swipe file: my AI prompt library
the prompts, skills, and projects that keep me sane
I know a lot of creators preface content with “a lot of you have been asking me…”
It’s my social media ick. AND…I have gotten a lot of DMs asking for the specific prompts I’ve used or skills I’ve set up in Claude.
This post is for those people.
I’ve gained back a significant amount of time, energy, and mental space with this setup (and have been using that time to log off and read/learn/live more). This post will be a living library of the prompts, skills, and projects I’ve set up in Claude.
I also want to be clear that I firmly believe in the regulation of AI and for a moratorium on new data center builds until we have established better protections for the residents and communities who live where new data center constructions are being announced. Erin Brockovich (yes, THE Erin Brockovich) has an excellent Substack reporting on this.
Before I dive into the prompt/skill/project library, allow me (well, allow Claude) to define these terms:
Chat — A single conversation with Claude: one continuous thread of back-and-forth. Claude remembers everything within that one chat, but each chat is separate from your other chats — start a new one and it doesn’t know what happened in the others (unless they’re grouped in a project, or memory is turned on). The chat is the basic unit — one conversation, start to finish. Product Talk
Prompt — The instruction or message you give Claude — what you type or say to tell it what you want. A prompt can be a one-line question or a long, detailed set of instructions. The more specific and clear it is, the better the response. Every one of the things in this post is, at bottom, a prompt: you’re just writing better ones.
Skill — A folder of instructions, scripts, and resources that Claude loads on its own to do a specialized task in a repeatable way. The key difference from a prompt: a skill is reusable and Claude only pulls it in when it’s relevant to what you’re doing — you don’t paste it in every time. Think packaged expertise you set up once (your brand voice, a recurring workflow) and Claude reaches for automatically. Skills are available on Free, Pro, Max, Team, and Enterprise plans, and need code execution turned on, and they work across the Claude apps, Claude Code, and the API.
Project — A self-contained workspace that groups related chats, reference files, and custom instructions in one place. You set the instructions and upload knowledge once, and everything in that space is used across all the chats inside that project — so Claude already has the background and follows your instructions in every chat without you repeating yourself. Chats inside a project still stay separate from each other; they just share the same setup. (On paid plans, project knowledge scales up automatically for large file sets.)
A prompt is one message → a chat is one conversation made of many prompts → a project is a workspace that groups many chats around shared context → a skill is reusable instructions Claude loads by itself whenever the task calls for it.
I currently use Littlebird and Claude (and Monologue for voice dictation). If I were starting from scratch, I’d actually start with Littlebird.
You can get a 2 month free trial of Litterbird with code PVH5W6DX.
Claude and ChatGPT rely on what you tell it, which is why a context hack is so important. With Littlebird, the app is constantly monitoring your screen so it gets to know you and how you work almost immediately.
What I like about Littlebird is you can connect it to multiple inboxes and calendars, whereas you can only use 1 with Claude. That said, you have to be careful what you share wlth Littlebird if your work requires extra privacy and security. You can pause “context enabled” (the screen viewing capability on Littlebird), which I do whenever I’m working on Rhoshan Pharma work or checking that email.
If the thought of setting up agents and workflows in Claude Cowork is too intimidating for you (spoiler - it is for me), Littlebird is an excellent alternative. It’s become my executive assistant, project manager, and has saved me so much time and mental space. I use Claude solely for thought partnership—on building Hyphenated, in my daily check-ins to keep me rested and sane, and doing some light copy-editing of my yapping (I do this in incognito chats so my words aren’t used to train AI).
prompt index:
prompt 1: context hack
This is the most important chat-turned-skill you’ll create.
A context hack tells Claude (or whichever tool you use) everything it needs to know about you. Think of this as your AI foundation—it governs what and how you the AI tool, and how to make it work best for you.
Even if you’ve been using it for a while, a context hack can help refine your existing projects and workflows to be more efficient. Here are the two versions - if you’re starting from scratch, and if you’ve been using AI for a while.
version 1 — you’re brand new to Claude
I'm starting to use you as a thinking partner for both my work and my personal life, and I want you to actually know me — not hand me generic answers. Interview me. Ask one question at a time, and let each answer shape the next question. Start broad, then get sharper. Cover: - who I am and the season of life I'm in right now - what I'm working toward - how I want you to talk to me: push back hard, or just execute? one clear recommendation, or a few options? - the people who matter, and how - when I'm at my sharpest, and when I'm running on empty - what you should never bring up unless I raise it first - what I want to keep doing entirely myself After about 8–10 questions, stop and write me a "context profile" — organized, in plain language — that I can save and paste back to you anytime.
version 2 — you already use Claude and want to sharpen what it knows
I’ve been using you for a while for a mix of work and personal life, and I want you to have a sharper, more accurate picture of me. Do this in two steps:
1. First, tell me what you already understand about me — my roles, what I’m working toward, how I like to work, what I care about, the people who matter. Be specific, and flag anywhere you’re guessing.
2. Then ask me 5–7 sharp questions to fill the gaps that matter most — the ones that would change how well you can help me. Don’t ask me things you can already infer. Prioritize: what you get wrong about me, where I want pushback vs. where I want you to just execute, what you should never raise unless I do, my energy across a day and a week, and what I want to keep doing myself with no help.
When I’m done answering, rewrite my context profile — tighter and more accurate — so I can save it and reuse it.
Once you’re done, save this as a markdown file and upload it as a skill. Whenever you start a new project, add this skill to it. I also recommend updating your context hack every few months or so (I like to have a monthly AI day where I refine this skill, update my projects, or tinker with my vibe-coded apps).
I wish I had done this before I set up my AI accountability coach. This has been the most valuable thing I’ve created using Claude, and I’m not exaggerating when I say it’s changed my life (and saved it at times, during a really tough year).
