Every week I talk to consultants and operators who use AI constantly but start from scratch every single time.

New meeting, new prompt. New client brief, new prompt. New research task, new prompt.

They spend 10 minutes writing a prompt that gets them 80% of what they need, use it once, and never see it again.

That's not a workflow. That's improvising.

The professionals getting the most out of AI aren't necessarily writing better prompts in the moment. They've built a library of prompts that work, organized so they can find and use them in seconds, and connected to a version control system that means they never lose a good one.

This week I'm walking you through exactly how I built mine: task-based folders in Cursor, prompt files that work across different clients and contexts, and GitHub as the central archive that keeps everything versioned and backed up.

Next week in Part 2: how to use Cursor's @ file references to call any prompt instantly, and how to build variables into your templates so one prompt does the work of twenty.

Let's build it.

Personal experience

I started saving prompts the way most people do. Copy-pasting into a notes app. Bookmarking conversations. Keeping a running document that quickly became impossible to navigate.

The problem wasn't saving the prompts. It was finding them when I needed them, and knowing which version was the one that actually worked.

I moved everything into Cursor and GitHub about six months ago. The shift was simple: one GitHub repository called /prompts, organized into folders by task type, with each prompt saved as its own markdown file. Cursor opens the whole repository as a project, which means every prompt I've ever written is searchable, editable, and one command away.

The library started with four folders and maybe a dozen files. It now has prompts covering research, client communications, analysis, and meeting preparation. Every time I write a prompt that works well, it goes into the library. Every time I improve one, GitHub tracks the change.

The compounding effect is real. By month three the library was saving me more time than it took to build. By month six I couldn't imagine starting an engagement without it.

The setup: 5 steps

Step 1: Create your GitHub repository
Go to github.com and create a new repository called prompts. Set it to private. This is your central archive. Everything lives here, versioned and backed up automatically every time you push a change.

Step 2: Build your folder structure
Inside the repository create four folders to start. Keep it simple:

/research — prompts for market research, competitive analysis, company briefings

/client-comms — prompts for proposals, status updates, difficult conversations, executive summaries

/analysis — prompts for root cause analysis, red-teaming, data interpretation, financial review

/meetings — prompts for meeting preparation, transcript summarization, action item extraction

You can always add folders later. Start with the four categories you use most.

Step 3: Create your first prompt files
Inside each folder create markdown files, one per prompt. Name them clearly so you know exactly what's inside without opening them:

research/company-brief.md
research/competitive-scan.md
client-comms/status-update.md
client-comms/proposal-draft.md
analysis/root-cause.md
meetings/pre-meeting-prep.md

The file name is your search term. Make it obvious.

Step 4: Write your first prompt
Open meetings/pre-meeting-prep.md and paste in the morning briefing prompt from Issue #3. That's your first saved prompt. It works, you've tested it, and now it lives in the library permanently.

Do the same for the RCA prompt and red-team prompt from Issue #4, and the monthly review prompt from Issue #5. You already have five prompts worth saving. Your library already exists.

Step 5: Open the repository in Cursor
Open Cursor, click File, then Open Folder, and select your local prompts folder. Cursor now treats your entire prompt library as a project. Every file is visible in the sidebar, searchable, and editable. When you want a prompt, you open the file, copy it, and go.

Next week in Part 2 we'll go further: using Cursor's @ file references to load prompts directly into your workflow without copying anything, and building variables into your templates so they adapt to any client or context.

Tool ratings
Ease of setup: 8/10 — takes about an hour to build the initial structure, runs itself after that
Time saved: 20 to 30 minutes per day once the library has 10 or more prompts
Cost: Free on GitHub's base plan and Cursor's free tier

Save this as research/client-brief.md in your library

You are a senior research analyst preparing a client briefing. Using only publicly available information, give me a concise brief on the following:

Company or organization: [CLIENT NAME]
Reason for the brief: [MEETING PREP, PROPOSAL, DUE DILIGENCE]
What I already know: [ANY EXISTING CONTEXT]

Structure the brief as follows:

1. WHO THEY ARE
What the organization does in two sentences. No jargon.

2. WHERE THEY ARE
Current strategic position: what's working, what's under pressure, what's changing.

3. WHAT THEY CARE ABOUT
Their likely priorities right now based on public signals: announcements, hiring patterns, leadership statements, market conditions.

4. WHAT I SHOULD KNOW
One thing about this organization that most people walking into a meeting wouldn't know but should.

5. THE SMART QUESTION
One question I can ask in the meeting that signals I've done my homework and opens a genuinely useful conversation.

Keep it tight. I need to read this in under five minutes.

Save this file. The next time you prep for a client meeting, open client-brief.md, fill in the three variables at the top, and paste it into Claude or Perplexity. That's the library working.

More to come on the next issue (Part 2)

GitHub Copilot now reads your entire repository as context, not just the file you're in

GitHub's latest Copilot update gives it visibility into your full project structure when generating suggestions and responses. For prompt library workflows this means Copilot can reference your existing prompts when helping you write new ones.

Why it matters to you: your prompt library gets smarter as it grows. The more prompts you have saved, the more context Cursor and Copilot have when helping you refine and build new ones.

Anthropic releases Claude's system prompt best practices guide for professional users

Anthropic published a detailed guide on structuring prompts for consistent, high-quality outputs across complex professional tasks, including frameworks for role definition, context setting, and output formatting.

Why it matters to you: the principles in that guide are exactly what your prompt library should be built on. Part 2 next week covers how to bake those principles into reusable templates.

Knowledge workers who document their AI workflows report 3x faster onboarding for new team members

A workplace AI adoption study found that teams with documented prompt libraries and AI workflows onboard new members significantly faster than those where AI use is informal and undocumented.

Why it matters to you: your personal prompt library isn't just a personal productivity tool. It's an asset that scales when you bring on team members or hand work to a junior.

Here is the exact sequence:

Create the GitHub repository tonight. Five minutes. Call it prompts, set it to private, done.

Create the four folders: research, client-comms, analysis, meetings. Two minutes.

Save the five prompts you already have from this newsletter series into the right folders. Issues #3, #4, and #5 gave you six prompts worth keeping. Put them in the library now.

Open the folder in Cursor. Your prompt library is live.

That's it for week one. You don't need twenty prompts to start. You need a system that makes it easy to add the next one. That's what you just built.

Going further: every time you write a prompt that gives you a great output this week, save it to the library before you close the window. Most good prompts get lost because saving them feels like one more thing to do. Make it a habit this week while the library is new.