Nate Herk's video promises the best GitHub repos for Claude after, he says, testing more than 100. The headline says six, but the spoken walkthrough numbers ten workflow picks. His eighth and ninth picks each pair two tools, making 12 linked repositories in the supplied resource list. This guide follows what the video actually covers and gives every project a direct source link.
Start with the job, not the install count. Nate's list moves from making a deliverable (video or a site), to operating a business (Workspace and research), to building and checking software (planning, app skills, browser tests, security, and SEO). His speed and savings examples are personal claims, not a controlled comparison. Open-source downloads also do not erase model, hosting, account, or cloud-service costs.
Watch Nate Herk's Walkthrough
Credit: Repo selection, examples, and personal workflow observations are Nate Herk's. Scrollcraft is his own project. The linked repositories are the source for installation requirements and current capabilities.
1. Create Video, Better Sites, and Better Copy
| Pick | Repository and useful first test | Check before selling it |
|---|---|---|
| 1. HyperFrames 00:24 | heygen-com/hyperframes turns coded HTML scenes into rendered video. Try one 15-second captioned product clip from footage you own; compare it with your current editing process. | Nate reports no per-render fee for the open-source tool, not a cost-free production pipeline. Budget for AI inference, source media, rendering hardware, revisions, and rights to music or assets. |
| 4. Scrollcraft 03:09 | nateherkai/scroll-craft is Nate's scroll-led site toolkit. Prototype one section that reveals a real product benefit, then check screenshots at desktop and mobile widths. | Motion is useful only if the page stays readable, accessible, fast, and conversion-focused. Confirm reduced-motion behavior and actual mobile performance. |
| 5. Humanizer 04:16 | blader/humanizer flags formulaic AI writing. Rewrite one homepage paragraph or outreach draft, then have its human author approve the meaning and tone. | It is an editing aid, not a fact-checker or a license to pass generated claims off as personal experience. Preserve evidence, names, and the client's voice. |
Nate's business angle is a repeatable deliverable, such as a small batch of approved short videos or a custom conversion page. The client pays for the result and the review, not for the fact that a GitHub repo exists.
2. Connect Business Work and Research
| Pick | Repository and useful first test | Access and cost boundary |
|---|---|---|
| 2. Google Workspace CLI 01:28 | googleworkspace/cli exposes Workspace tasks to an agent-friendly command line. Try a read-only weekly summary from a test folder or calendar. | The repository says it is not an officially supported Google product. Setup needs a Google Cloud project and OAuth. Grant only necessary scopes; require review before email sends, document edits, or sharing changes. |
| 3. Apify MCP Server 02:11 | apify/apify-mcp-server connects an assistant to Apify Actors. Try one small, permitted public-data question and record the source URL, date, and result quality. | The connector is open source; cloud Actors can cost money. Review website terms, privacy law, rate limits, and personal-data handling. Do not turn a scrape into unsolicited outreach without a lawful process and human review. |
3. Plan and Build the Product
| Pick | Repository and first test | What remains your job |
|---|---|---|
| 6. Addy Osmani's Agent Skills 04:57 | addyosmani/agent-skills uses a spec, plan, and build sequence. Write a one-page booking-flow spec with cancellations, double bookings, and time zones before generating code. | A written plan helps preserve requirements across sessions, but a client must still approve scope and acceptance tests. |
| 7. Expo Skills 06:20 | expo/skills guides React Native and Expo workflows. Build one device-tested screen and one task, such as viewing and confirming an appointment. | A phone preview is not a production release. Verify platform policies, accessibility, device behavior, backend security, and current developer-account fees separately. |
| 8a. Supabase Agent Skills 06:55 | supabase/agent-skills supplies backend guidance. Test that one signed-in user cannot read or change another user's record. | Skills are advice, not proof of correct row-level security. Keep production credentials out of prompts and have a qualified reviewer inspect access rules. |
| 8b. Stripe Best Practices 06:55 | stripe/ai: stripe-best-practices gives payment-integration guidance. Run test-mode checkout and webhook replay before accepting money. | Verify webhook signatures, idempotency, refunds, and failure paths against Stripe's current documentation. A generated payment flow needs human review. |
4. Test, Secure, and Measure
| Pick | Repository and first test | Limit |
|---|---|---|
| 9a. Playwright CLI 07:34 | microsoft/playwright-cli lets an agent operate a real browser. Test the booking flow, mobile layout, and failure messages as a customer would. | A passing click-through covers only the cases you actually ran. Keep regression tests and manually inspect visual and accessibility issues. |
| 9b. Claude Security 08:11 | Anthropic's Claude Security plugin reviews code and proposes fixes. Run it on a small authorized project, then reproduce each finding. | The plugin operates with the user's Claude Code permissions; it is not a sandbox. Review patches and rerun tests before merging. It cannot certify a product secure. |
| 10. OpenSEO 08:32 | every-app/open-seo offers SEO workflows and data connections. Audit one page and verify each opportunity against Search Console or another first-party source. | Check current data-source pricing, quotas, hosting, retention, and client authorization. A ranking report is a lead for investigation, not a ranking guarantee. |
Where to Start Without Installing Everything
- Name one recurring buyer problem. A weekly report, a landing page that fails to explain the offer, or a booking flow that loses customers is clearer than “AI automation.”
- Choose one repo and one sample. Reproduce a non-sensitive task manually first, then use the tool and compare time, quality, cost, and corrections.
- Inspect the trust boundary. Read the README, license, permissions, outbound services, and maintenance status. Use test accounts and minimum access.
- Deliver a reviewed outcome. Demonstrate the result to one reachable buyer. Ask whether it solves the costly problem before turning the prototype into a retainer.
Nate closes with a similar point: buyers do not pay for a free repo; they pay for a useful solution. His brief token-saving discussion at 08:56 names additional tools, but the supplied links do not include the cited benchmark. Treat those savings figures as unverified until you can inspect the test and reproduce it in your own workflow.
Jump to a Segment
| Time | Topic | Time | Topic |
|---|---|---|---|
| 00:00 | How Nate chose the picks | 00:24 | HyperFrames video editing |
| 01:28 | Workspace CLI and Apify | 03:09 | Scrollcraft and Humanizer |
| 04:57 | Planning with Agent Skills | 06:20 | Expo, Supabase, and Stripe |
| 07:34 | Browser tests and security | 08:32 | OpenSEO and real data |
| 08:56 | Token-saving caveats | 09:46 | Where to start |
Build a Small Offer From One Skill
This prompt turns the article's tools into a buyer-first experiment. It works in a general AI assistant; no repository or paid account is required to begin.
Find the paid outcome
Choose one bottleneck, one deliverable, and one buyer.
Sources and Limits
Creator: Nate Herk's full video and the supplied transcript. Source code and documentation: each of the 12 repository links above, plus Stripe's webhook guide. The transcript is the source for Nate's personal speed, pricing, and business examples; the linked projects are the sources for current setup and feature details. This article does not independently benchmark the repos or verify client revenue claims. Check each project's current license, API costs, and security requirements before adopting it.