The Execution Layer of AI
Most people use AI as a chatbot. You're about to turn it into an execution machine that writes code, deploys websites, sends emails, posts content, manages your calendar, and runs your entire business — while you sleep.
The Blueprint
Think of it like building a person. Each layer makes the whole system smarter, faster, and more capable.
Your entire knowledge base in markdown files. Every document, note, and piece of context lives here — linked like a Wikipedia for your business. Claude reads this as its long-term memory. The more you feed it, the smarter it gets.
Skills are markdown files that teach Claude how to do specific things. Want it to write emails like a pro? There's a skill for that. Deploy websites? Skill. Generate content? Skill. More skills = more capabilities.
CLIs and integrations let Claude actually do things in the real world — push code to GitHub, deploy to Vercel, send emails via Gmail, post to social media, manage your calendar, query databases. These are the arms and legs.
Wispr Flow is a voice-to-text app that runs on your Mac. Instead of typing prompts into Claude Code, you just talk. It transcribes your speech in real-time, directly into any text field — including Claude Code's terminal.
The average person types 40 words per minute. Speaking? 150+ WPM. That's nearly 4x faster input. For long prompts, project briefs, and creative direction — talking beats typing every time.
Works in Claude Code, Slack, email, Notion — anywhere you can type. Runs locally on your Mac. Free tier available.
The Gap
Chat AI is a brain in a jar. Claude Code is a brain with hands, legs, eyes, and a toolbelt.
| Capability | Chat AI (ChatGPT, Gemini, Claude.ai) | Claude Code |
|---|---|---|
| Write code | Shows code you copy-paste manually | ✓ Writes files directly on your computer |
| Edit existing files | You paste code in, it suggests changes | ✓ Reads your actual project, edits in-place |
| Run terminal commands | ✕ Can't touch your terminal | ✓ Runs npm, git, curl — any CLI natively |
| Deploy a website | "Here's how to deploy..." | ✓ Runs vercel deploy, gives you the live URL |
| Send emails | "Here's a draft..." | ✓ Drafts AND sends via Gmail integration |
| Manage calendar | "Here's what I'd schedule..." | ✓ Creates & edits events directly |
| Automate browsers | ✕ Can't | ✓ Opens browsers, fills forms, takes screenshots |
| Manage databases | "Here's the SQL..." | ✓ Runs migrations, queries, manages auth |
| Process video/audio | ✕ Can't | ✓ Transcribes, edits, generates |
| Post to social media | ✕ Can't | ✓ Posts to 10+ platforms via Upload Post API |
| Work autonomously | Waits for your next message | ✓ Multi-step tasks, sub-agents, parallel work |
| Persistent memory | Limited chat history | ✓ Obsidian vault + CLAUDE.md + Supermemory |
The Setup
No terminal commands. No coding. Just copy each prompt below and paste it directly into Claude Code. Claude handles the rest.
First, you need a Claude account. Go to claude.ai and pick a plan:
Download & install Claude Code for your OS (Mac, Windows, or Linux)
One download. Installs everything you need — no terminal commands required.
↓ Download Claude CodeAfter installing, open Claude Code and log in with your Claude account. That's it — you're in.
This is the most important step. Copy this prompt and paste it into Claude Code. It builds your knowledge base, writes your master instruction file, and then asks which folder you want it to absorb into your wiki.
I need you to build my entire AI assistant foundation. Do everything below:
PHASE 1 — OBSIDIAN VAULT SETUP:
I should already have Obsidian installed (from obsidian.md).
Create a vault folder in my Documents folder called "SuperBrain"
(detect my OS and use the right path automatically).
Build this structure:
- 00-Inbox/ (new unsorted notes land here)
- 01-Projects/ (one subfolder per active project)
- 02-Areas/ (ongoing areas: business, health, finance, relationships, learning)
- 03-Resources/ (reference material, research, frameworks, templates)
- 04-Archive/ (completed or paused items)
- Templates/ (reusable note templates)
- _Maps of Content/ (index pages that link to everything, like a Wikipedia home page)
Create these templates in Templates/:
1. Project Note — title, status, goals, tasks, links, next action
2. Meeting Note — date, attendees, agenda, decisions, action items
3. Daily Note — date, top 3 priorities, log, reflections, wins
4. Person/Contact — name, company, relationship, context, notes
5. Research Note — topic, sources, key findings, quotes, gaps
Every file MUST have YAML frontmatter:
---
title: "Note Title"
date: YYYY-MM-DD
tags: [tag1, tag2]
status: active/complete/paused
type: project/meeting/daily/person/research/resource
---
Use [[wiki-links]] for all internal references.
Use #tags for categorization.
Use > [!note], > [!tip], > [!warning] for callouts.
Create a _Home.md dashboard that links to every Map of Content.
Create a Map of Content for each area (Business MOC, Health MOC, etc.)
PHASE 2 — CLAUDE.md (My Master Instruction File):
Create a CLAUDE.md file in my current working directory with:
- Identity: "You are my AI executive assistant, builder, and strategic partner.
Communicate directly: bullets, action-oriented, no fluff, no filler.
Lead with action. Challenge weak thinking."
- Link to my Obsidian vault (the path you just created)
- Rules:
* Ask before: pushing code, deploying, sending emails/messages
* Read files before editing them
* Never share API keys or secrets
* Save all research and notes as Obsidian-formatted markdown
* Update this CLAUDE.md whenever a new tool or skill is installed
- Tools section: check what CLIs and tools are already installed on this
machine (gh, vercel, node, python, whisper, playwright, etc.) and list them
- Add a "Current Projects" section (leave blank for me to fill in)
PHASE 3 — KNOWLEDGE ABSORPTION:
Now ask me: "Which folder on your computer has files you want me
to absorb into your Super Brain? (Documents, PDFs, notes, anything)"
When I give you a folder path:
1. Scan every file in that folder recursively (PDF, .doc, .docx, .txt,
.md, .csv, .json, .html, images with text, etc.)
2. For each file, extract the key information and convert it into a
clean Obsidian markdown note with proper frontmatter
3. Organize notes into the right vault folders (Projects, Areas,
Resources, etc.) based on content
4. Create [[wiki-links]] between related notes automatically
5. Build a Map of Content that indexes everything you absorbed
6. Give me a summary: X files processed, X notes created,
X links generated, organized into these categories
The goal is to turn any messy folder into a structured, searchable,
interlinked knowledge base — like a personal Wikipedia.
PHASE 4 — VERIFICATION:
Show me:
- The vault structure you created
- The CLAUDE.md contents
- How many tools you detected on my system
- Confirmation that everything is ready
Then say: "Your Super Brain is live. What folder should I absorb first?"
Pick the capabilities you want below. Each card has a ready-to-paste prompt — click it, copy the prompt, paste it into Claude Code. Claude installs and configures everything for you.
Know Your Controls
Claude Code asks before doing things. These modes control how much freedom you give it. Start with Ask, graduate to Auto as you build trust.
Claude asks permission before every action — reading files, editing code, running commands. Safest mode. You approve each step. Best when you're learning or doing something sensitive.
Claude can read and write files freely without asking, but still asks before running terminal commands. Great for coding sessions where you trust the edits but want control over deployments and installs.
Claude creates a detailed plan first, then waits for your approval before executing anything. Use this for big tasks where you want to review the strategy before Claude starts building.
Claude runs everything without asking — file edits, terminal commands, installs, all of it. Maximum speed, zero interruptions. Only use this when you fully trust the task and can undo mistakes.
When Things Go Wrong
Claude is built to debug itself. When something breaks, don't Google it — tell Claude to fix it.
Just paste the error. Say: "This happened. Fix it." Claude reads the error, finds the root cause, and fixes it. You don't need to understand the error — Claude does.
Say: "Undo that and try again. What I actually want is..." Claude can undo its own changes and take a different approach. Be specific about what you DO want, not just what went wrong.
Say: "I'm trying to [goal]. I don't know how. Figure it out for me." Claude will research, plan, and execute. You describe the destination — Claude finds the route.
"Figure it out"
Claude researches and solves independently
"Don't ask me, just do it"
Claude makes decisions instead of asking questions
"Explain this like I'm 10"
Gets simple explanations instead of technical jargon
"What would you recommend?"
Claude gives opinionated advice instead of listing options
"Do these 3 things in parallel"
Spawns subagents for simultaneous execution
"Undo everything since [point]"
Claude reverts its own changes to a known good state
Pick Your Superpowers
Each card below gives Claude the ability to do something new. Click to see the prompt, copy it, and paste it into Claude Code. Claude handles all the installation.
★ = Start with these ◆ = Add when you need them
I need you to set up my deployment tools. Do all of this:
GitHub CLI Docs: https://cli.github.com/manual/
Vercel CLI Docs: https://vercel.com/docs/cli
1. Install the GitHub CLI if I don't have it (check with `gh --version`)
- If missing: brew install gh (Mac) or download from cli.github.com
- Run `gh auth login` and walk me through connecting my GitHub account
2. Install the Vercel CLI if I don't have it (check with `vercel --version`)
- If missing: npm install -g vercel
- Run `vercel login` and walk me through connecting
3. Verify both are working by running the version commands.
4. Update my CLAUDE.md to note these tools are now available.
If you hit any issues, refer to the docs linked above.
From now on, when I say "deploy this" you should use these tools.
What you unlock: Say "deploy this" and Claude pushes your code to GitHub and puts your site live on the internet. Free tier covers most projects.
I need you to set up Upload Post for social media cross-posting.
Docs: https://docs.upload-post.com/
1. Install the Upload Post SDK:
npm install upload-post
2. I need to sign up at upload-post.com and get my API key.
Walk me through the signup process.
3. Once I give you the key, store it securely:
- Save in .env.local as UPLOAD_POST_API_KEY
- Verify .env* is in .gitignore
- NEVER hardcode it in source files
4. Create a reusable helper script using the SDK:
import { UploadPost } from 'upload-post';
const client = new UploadPost(process.env.UPLOAD_POST_API_KEY);
That accepts: content, platforms, and optional schedule time.
5. Update my CLAUDE.md to note Upload Post is configured.
Refer to the full API reference at:
https://docs.upload-post.com/api/reference/
Platforms supported: TikTok, Instagram, YouTube, LinkedIn,
Facebook, X, Threads, Pinterest, Reddit, Bluesky, Google Business.
Free tier = 10 uploads/month (1 upload hits multiple platforms).
What you unlock: Tell Claude "post this to Instagram and TikTok" and it handles everything — formatting, uploading, even scheduling for later.
I need you to set up audio/video transcription on my computer.
1. Check if Python is installed (`python3 --version`).
If not, install it for my OS.
2. Check if ffmpeg is installed (`ffmpeg -version`).
If not, install it (brew install ffmpeg on Mac,
or download from ffmpeg.org on Windows).
3. Install OpenAI Whisper: pip install openai-whisper
Docs: https://github.com/openai/whisper
4. Test it works by running: whisper --help
5. Update my CLAUDE.md to note Whisper is available.
If you hit any issues, refer to: https://github.com/openai/whisper
From now on, when I give you an audio or video file,
you should transcribe it using the "medium" model
(best balance of speed and accuracy), then:
- Save the transcript as an Obsidian markdown note
- Add YAML frontmatter (title, date, source, tags)
- Break it into logical sections with headers
- Bold the most important quotes and insights
- Add a TL;DR summary at the top
What you unlock: Drop any audio or video file into Claude Code and get a beautifully formatted transcript saved to your Obsidian knowledge base. No API costs — runs on your machine.
I need you to set up browser automation with Playwright.
Docs: https://playwright.dev/docs/intro
1. Install Playwright and its browser engines:
npx playwright install
2. Verify it works by taking a screenshot of google.com
and showing it to me.
3. Update my CLAUDE.md to note Playwright is available.
From now on, when I need you to interact with a website
(fill a form, take a screenshot, check if something looks right,
scrape visible content), use Playwright to do it.
If you hit any issues, refer to: https://playwright.dev/docs/intro
I need you to set up Firecrawl for web scraping and research.
CLI Docs: https://docs.firecrawl.dev/sdks/cli
API Docs: https://docs.firecrawl.dev/api-reference/introduction
1. Install the Firecrawl CLI:
npm install -g firecrawl-cli
2. I need an API key from firecrawl.dev — walk me through
signing up (free tier = 500 credits, no card required).
3. Once I give you the key (starts with fc-), authenticate:
firecrawl login --api-key fc-MY-KEY
Also store in .env.local as FIRECRAWL_API_KEY
4. Verify it works: firecrawl scrape https://example.com
5. Update my CLAUDE.md to note Firecrawl CLI is available.
Available commands:
- firecrawl scrape [url] — extract clean content from a page
- firecrawl crawl [url] — crawl entire site recursively
- firecrawl map [url] — discover all URLs on a site
- firecrawl search [query] — web search with content extraction
From now on, when I say "research [topic]" or "scrape [url]",
use Firecrawl to extract the data, then save findings as
Obsidian markdown notes with proper frontmatter and links.
I need you to set up the NotebookLM CLI for research.
Docs: https://github.com/teng-lin/notebooklm-py
CLI Reference: https://github.com/teng-lin/notebooklm-py/blob/main/docs/cli-reference.md
1. Check if Python 3.10+ is installed: python3 --version
If not, install it for my OS.
2. Install with browser support for login:
pip install "notebooklm-py[browser]"
3. Install browser engine: playwright install chromium
4. Log in with your Google account:
notebooklm login
Walk me through the browser authentication flow.
5. Verify it works: notebooklm notebooks list
6. Update my CLAUDE.md to note NotebookLM CLI is available.
Available commands:
- notebooklm notebooks create "Topic Name"
- notebooklm sources add [url/file] --notebook [id]
- notebooklm audio generate --notebook [id] (podcasts)
- notebooklm slides generate --notebook [id] (decks)
- notebooklm quiz generate --notebook [id] (flashcards)
- notebooklm chat "question" --notebook [id]
Note: This is an unofficial CLI (not from Google). It uses
reverse-engineered APIs. Great for personal research but
may break if Google changes their endpoints.
I need you to set up the Google Workspace CLI.
Docs: https://github.com/googleworkspace/cli
npm: https://www.npmjs.com/package/@googleworkspace/cli
1. Install the Google Workspace CLI:
npm install -g @googleworkspace/cli
2. Set up authentication (requires gcloud CLI):
gws auth setup
Then: gws auth login
Walk me through connecting my Google account.
3. Verify it works by listing my next 5 calendar events:
gws calendar events list --max-results 5
4. Update my CLAUDE.md to note Google Workspace CLI is available.
The `gws` command covers 40+ Google Workspace APIs:
- Gmail: gws gmail messages list / send / read
- Calendar: gws calendar events list / create / delete
- Drive: gws drive files list / download / upload
- Sheets: gws sheets values get / update
- Docs: gws docs get / create
It also has 100+ built-in agent skills and supports
structured JSON output on every command.
If gcloud is not installed, install it first:
https://cloud.google.com/sdk/docs/install
I need you to install the GSD (Get Shit Done) framework.
Docs: https://github.com/get-shit-done-cc/gsd
1. Run: npx get-shit-done-cc@latest
2. Follow the interactive setup prompts.
3. Verify it's working by running /gsd-help
4. Update my CLAUDE.md to note GSD is available.
This framework lets me say /gsd-new-project to start
a structured project with specs, phases, and quality gates.
It prevents context rot on long sessions by using fresh
context windows and sub-agents for each phase.
If you hit issues, refer to the docs above for troubleshooting.
I need you to install the GSAP animation skills.
GSAP Docs: https://gsap.com/docs/v3/
Skills: https://github.com/greensock/gsap-skills
1. Run: npx skills add greensock/gsap-skills
2. Verify the 8 skills are installed (core, timeline,
scrolltrigger, plugins, utils, react, performance, frameworks)
3. Update my CLAUDE.md to note GSAP skills are available.
From now on, when I ask for scroll animations, page transitions,
or interactive effects on a website, use GSAP best practices
from these skills. You now know the correct API for animations,
timelines, scroll-triggered effects, and React integration.
Refer to https://gsap.com/docs/v3/ for any API questions.
I need you to set up Remotion for programmatic video creation.
Docs: https://www.remotion.dev/docs/
API Reference: https://www.remotion.dev/docs/api
1. Create a new Remotion project in my current folder:
npx create-video@latest my-videos
(select the default template when prompted)
2. Verify it works by starting the Remotion Studio:
cd my-videos && npx remotion studio
Confirm the preview loads in my browser.
3. Install the Remotion skills for Claude Code:
npx skills add remotion/skills
These teach you the correct Remotion API patterns.
4. Update my CLAUDE.md to note Remotion is available.
Refer to https://www.remotion.dev/docs/ for troubleshooting.
From now on, when I ask you to create a video, you should:
- Write a Remotion composition using React components
- Use <AbsoluteFill>, <Sequence>, useCurrentFrame()
- Add animations with interpolate() and spring()
- Render to MP4 using: npx remotion render src/index.ts
CompositionName out/video.mp4
Types of videos I might ask for:
- Social media clips with text animations and transitions
- Product demo videos with screenshots and callouts
- Data visualization videos (charts that animate)
- Branded intro/outro sequences
- Quote cards with animated text for Instagram/TikTok
What you unlock: Describe a video and Claude writes the code, previews it in Remotion Studio, and renders it as an MP4. No video editing software needed — just words in, video out.
I need you to install the official Anthropic document skills.
Docs: https://github.com/anthropics/skills
1. Run: npx skills add anthropics/skills
2. This installs skills for: PDF, DOCX, PPTX, XLSX creation,
plus the Skill Creator (for building custom skills).
3. Verify the skills are installed.
4. Update my CLAUDE.md to note document skills are available.
From now on, when I ask you to create a presentation, report,
spreadsheet, or PDF, you can generate the actual file directly.
Refer to https://github.com/anthropics/skills for full list.
I want to set up scheduled tasks so you can do things
automatically on a recurring basis — even when I'm not here.
Claude Code has built-in scheduled tasks. Here's how to use them:
RECURRING TASKS (runs daily, weekly, etc.):
Tell me something like:
"Schedule a task called 'morning-briefing' that runs every
weekday at 9am. It should check my email, summarize what's
new, and create a daily briefing note in my Obsidian vault."
"Schedule a task called 'social-content' that runs every
Monday at 10am. It should draft 5 social media posts for
the week based on my recent content and save them as drafts."
"Schedule a task called 'competitor-check' that runs every
Friday at 3pm. It should check competitor websites for
changes and summarize findings."
ONE-TIME TASKS (runs once at a specific time):
"Schedule a one-time task for tomorrow at 2pm to remind me
about the client call and prep a briefing document."
MANAGING TASKS:
"Show me all my scheduled tasks"
"Pause the morning-briefing task"
"Delete the competitor-check task"
"Update the social-content task to run on Tuesdays instead"
HOW IT WORKS:
- Tasks are stored as skill files in ~/.claude/scheduled-tasks/
- Each task is a complete prompt — Claude runs it fresh each time
- Tasks run in your local timezone (not UTC)
- You get notified when each task completes
- Tasks can read/write files, call APIs, use all your superpowers
BEST SCHEDULED TASKS FOR BUSINESS OWNERS:
- Daily inbox summary + action items (every morning)
- Weekly content batch (drafts for the whole week)
- Daily sales pipeline check (pull CRM data)
- Weekly competitor monitoring (scrape + summarize)
- Monthly report generation (pull metrics, create PDF)
- Daily journal prompt (create today's Obsidian note)
I want to set up Dispatch — Claude Code's parallel agent system.
WHAT DISPATCH DOES:
Instead of doing tasks one at a time, Dispatch lets Claude
split work across multiple agents running simultaneously.
Think of it like having 5 employees working at once instead of 1.
HOW TO USE IT:
When you have a big task, tell Claude to "dispatch" or
"run these in parallel." For example:
"Dispatch these 3 tasks in parallel:
1. Research competitor pricing for [product]
2. Draft an email sequence for my launch
3. Audit my landing page for conversion issues"
Claude will spin up separate agents for each task.
They all run at the same time, then report back with results.
BEST USE CASES:
- Research multiple topics simultaneously
- Generate content for different platforms at once
- Audit multiple pages/files in parallel
- Run code fixes across multiple files
- Compare 5 competitors at the same time
SETUP:
Dispatch is built into Claude Code — no installation needed.
Just tell Claude what you want done in parallel.
Add this to your CLAUDE.md under a "## DISPATCH RULES" section:
## DISPATCH RULES
- When I say "dispatch" or "in parallel", use the Agent tool
to spawn multiple subagents for simultaneous execution
- Each agent gets a clear, self-contained task
- Prefer dispatch for any task with 2+ independent subtasks
- Report results from all agents in a single summary
- Default to background agents unless I need results before next step
I want to connect my iMessage to Claude Code so I can
read and send messages through you.
OPTION 1 — OFFICIAL ANTHROPIC PLUGIN (Recommended):
Claude Code has a built-in iMessage plugin. Set it up:
1. Grant Full Disk Access to your terminal app:
System Settings → Privacy & Security → Full Disk Access
→ Add Terminal (or iTerm, Warp, whatever you use)
2. Tell Claude Code:
"Install the official iMessage plugin and run /imessage:configure"
3. That's it. The plugin reads ~/Library/Messages/chat.db
directly with built-in access control (allowlist-based).
OPTION 2 — ALL-IN-ONE APPLE MCP (Messages + Notes + Mail + Calendar):
If you want Claude to access ALL your Apple apps, not just iMessage:
1. Same Full Disk Access step as above.
2. Tell Claude Code:
"Install the Apple MCP server by running:
bunx --no-cache apple-mcp@latest
Then add it as an MCP server."
3. This connects Messages, Notes, Contacts, Mail,
Reminders, Calendar, and Maps — all in one server.
WHAT YOU CAN DO:
- Read recent messages from any contact or group chat
- Search your entire iMessage history by keyword or date
- Send iMessages to contacts (requires your approval each time)
- Summarize long group chat threads you missed
- Draft and send replies
SECURITY:
- All data stays LOCAL on your Mac — nothing leaves your machine
- The plugin uses an allowlist — only contacts you approve
- Sending always requires your explicit confirmation
- Never store message content in CLAUDE.md or committed files
Add to CLAUDE.md:
## iMessage
- iMessage plugin is connected (read + send, allowlist-based)
- NEVER store message content in files or commits
- ALWAYS ask permission before sending any message
- Use for: reading messages, drafting replies, summarizing threads
Copy & Paste Ready
Pre-built prompts for the most common workflows. Click to expand, copy, and paste directly into Claude Code.
I want you to absorb an entire folder into my Obsidian Super Brain.
Folder to absorb: [DRAG A FOLDER HERE OR TYPE THE PATH]
My vault is at: ~/Documents/SuperBrain/
(if the vault isn't set up yet, run the Super Brain setup from Step 2 first)
For every file in this folder (recursively scan all subfolders):
1. Read the file — handle PDFs, .doc, .docx, .txt, .md, .csv, .html,
.json, images with text, spreadsheets, presentations, anything
2. Extract the key information, ideas, frameworks, and data
3. Convert into a clean Obsidian markdown note with proper frontmatter:
---
title: "[Descriptive Title]"
date: [today's date]
source: "[original filename and path]"
tags: [auto-generate relevant tags]
status: active
type: resource
---
4. Organize each note into the right vault folder based on content:
- Business-related → 02-Areas/Business/
- Project-specific → 01-Projects/[project-name]/
- Reference material → 03-Resources/[category]/
- People/contacts → 03-Resources/People/
5. Create [[wiki-links]] between related notes automatically
(if Note A mentions a concept from Note B, link them)
6. If you find recurring themes, create a Map of Content that
indexes all notes on that theme
After processing, give me:
- Total files scanned vs successfully converted
- Notes organized by category
- Top themes discovered
- Key connections found between notes
- Any files that couldn't be processed (and why)
Then update _Home.md to include links to the new content.
# Step 1: Install the SDK
npm install upload-post
# Step 2: Sign up at upload-post.com and get your API key
# Step 3: Connect your social accounts (one-time browser setup)
# Docs: https://docs.upload-post.com/
After installing, use this prompt in Claude Code to post content:
Post this video to TikTok, Instagram, and YouTube using
the Upload Post SDK (npm package: upload-post).
Video file: /path/to/video.mp4
Title: "Your caption here"
User profile: "your-profile-name"
Platforms: TikTok, Instagram, YouTube
Use the SDK like this:
import { UploadPost } from 'upload-post';
const client = new UploadPost(process.env.UPLOAD_POST_API_KEY);
Refer to the API docs for the correct method:
https://docs.upload-post.com/api/reference/
Schedule for tomorrow at 9am EST if possible.
My API key is stored in .env.local as UPLOAD_POST_API_KEY.
# Install Whisper (requires Python)
# Docs: https://github.com/openai/whisper
pip install openai-whisper
# Install ffmpeg (required for audio processing)
# macOS:
brew install ffmpeg
# Windows:
choco install ffmpeg
Use this prompt to transcribe and organize:
Transcribe this file using Whisper and save as markdown:
Input: /path/to/video-or-audio.mp4
Output: ~/Documents/SuperBrain/03-Resources/Transcripts/
Use the "medium" model for best speed/accuracy balance:
whisper "/path/to/file.mp4" --model medium --output_format txt
Then convert the transcript to a well-formatted Obsidian note:
- Add YAML frontmatter (title, date, source, tags)
- Break into logical sections with headers
- Bold key quotes and insights
- Add a TL;DR summary at the top
- Tag with relevant topics
Run a full research pipeline on this topic: [YOUR TOPIC]
Step 1: Search YouTube for the top 20 videos on this topic
from the last 6 months using yt-dlp. Get title, channel,
views, duration, and URL for each.
Step 2: Pick the 5 best videos based on views, relevance,
channel authority, and recency.
Step 3: Transcribe each video using Whisper (medium model).
Step 4: Create a NotebookLM notebook with all 5 transcripts
as sources. Run analysis: trends, outliers, gaps, key quotes.
Step 5: Produce a research brief as an Obsidian markdown note:
- Title, date, topic, tags in YAML frontmatter
- Executive Summary (3 bullet points)
- Key Findings (organized by theme)
- Notable Quotes (with source attribution)
- Gaps & Opportunities (what nobody is talking about)
- Sources (linked YouTube URLs)
Save to: ~/Documents/SuperBrain/03-Resources/Research/
Deploy this project to Vercel:
1. Run `vercel link` to connect to my Vercel account
2. Run `vercel` for a preview deployment
3. Share the preview URL with me
4. If I approve, run `vercel --prod` for production
If this is a new project, initialize it first:
- vercel link (select "Create new project")
- Set the framework (Next.js, static, etc.)
- Pull environment variables: vercel env pull .env.local
Take this content idea and create posts for multiple platforms:
Idea: [YOUR IDEA OR TOPIC]
Create:
1. Twitter/X thread (5-7 tweets, hook first, CTA last)
2. LinkedIn post (professional tone, story format, 1300 chars)
3. Instagram caption (with emoji, hashtags, CTA)
4. TikTok/Reels script (15-60 sec, hook in first 3 sec)
5. YouTube Shorts script (under 60 sec)
6. Email newsletter paragraph (for my subscriber list)
Rules:
- Lead with a hook that stops the scroll
- Use the RCQHT formula: Recognition, Curiosity,
Qualifier, Hand-Raiser, Takeaway
- Each piece should stand alone but drive to the same CTA
- Include platform-specific formatting (threads, hashtags, etc.)
- Tone: direct, energetic, value-first
Then use Upload Post to schedule all posts for tomorrow 9am.
The Team
Each agent is a specialist with specific tools and skills. They can work in parallel, hand off tasks, and operate autonomously.
~/.claude/agents/.
It contains the agent's role, personality, available tools, and rules. Claude Code can switch between
agents or run them in parallel. One founder. Eleven specialists. Zero payroll.
This isn't about replacing people. It's about giving one person the execution power of an entire team. Every email sent, every post published, every line of code deployed — all from one terminal.
Obsidian is the brain. Skills are the muscles. CLIs are the hands. Agents are the team.
The more you build, the smarter the whole system gets.
Cheat Sheet
No commands to memorize. Just say what you want in plain English.
"Deploy this project to Vercel and give me the live URL"
"Create a GitHub repo for this project and push all my code"
"Transcribe this video file and save it as a text document"
"Scrape this URL and summarize the key information for me"
"Post this video to TikTok, Instagram, and YouTube with this caption"
"Draft an email to my list about the new product launch"