A 13-minute walkthrough of the six Claude Code skills Nate Herk says are the only ones businesses will actually pay for — plus how to package them into an AI-automation offer.
The six Claude Code skills that generate client revenue are the boring ones that save time, cut costs, or eliminate mistakes—not the flashy ones—and you sell them by marketing the business outcome, not the technical workflow.
Six Claude Code skills survive real client work because they save time, cut costs, or prevent mistakes � the only outcomes businesses actually pay for. The stack works as a chain: Skill Creator builds custom skills from plain-English descriptions, Superpowers forces senior-engineer discipline by planning and testing before writing, GSD spawns clean-context subagents to stop mid-session context rot, /review and /ultrareview catch verified bugs before merge, Context Mode strips tool-call garbage from the window and rebuilds session state on compaction, and Claude Mem persists project knowledge across sessions through a local vector database. Frontend Design is a bonus for non-AI-looking UIs. Sell the outcome � hours saved, fewer errors, faster delivery � not the workflow; start with one skill, ship a demo, expand from there.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →
Pattern interrupt that says most YouTube-famous skills are useless to real clients. Names the criteria: save time, save money, remove mistakes — and promises the six that pay.
Lists the verticals he's shipped to (real estate, HVAC, coaches, marketing) to earn permission, then restates the value: 'if you want to make money selling AI in 2026, start with what businesses already pay for.'
Official Anthropic skill that writes other skills from plain-English descriptions. Frames it as 'the factory that builds the product' — not what a client pays for, but what makes every other paid skill possible. Install command: /plugin install skill-creator@claude-plugins-official.
Quick aside: a skill is a markdown file teaching Claude one job, a plugin is a bigger package that can bundle skills + hooks + MCP servers. Also addresses comments-section conspiracies that he's an AI avatar.
Community skill (150K GitHub stars) that forces Claude to plan, test, review in two stages, and work in an isolated environment before writing code. Pitch: gets you from 60% to 80% one-shot quality, cuts debugging cycles and token costs. Tied directly to client-grade production work.
Names 'context rot' — the mid-session collapse where Claude forgets requirements and cuts corners — as the wall every Claude Code user hits.
Solution: spawn fresh sub-agents per task, each with a clean context window. Built-in quality gates catch scope reduction and enforce security. Autonomous mode lets it plan/execute/commit unattended. Pairs with Superpowers — Superpowers gives the process, GSD gives the clean environment to run it.
Built into Claude Code, not installed. /review runs a fast local structured code review; /ultra-review (launched with Opus 4.7) uploads your branch to a cloud sandbox, spins up parallel reviewer agents, and only surfaces independently reproduced bugs. Requires Claude Code 2.1.86+ and a Claude account. Pro/Max get 3 free runs, then $5–$20 per run.
Every tool call dumps raw output into the context window — Playwright snapshots, GitHub issues, logs. After 30 minutes, 40% of context is junk. Context Mode routes tool calls through a sandbox subprocess and returns only the slice Claude needs: a 56 KB Playwright snapshot becomes 299 bytes.
Second half of Context Mode: tracks every meaningful event (file edits, tasks, decisions, errors) in a local SQL DB so when Claude compacts, it rebuilds the session snapshot and injects it back. Sessions that died at 30 min now run 3 hours. Install with two slash commands + restart.
Where Context Mode keeps the current session clean, Claude Mem carries knowledge across sessions. Hooks into the lifecycle, auto-captures edits/decisions/bug-fixes via Claude's own agent SDK, compresses into semantic summaries, stores in local SQLite with vector search. Auto-generates folder-level CLAUDE.md files. Three-layer retrieval claims ~10× token savings vs dumping everything at session start.
Official Anthropic frontend-design skill — installs globally, makes Claude Code's UI output look 'a lot less AI generated.' Notes this is now baked into Claude Design (Anthropic Labs) but you still want the skill installed if you bring Claude Design output back into Claude Code.
The actual business lesson: stop pitching 'I'll build you an AI workflow.' Pitch saving 10 hours/week, cutting admin mistakes, speeding up or growing leads. Beginner playbook: pick ONE of the six, build a few workflows with it, show a demo — the business owner doesn't care about your experience, they care about value. CTA to free Skool community resource guide + next video.
The whole video is one move — burn the field of 100+ options down to six, name the criteria upfront, and tie each item to a paying business pain.
“They're simple, they're boring, but they are effective.”
“If you're watching this and you wanna make money selling AI in 2026, don't overthink it. Just start with what the businesses are already paying for.”
“The skill itself isn't what a client pays for. But every single skill that you're going to sell... they all come out of this one. It's the factory that builds the product.”
“Superpowers slows Claude down just enough to actually think the problem through, and it doesn't let Claude skip steps.”
“If you can get somewhere like 80% of the way there rather than just like 60% of the way there, that's a win.”
“Somewhere around halfway through your context window, it starts to fall apart. And this is essentially called context rot.”
“After about thirty minutes of real work, 40% of your context window is just garbage.”
“A 56 kilobyte playwright snapshot becomes 299 bytes.”
“Sessions that used to fall apart around thirty minute mark now run for three hours.”
“You need to be selling the outcome. You need to be selling saving the business owner ten hours a week or cutting their admin mistakes.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
Nate opens by burning the rest of the genre — '100+ skills tested, most don't matter' — then immediately narrows the field to six. The title promises a filter and the first 22 seconds deliver the filter's criteria: save time, save money, or remove mistakes. From there it's a pure curation play built on his 400-hour credential and his portfolio of real-estate, HVAC, coaching and marketing clients.
Nate's whole curation criteria. Anything that doesn't pass one of these three doesn't make the list — and these are also what the business owner is actually buying.
The full curated stack — Skill Creator is the factory, Superpowers gives Claude the engineering process, GSD gives it clean sub-agent context, /review and /ultra-review catch bugs pre-merge, Context Mode trims tool-call garbage in-session, Claude Mem carries memory across sessions, and Frontend Design fixes the AI-generic UI look.
Superpowers controls HOW Claude writes code (plan → test → review). GSD controls the ENVIRONMENT it writes code in (clean sub-agent contexts). One gives the process, the other gives the substrate the process runs on. They compose.
Daily flow: /review for everything (fast, local, cheap). /ultra-review only for high-stakes merges — payments, auth, DB migrations — where production bugs cost more than the $5–$20 run.
The pitch frame for AI-automation freelancers. Don't sell the workflow, don't sell the skill, don't sell the AI — sell what the business owner already wants to buy.
The beginner cold-start sequence. Resists the temptation to learn all six before pitching anything.
“I threw all of this into a resource guide that you can access for completely free inside of my free school community. You can check that out using the link in the description.”
Soft, low-pressure pitch tucked after the big 'sell the outcome' lesson. Followed immediately by an end-screen video card and a like-the-video ask. Note: he says 'school' but means Skool — the spelling drift is verbal habit.
00:00
00:03
00:04
00:0500:10
00:20
00:24
00:27
00:3100:3900:49
00:49
00:53
00:58
01:08
01:12
01:18
01:22
01:24
01:2601:30
01:3701:44
01:54
01:55
01:58
02:02
02:12
02:17
02:20
02:26
02:36
02:39
02:4902:55
02:58
03:02
03:05
03:0803:10
03:18
03:25
03:27
03:31
03:4103:41
03:46
03:50
03:5504:02
04:10
04:14
04:19
04:22
04:2504:28
04:32
04:3504:39
04:4704:57
04:57
05:07
05:14
05:24
05:34
05:38
05:46
05:5605:58
06:06
06:0906:14
06:20
06:24
06:2906:31
06:34
06:4406:53Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
11:19How five lightweight TypeScript plugins turn the Claude Code desktop UI into a multi-agent control center with live cache tracking, stream censorship, and concurrency locks.
October 2nd
26:42A 26-minute field guide to Claude Code subagents — when to use them, how to build them, and how to save money by matching model to task.
June 9th
24:55A 25-minute live build that covers why agentic workflows command premium fees, how to structure them with the WAT framework in Claude Code, and how to sell the result on value rather than hours.
March 8th
10:00:05A 10-hour zero-to-pro curriculum that teaches Claude Code from first command to paid client — no code written by hand.
March 12th
25:48Seven identical prompts, two Claude models, and one real question: when does paying double for Opus actually buy you something?
September 29th
20:34Same prompts, same brand kits, two coding agents — one landing page built by each, eight times over, scored on design and on the bill.
August 26th