10 Best GitHub Repos for Claude Code in 2025
The top GitHub repos and skills builders use to extend Claude Code into video editing, scraping, app building, testing, and SEO.
Which GitHub repos actually extend Claude Code into real work?
The ten repos worth installing cover video editing (Hyperframes), Google Workspace automation, web scraping (Apify), scroll-based website design (Scrollcraft), copy editing (Humanizer), spec-driven planning (Agent Skills), mobile app builds (Expo’s official skills), backend security (Supabase and Stripe skills), browser testing (Playwright CLI and Claude Security), and SEO auditing (Open SEO). Each one solves a specific production bottleneck rather than just generating boilerplate code, which is what separates them from the thousands of throwaway repos built for Claude.
TL;DR
- Hyperframes turns natural-language prompts into rendered MP4 videos with captions and motion graphics, and it’s free and open source with no per-render fees.
- Google Workspace CLI gives Claude command-line access to Gmail, Drive, Calendar, Docs, Sheets, and Slides with over 40 pre-built skills, useful for automating small-business admin work.
- Apify’s MCP server connects Claude to thousands of ready-made scrapers for sites like Instagram and TikTok that block normal scraping, though the scrapers themselves run on Apify’s cloud and can cost money beyond the free $5/month credit.
- Agent Skills from Addy Osmani moves project planning out of chat and into spec, plan, and build files, so a fresh session can pick up exactly where a compacted or context-rotted one left off.
- Playwright CLI from Microsoft opens a real browser to click through an app like a user would and write repeatable tests, using far fewer tokens than an MCP-based alternative.
- Claude Security, Anthropic’s own plugin, runs multiple agents against your codebase looking for vulnerabilities and produces a plain-English report with patches.
- Open SEO plugs Claude into keyword, ranking, and backlink data so audits point to specific pages and terms instead of generic advice, though free trial credits are limited.
What problem does each repo actually solve?
Most Claude Code repos duplicate something the model can already sort of do. The ones worth keeping fix a specific failure mode.
Video editing is slow and expensive to outsource. Hyperframes, built by Haygen, lets Claude generate HTML and animations that render into an actual MP4 on your machine. You can build skills on top of it for recurring formats like YouTube intros, Instagram Reels, or course videos, which turns editing from a bottleneck into a few minutes of prompting.
Scraping is blocked by design on most social platforms. Sites like Instagram and Reddit either block automated requests outright or require JavaScript rendering that leaves Claude looking at an empty page. Apify routes around this with thousands of maintained scrapers, so a request like pulling Chicago bakeries’ Instagram follower counts and bio links becomes a spreadsheet instead of a dead end.
Generic AI-built websites look generic because Claude defaults to safe, predictable layouts. Scrollcraft addresses this by giving Claude an actual scroll engine plus a set of accumulated design rules, along with a built-in checker that screenshots scroll positions so layout bugs get caught before a human has to flag them.
AI-written copy has tells: rhythmic “not this, but that” constructions, everything grouped in threes, inflated language. Humanizer strips these patterns out using the same style of guide Wikipedia editors use to catch AI-generated text, which matters because copy that reads like it was written by a person is a much easier sell than copy that doesn’t.
Why does context management matter so much for longer builds?
Long Claude Code sessions degrade. As a session runs, earlier context gets compacted or summarized away, a problem often called context rot, and the model starts giving answers that ignore details established earlier in the build. This is the exact failure mode Addy Osmani’s Agent Skills repo targets.
Instead of keeping the plan in the chat, Agent Skills writes it into files through three commands: /spec produces a real specification, /plan breaks that spec into discrete tasks, and /build works through those tasks one at a time with a check after each step. Because the plan lives in files rather than in the conversation, a fresh session can resume exactly where a messy one left off. The spec step also surfaces edge cases non-developers tend to miss, like double bookings, refund handling, or customers in different time zones, which also doubles as a document you can hand a client before any code gets written.
Superpowers, covered in an earlier roundup, still works for this same purpose, but Agent Skills is described as more approachable for people newer to the workflow. Matt Pocock’s skills are a notable alternative worth checking out for more developer-oriented builds.
How do you take a build from prototype to something a client can actually use?
Remy doesn't build the plumbing. It inherits it.
Other agents wire up auth, databases, models, and integrations from scratch every time you ask them to build something.
Remy ships with all of it from MindStudio — so every cycle goes into the app you actually want.
Once a client asks for something real, like a booking and deposit system for a cleaning company, the requirements change. Expo’s official skills handle the mobile side: they include Expo’s list of traits that make an app look obviously AI-built, along with current app store submission rules. Claude doesn’t always load these skills automatically, so they need to be called by name. Shipping to TestFlight or the App Store also requires a $99/year Apple developer account.
On the backend, Supabase and Stripe’s official skills enforce rules that are easy to get wrong when vibe coding, like ensuring users can only see their own data or confirming a payment genuinely originated from Stripe. The recommendation is to install only the “Stripe best practices” skill rather than the full plugin, since the full plugin can try to route every vendor decision through Stripe. Setting the Supabase connector to read-only and testing as a single logged-in user before launch is a basic safeguard against data leaks.
Before anything ships, Playwright CLI from Microsoft opens an actual browser and clicks through the app the way a user would, writing reusable tests in the process. Because it’s a command-line tool rather than an MCP server, it doesn’t dump entire pages into Claude’s context, which keeps it lighter on tokens and faster to run. Claude Security, Anthropic’s own plugin, complements this by running multiple agents through the codebase to hunt for vulnerabilities, with agents cross-checking each other’s findings before generating a plain-English report and patches.
Is Open SEO worth adding after launch?
Once a site or app is live, the remaining problem is discoverability. Open SEO connects Claude to keyword data, search rankings, and backlink information, which lets it generate audits pointing to the exact page and keyword where a client is ranking rather than generic SEO advice. Agencies use audits like this to justify ongoing SEO retainers. The tradeoffs are real: free trial credits are limited, and by default your data sits on the provider’s server unless you self-host.
What about token-saving repos like Caveman or RTK?
Several repos marketed as token or cost savers didn’t hold up well under testing. JetBrains ran Caveman, RTK, and Ponytail against real coding tasks: Caveman cut roughly 8.5% of output tokens despite advertising closer to 65%, RTK saved essentially nothing and sometimes increased cost at low effort settings, and Ponytail was the only one that delivered real savings, around 10% per task, about half its advertised number, though it has an open bug on Windows that can stall messages for several seconds. A separate repo called Codeburn tracks where tokens actually go and flags waste, like an MCP server that loads on every message but rarely gets used. Two previously recommended memory tools, Claude Mem and Context Mode, are worth holding off on for now: Claude Mem sends usage data by default, and Context Mode has an open bug that can burn through usage unexpectedly.
Frequently Asked Questions
Do I need to install all ten repos at once?
No. Installing everything at once makes it hard to tell which repo caused a problem if something breaks. Pick the repo that solves the specific bottleneck you’re currently facing, and add others as new problems come up.
Are these repos free to use?
The repos themselves are open source and free to install. Some depend on paid services behind them, such as Apify’s scraper credits beyond the free $5 monthly allowance, or Open SEO’s limited free trial data.
Which repo is the easiest starting point for beginners?
The Google Workspace CLI, Scrollcraft, and Humanizer are reasonable starting points since they solve common, narrow problems (document automation, site design, and copy quality) without requiring backend or security knowledge.
Why use Playwright CLI instead of an MCP-based testing tool?
Playwright CLI runs as a command-line tool rather than dumping full page contents into Claude’s context window, which makes it lighter on tokens and faster while still opening a real browser to click through an app and catch bugs before launch.
Is it worth paying for the full Stripe plugin or just the best-practices skill?
The guidance is to start with just the “Stripe best practices” skill. The full plugin can attempt to route every vendor decision through Stripe, which adds complexity most small builds don’t need at the outset.
