Paste this guide into Claude Code and it will walk you through every step interactively.
10 Commands You Can Run Today
What this gives you: 10 copy-paste prompts you can run right now to see Claude Code in action. Generate a landing page, audit your security, deploy to Vercel, create invoices - real results in under 5 minutes each.
No setup needed. No configuration. Just open your terminal, navigate to any project folder, type claude, and paste one of these prompts. Each one delivers something real in under 5 minutes.
Before you start: Install Claude Code with npm install -g @anthropic-ai/claude-code and authenticate with claude auth. Then cd into any project folder and type claude to open the interactive session. Paste any prompt below.
1. Security Audit Your Project
The most important one. Most projects have secrets, vulnerabilities, and bad practices hiding in plain sight. This command finds them in seconds - things it would take you hours to audit manually.
Perform a full security audit of this codebase. Check for:
1. API keys, tokens, or passwords hardcoded anywhere (even in comments)
2. .env files that might be committed to git (check .gitignore)
3. Dependencies with known CVEs (check package.json or requirements.txt)
4. SQL injection risks or unsanitised user inputs
5. Exposed admin routes or endpoints without authentication
6. CORS policies that are too permissive
7. Sensitive data being logged to console or files
8. Missing HTTPS enforcement or insecure redirects
For each issue found: show the file path and line number, explain the risk,
and give me the exact fix to apply.
What you get: A prioritised list of security issues with file locations and copy-paste fixes. Clients pay security consultants thousands of pounds for this. You get it in 3 minutes.
2. Generate a README from Your Codebase
Every project deserves a clear README. Nobody writes them because it is tedious. Claude reads the entire codebase and writes one that is actually accurate - not the generic template you copied from somewhere and forgot to update.
Read through the entire codebase and generate a comprehensive README.md that includes:
1. Project name and one-sentence description (what it does, for whom)
2. Tech stack with versions (pull from package.json or requirements.txt)
3. Prerequisites (what needs to be installed before setup)
4. Installation steps (exact commands, in order)
5. Environment variables needed (list each one, what it does, example value)
6. How to run in development
7. How to run in production / deploy
8. Project structure - key directories and what lives in each
9. API endpoints or key functions (if applicable)
10. How to run tests (if tests exist)
Use real values from the actual codebase - do not use placeholder text.
Format it in clean markdown with code blocks for all commands.
What you get: A README that someone new to the project can actually follow. Especially useful when handing off to a client or bringing in a new developer.
3. Explain Any Codebase Architecture
Inherited someone else's code? Taking over a client project? This prompt produces a clear architectural overview in plain English - no assumed knowledge, no jargon, just "here is what this thing does and how."
You are a senior software architect explaining this codebase to a technical but
unfamiliar developer joining the team today. Read through all the key files and explain:
1. What this application does (in one paragraph, plain English)
2. The overall architecture pattern (MVC, microservices, monolith, serverless, etc.)
3. The data flow - how does a request enter the system and flow through to a response?
4. The key files and directories - what is the purpose of each important one?
5. The database schema - what data is stored and how is it related?
6. External services and APIs the app depends on
7. Where the business logic lives vs where the infrastructure code lives
8. What the hardest parts of this codebase are to understand, and why
9. What you would refactor first and why
10. Any known technical debt or workarounds you spotted in comments or code
Be specific. Reference actual file names, function names, and variable names
from the codebase so the explanation is grounded in what is really there.
What you get: A 1,000-2,000 word architectural overview that saves you 4-6 hours of spelunking through an unfamiliar codebase. Share it with your team as onboarding documentation.
4. Refactor the Worst Functions
Every codebase has that one file. The one everyone avoids. The one that is 500 lines long with nested ifs and variables named temp2. This command finds it and fixes it.
Identify the 3 worst functions or code sections in this codebase.
"Worst" means: hardest to understand, most likely to have bugs,
most difficult to maintain, or most violated best practices.
For each one:
1. Show me the current code with line numbers
2. Explain specifically why it is problematic (not just "it's messy")
3. Rewrite it as it should be written - cleaner, more readable, same behaviour
4. Show me the diff so I can see exactly what changed
5. List any edge cases the original code handled (intentionally or accidentally)
that need to be preserved in the refactor
After refactoring, confirm the function signature and return type are unchanged
so callers do not break.
What you get: Three refactored functions with explanations. Apply the changes, run your tests, and commit. Teams spend entire sprint weeks on refactoring work that Claude can do in 10 minutes per session.
5. Write Unit Tests
The most universally skipped part of software development. Not because developers do not know how - because it is tedious. This makes it instant.
Write a comprehensive unit test suite for this codebase.
Use the existing test framework if there is one (check for jest.config.js,
pytest.ini, vitest.config.ts, etc.). If there is none, use the most appropriate
one for this language and add it to package.json / requirements.txt.
For each key function or module:
1. Write tests for the happy path (normal inputs, expected outputs)
2. Write tests for edge cases (empty inputs, null values, max values, empty arrays)
3. Write tests for error cases (invalid inputs, missing required fields)
4. Write tests for boundary conditions (off-by-one, type coercion, async timing)
Do NOT write trivial tests that just check if a function exists.
Each test should assert something meaningful that could actually catch a real bug.
After writing the tests, run them and fix any that fail due to bugs
you discovered in the implementation (not test bugs - real bugs).
What you get: A working test suite with real coverage. The test run at the end often reveals bugs that have been sitting in production, unfound, because nobody was testing those paths.
6. Build a Landing Page in One Prompt
This one is genuinely impressive. Give Claude the context and it builds a complete, styled, responsive landing page that you can deploy. Not a template - a page with your actual offer, your actual copy, your actual structure.
Build a complete landing page as a single HTML file (inline CSS and JS, no external
dependencies except Google Fonts).
Business: [Your business name]
Offer: [What you sell in one sentence]
Target customer: [Who it is for]
Main benefit: [The #1 outcome they get]
Social proof: [A real result or testimonial - make one up if you do not have one]
CTA: [What you want them to do - Book a call / Sign up / Get a demo]
CTA link: [Your URL]
Design requirements:
- Dark background (#0a0a0a), white text, lime green (#a3e635) accent
- Mobile-first, fully responsive
- Sections: hero with headline + subheadline + CTA, 3 benefits, social proof,
second CTA at the bottom
- Hero headline must be bold, specific, and outcome-focused (not "Welcome to...")
- All buttons open CTA link in a new tab
- No placeholder text - every word should be final copy ready to publish
Output the complete HTML file content, ready to save and deploy.
What you get: A deployable landing page. Save the output as index.html, drag it into Vercel, and it is live in 60 seconds. The whole thing takes under 5 minutes.
7. Create a Shell Script Automation
Repetitive terminal tasks are the easiest thing to automate. If you do something more than twice, it should be a script. This prompt turns any workflow description into a working bash script.
Write a bash script that automates the following workflow I do manually every week:
[Describe your manual workflow here. Example below:]
Every Monday morning I:
1. SSH into my VPS (ssh my-vps)
2. Pull the latest code for 3 different projects (git pull in each directory)
3. Restart each project's pm2 process
4. Check that all processes are running (pm2 list)
5. Check disk usage to make sure we are not over 80%
6. Send myself a Telegram message with a summary of what happened
Write this as a bash script that:
- Has error handling (stops and alerts me if any step fails)
- Shows clear output for each step with timestamps
- Sends a Telegram message at the end with success/failure status
- Can be added to crontab to run automatically
Include the crontab entry I need to add at the bottom as a comment.
What you get: A working bash script with error handling, logging, and a crontab entry. The time you spent doing this manually every week becomes zero.
8. Analyse a CSV with Insights
You have data. You do not have insights. This command turns a CSV file into a clear analysis with actual recommendations - not just charts, but conclusions.
Read the CSV file at [path/to/file.csv] and perform a full data analysis.
First, describe what the data contains: columns, data types, row count,
date range if applicable.
Then analyse it for:
1. Key trends over time (if there is a date column)
2. Top performers and bottom performers (whatever the main metric is)
3. Anomalies or outliers that stand out and might need investigation
4. Correlations between variables (what moves together?)
5. Segments or groups in the data (are there natural clusters?)
6. Missing data or quality issues
Finally, give me 5 specific, actionable recommendations based on what you found.
Each recommendation should be: specific (not "improve X"), tied to the data
(reference actual numbers), and actionable (something I can actually do this week).
Format the output with clear sections and include the specific numbers that
support each finding.
What you get: A proper data analysis with recommendations. Hand this to a client and it looks like you spent a day on it. It took 5 minutes.
9. Deploy to Vercel from Terminal
For anyone who has never deployed to Vercel via CLI - this is the fastest way to go from local code to live URL. This prompt walks Claude through setting it up end-to-end.
Help me deploy this project to Vercel from the terminal.
Steps:
1. Check if Vercel CLI is installed (vercel --version), install if not (npm i -g vercel)
2. Check if this project is already linked to Vercel (.vercel directory)
3. If not linked: run `vercel` (interactive setup) and guide me through the prompts
4. Identify the build command and output directory from package.json or framework
5. Check if there are environment variables needed (look for .env.example or .env)
- List each one and ask me which values to add to Vercel
6. Set the environment variables in Vercel using `vercel env add`
7. Deploy to production with `vercel --prod`
8. After deployment: test the live URL returns a 200 status
9. If there is an error: read the build logs and fix the issue before redeploying
Give me the exact commands to run at each step.
What you get: A live URL. The commands, the env var setup, the verification - all handled. Typical time from "I want this live" to live URL is under 10 minutes.
10. Generate an Invoice
The most satisfying one. Describe the work you did, and Claude generates a professional invoice as an HTML file you can print to PDF and send. No invoicing software required.
Generate a professional invoice as a clean HTML file I can print to PDF.
Invoice details:
- From: [Your name / company name]
- From address: [Your address]
- From email: [Your email]
- Invoice number: INV-[MONTH]-[YEAR]-001
- Invoice date: [Today's date]
- Due date: [30 days from today]
Bill to:
- [Client company name]
- [Client contact name]
- [Client address]
Line items:
- [Service description 1]: [Amount]
- [Service description 2]: [Amount]
- [Add more as needed]
Design: Clean, minimal, professional. White background, dark text.
Include your name/logo at top, itemised table in the middle,
subtotal / VAT (20%) / total at the bottom, payment terms below.
Bank details at the bottom: [Your bank name, sort code, account number]
Output the complete HTML. I will open it in a browser and Print to PDF.
What you get: A print-ready invoice HTML file. Open in Chrome, Cmd+P, Save as PDF, send. Looks more professional than most invoicing software output because you control every element of the design.
What to Do Next
You have just seen 10 things Claude Code can do in under 5 minutes each. These are the basics. The real power comes when you start combining them - security audit your project, refactor the issues it finds, write tests to lock in the fixes, deploy the clean version, and analyse the usage data. That is an entire software improvement cycle that used to take a sprint.
Beyond these quick wins, the more advanced patterns - multi-agent systems, parallelization, auto-research loops, VPS deployment - are what separate teams that use AI as a novelty from teams that use it as genuine infrastructure.
The learning curve is short. The ceiling is high. And the compounding effects of each improvement add up fast.
Want me to build your entire system?
From VPS setup to multi-agent orchestration to automated research loops - I can design and deploy the full stack for your business. One call to scope it out.