Claude Code in the Wild with Tanay Mishra | Clay Bootcamp Session Recap

Guest Speaker

Meet Tanay Mishra

Tanay Mishra

Claude Code Coach at Clay Bootcamp

Tanay teaches founders, GTM teams, and career-switchers how to build real tools with Claude Code, no technical background needed. He is the coach behind the Clay Bootcamp Claude Code cohort program and has taught the Clay team itself how to use Claude Code. He has been using and teaching Claude Code for over a year, and his approach is simple: plan first, build fast, and never let anyone gatekeep you from building.

Session Overview

What we covered

📚

Foundations first

Why Claude Code is different from ChatGPT, how it actually manipulates files on your computer, and what makes it so powerful for non-technical builders.

💡

Plan before you build

Tanay's core framework: spend an unreasonable amount of time planning with Claude before a single line of code gets written. This is the biggest separator between power users and everyone else.

🛠️

Live build-along

We built a real email enrichment skill using Blitz API. Starting from a CSV of prospects, Claude Code found email addresses for all contacts using a single slash command.

⚡

Skills and reusability

How to turn a one-off build into a reusable skill you can run any time with one command. Next time you have 1,000 prospects, just type /enrich.

The Build

What we built, step by step

Starting from nothing, we built a working email enrichment skill in about 45 minutes using plain English. Here is exactly how it happened.

Create a project folder and launch Claude Code

One folder per project. On Mac, right-click the folder, select Services, then New Terminal at Folder. Type claude to launch. Every project gets its own directory. This keeps Claude focused and prevents context bleed between projects.

First tiny win: create a hello.md file

Tanay asked Claude to create a file called hello.md that says "I did it, Claude Code is working." The file appeared instantly. This proves everything is running and shows the core capability: Claude Code can actually manipulate files on your computer. ChatGPT cannot do this.

Plan before building with the golden prompt

Before writing a single line of code, Tanay used his golden prompt to have Claude ask clarifying questions. Claude interviewed him about the goal, the data format, the API, and edge cases. Only once all questions were answered did the build begin.

Store the API key securely in a .env file

When connecting to Blitz API, Claude Code created a hidden .env file and stored the key there. API keys should never leave your computer or be committed to code. Tell Claude: "I've put the key in a .env file, look it up from there."

Test on one contact first

The prompt specified: do not enrich everyone yet. Test on one contact first. Claude tested on Amy Chen and successfully retrieved her email address. Only then did we move forward. This is the right way to build: prove it works before scaling it.

Save progress with a CLAUDE.md file

Tanay asked Claude to create a CLAUDE.md file documenting everything that was built. This is your project's save file. If your computer crashes or the session ends, you come back, read the CLAUDE.md, and pick up exactly where you left off without re-explaining anything.

Create a reusable /enrich skill

Tanay saved the entire process as a Claude Code skill called /enrich. After restarting Claude, he typed /enrich, selected all contacts, and Claude enriched every contact in the CSV automatically. All eight emails found in one command. That same skill will work with a CSV of 1,000 people next week.

Frameworks

Tanay's rules for Claude Code

One folder per project

Never work from your desktop or root folder. Create a dedicated directory for every project. Claude will co-mingle files and context if you do not.

Plan before you build

The biggest separator between power users and everyone else is how long they spend planning. Make Claude ask you questions first. Do not let it write a single line of code until it has fully understood your goal.

Every session starts and ends with the .md file

Start by reading your CLAUDE.md. End by updating it. The more context you give Claude, the better the output. Documentation is not overhead. It is your competitive advantage.

Skills are brainless, repeatable workflows

A skill is a sequential set of steps Claude can follow without thinking. The moment anything changes, update the skill. If it requires judgment, it is not ready to be a skill yet.

Your edge is business knowledge, not code

You do not need to understand the code. Your biggest differentiator is domain expertise, business judgment, and knowing what problem actually needs to be solved.

Prompts from the session

Copy these and use them

These are the exact prompts Tanay used. The point is not to copy them word for word. It is to see that you can talk to Claude in plain English and get real results.

First tiny win

Create a file here called hello.md that says: I did it, Claude Code is working.

Golden prompt (use this to start every project)

Hey Claude, I want to build [describe your idea]. Before you start, ask me at least five clarifying questions to make sure you understand exactly what I want. Do not build anything just yet.

The build prompt (used in this session)

Hey Claude, I want to build a Claude Code skill that enriches a contact giving a person's name and company. I want to go and find their email address using Blitz API. Please ask me five clarifying questions and then make sure you create a step-by-step plan. Do not build anything just yet. Before you build the skill, we must test it on at least one contact. If it works, after that we can work on formalizing the skill, but do not build it just yet.

Save your progress

Can you create a CLAUDE.md file for this project? Document everything we have built so far, what this project does, and how to use it. I just want to save my progress.

Create a reusable skill

Save this as a reusable Claude Code skill called enrich with a skill.md file that explains what it does, what inputs it needs, and how to run it so I can just run this anytime I want.

Resume a session after a break

Let's pick up where we left off. Please read the CLAUDE.md file and let me know when you are ready to continue.

Commands cheat sheet

Useful commands for the terminal

These are the commands mentioned in the session. Save this page and refer back as you get started.

Terminal

# Install Claude Code (run once)

$npm install -g @anthropic-ai/claude-code

# Navigate to your project folder

$cd my-project

# Or create and enter a new folder

$mkdir my-project && cd my-project

# Launch Claude Code

$claude

# Claude Code slash commands (type these inside Claude Code)

>/model# Switch model (Sonnet for Pro, Opus for Max)

>/init# Auto-create a CLAUDE.md file for your project

>/skills# See all available skills

>/enrich# Run the enrich skill built in this session

# Basic terminal navigation

$ls# List files in current folder

$pwd# Show your current folder path

$cd ..# Go up one folder level

On Windows: right-click a folder and choose "Open in Terminal". Install Node.js from nodejs.org first, then run the install command above.

Q&A Highlights

Questions Tanay answered

Where do the email addresses actually come from? Is this legal?

Most email addresses are available in the public domain through published blogs, LinkedIn profiles, and other sources. Data providers like Blitz, Prospeo, and BrightData aggregate this public data. Tools like Clay run a waterfall: if Blitz does not have the email, it tries Prospeo, then BrightData, until it finds a match. Nothing is being hacked.

What is the safest way to store API keys so they do not get leaked?

Create a hidden .env file in your project folder and store your keys there. Never let API keys leave your computer. The moment they do, consider them compromised. Tell Claude "I've put the key in a .env file, look it up from there." Never paste API keys directly into a chat or commit them to code.

CLI versus VS Code. Which should I use?

The terminal is the most powerful place to run Claude Code. Everything else (VS Code, Claude Cowork, Cursor) gives you a subset of that power. Yes, there is a learning curve. But working in the terminal unlocks everything Claude Code can do. Start there. If you are brand new and just want to ship something first, VS Code is fine as a starting point.

Can I have one big project folder with sub-projects inside it?

Yes, if the projects share a common goal and need to share data between them, nesting is fine. The rule is: if projects are independent and do not need to reference each other, separate them into their own folders. If they work toward the same overarching outcome, one parent folder with nested project folders and a top-level CLAUDE.md works well.

What is the difference between Opus 4.6 and Sonnet?

Opus 4.6 is significantly more powerful for complex, multi-step tasks. One attendee described it as a Ferrari versus a Ford Fiesta. Sonnet is capable and will get you far, especially for straightforward builds. If you are on the Pro plan ($20/mo), use Sonnet. If you are on Max ($100 or $200/mo), use Opus. When you start hitting your usage limits frequently, that is the sign to upgrade.

What if my skill needs to change? Do I rename it?

You update the skill. A skill should be a brainless, sequential workflow that Claude can follow without thinking. The moment the API endpoint changes, your CSV structure changes, or you swap data providers, you go back and update the skill file. A skill that requires judgment is not ready to be a skill.

Can Claude Code connect to Salesforce and update it directly?

Yes, absolutely. Claude Code can connect to Salesforce through its API. Just be careful about the guardrails and permissions you give it when writing back to a production CRM. Test on a sandbox first.

Once we have emails, can Claude Code connect to a sequencer and start outreach?

Yes, through the sequencer's API. Claude Code can make API calls to almost any tool that has one. This is exactly the kind of end-to-end GTM workflow the cohort teaches: find prospects, enrich them, write personalised emails, push to a sequencer.

Tools mentioned