across old chats, vague memory, random docs, and half-written instructions, Claude will produce scattered output.

An organized .claude folder gives Claude:

Your rules
Your stack
Your taste
Your workflows
Your constraints
Your project memory
Your execution style

That is where the leverage comes from.

Brutal truth

The prompt is not the advantage.

The system around the prompt is the advantage.

If you keep explaining the same thing to Claude over and over, you are not using AI efficiently.

You are doing manual onboarding every single time.

This folder fixes that.

Recommended first setup

Start here:

Add the .claude folder to your repo.
Fill out CLAUDE.md.
Add your stack into stack.md.
Add non-negotiable rules into rules.md.
Add repo style into style.md.
Add known problems into gotchas.md.
Create three agents:
- Code reviewer
- Debugger
- Researcher
Add one workflow for your most repeated task.
Improve the folder every time Claude gets something wrong twice.

What this is

What you get

Folder structure

How to use it

Step 1: Drop the folder into your repo

Step 2: Fill in the files Claude keeps needing

Step 3: Add agents for repeated work

Step 4: Add workflows for repeated execution

Step 5: Keep improving the folder

File descriptions

CLAUDE.md

rules.md

persona.md

stack.md

style.md

env.md

gotchas.md

api.md

tasks.md

todo.md

changelog.md

mcp.json

settings.json

ignore.md

ports.txt

deploy.sh

Folder descriptions

agents/

hooks/

skills/

commands/

output-styles/

memory/

workflows/

context/

guides/

prompts/

snippets/

scripts/

permissions/

routines/

plugins/

subagents/

examples/

teams/

The simple rule

Best use cases

Why this matters

Brutal truth

Recommended first setup

Follow for more

Related in Open-source AI

`CLAUDE.md`

`rules.md`

`persona.md`

`stack.md`

`style.md`

`env.md`

`gotchas.md`

`api.md`

`tasks.md`

`todo.md`

`changelog.md`

`mcp.json`

`settings.json`

`ignore.md`

`ports.txt`

`deploy.sh`

`agents/`

`hooks/`

`skills/`

`commands/`

`output-styles/`

`memory/`

`workflows/`

`context/`

`guides/`

`prompts/`

`snippets/`

`scripts/`

`permissions/`

`routines/`

`plugins/`

`subagents/`

`examples/`

`teams/`