I dug through GitHub Trending and pulled the 5 repos that actually matter this week. Below each one: what it does, what I'd use it for, the real install command, and a link straight to the repo.
→An ex-Google Maps PM open sourced his spy-satellite simulator: a photorealistic 3D globe streaming live aircraft, ships, satellites, and around 800 public cameras. 10,475 stars in 4 days.
→A laid-off geophysics PhD built his whole job hunt into Claude Code, ran 69 applications into 20 first interviews, and got hired off his own tool. 37,529 stars.
→544 reverse-engineered GPT-Image-2 prompts, packaged as an installable agent skill for Claude Code and Codex. Biggest star gain on all of GitHub this week at +11,300.
→Anthropic's community plugin marketplace gained 2,100 of its 2,514 stars this week, after one engineer's /eli5 post passed 262K views in hours.
🔒Drop your email for the full breakdown
All 5 repos with install commands, the gotchas nobody mentions, and my honest call on which ones deserve a spot in your stack. Plus the weekly builds I ship with these tools.
Join 3,000+ builders already getting this. No spam.
ALL 5 REPOS — WHAT THEY DO, HOW TO INSTALL
01
bilawalsidhu / gods-eye-view
A spy-satellite simulator in your browser from Bilawal Sidhu, who spent 6 years as a product manager on Google's 3D Maps. Photorealistic 3D globe with 13 live layers: aircraft on real transponder data, ships, satellites, earthquakes, NASA fire detections, and around 800 public cameras projected into the 3D cities. Click any airliner and ride its cockpit down on real telemetry, in night vision if you want. A voice agent with 28 tools flies the whole thing: say 'take me to Tokyo' or 'how many flights are over Texas right now' and it does it.
Stars
10,475 (+9.6k · launched Aug 24)
Layers
13 live · 10 need no key
License
MIT
git clone https://github.com/bilawalsidhu/gods-eye-view
npm install && npm run dev
Use it for: the best demo you can run locally this weekend · watching real flights, fires, and ships on one screen · reading how a production CesiumJS app is put together. One real cost: the photorealistic globe needs a Google Maps 3D-tiles key. The first 1,000 sessions a month are free, then about $6 per 1,000, so set a billing cap before you share your instance. Node 24 or 26 only.
Mads Lorentzen, a geophysics PhD in Copenhagen, got laid off in late 2025 and built his entire job hunt into Claude Code slash commands. /setup builds your profile from your CV, /scrape finds postings, /rank scores them against you, /apply drafts a tailored CV and cover letter as LaTeX PDFs, and a second reviewer agent with fresh context critiques every draft before you see it. It compiles the PDFs, visually inspects them, and checks the text layer that applicant tracking systems actually read. His own run: 69 applications, 20 first interviews, one signed offer. He started as an AI engineer in June.
Stars
37,529 (+4.3k this week)
Pipeline
/scrape → /rank → /apply → /interview
License
MIT
gh repo fork MadsLorentzen/ai-job-search
Use it for: an actual job search, it has a hard rule against fabricating skills · studying a real drafter-reviewer agent loop wired in plain Claude Code. Two warnings: GitHub forks are always public, so keep your personal data in a private mirror, and it needs Python 3.10+, Bun, and a LaTeX install.
544 reverse-engineered GPT-Image-2 prompt cases across 13 categories, from product shots to infographics to posters, each broken into an atomic schema: subject, lighting, materials, layout, details. The whole library ships as an installable agent skill, so Claude Code or Codex assembles a production-grade prompt on request instead of you freestyling one. There are also 20+ fill-in-the-blank templates and a gallery site. Biggest star gain on all of GitHub this week.
Stars
24,104 (+11.3k this week)
Cases
544 across 13 categories
License
MIT (code)
npx skills add freestylefly/awesome-gpt-image-2
Use it for: repeatable product shots, posters, and thumbnails from a model you already use · learning what actually moves GPT-Image-2 output by reading 544 worked examples. Heads up: the prompt bodies are still mostly Chinese, the skill handles that for you, and the maintainer says third-party case content carries no commercial-use guarantee.
An agent skill that draws architecture, workflow, sequence, data-flow, and lifecycle diagrams as self-contained HTML with crisp PNG, SVG, and WebM export. What separates it from every Mermaid pipeline is a fail-closed validation gate: schema, layout, and label-clearance checks all have to pass before a new diagram replaces the last good version, so a broken render never ships. Architecture nodes can also pin to a specific git commit and line range, so the diagram carries evidence. Works with Claude Code, Codex CLI, Cursor, and OpenCode. Hit #1 on GitHub Trending on Tuesday.
Stars
26,383 (+8.5k this week)
Types
5 diagram types, all validated
License
MIT
npx skills add tt-a1i/archify -g
Use it for: system diagrams for docs, decks, and posts that your agent can regenerate on demand · before/after diagram diffs on pull requests. It validates structure, not reality: nothing checks the diagram against your live infrastructure unless you pin nodes to code yourself. Needs Node.js.
Anthropic's community plugin marketplace for Claude Code and Claude Cowork: 2,282 plugins that passed automated security scanning, covering Stripe, Figma, AWS, Playwright, and Kubernetes, plus stranger things like a 12-role engineering team and 16 agents that debate your question. It blew up this week after Thariq Shihipar, an engineer on the Claude Code team, posted the /eli5 skill, which explains any topic as a picture-heavy artifact. His post passed 262K views in hours, and eli5 installs from this exact marketplace.
Stars
2,514 (+2.1k this week)
Plugins
2,282 · automated scanning
License
Apache 2.0
claude plugin marketplace add anthropics/claude-plugins-community
claude plugin install eli5@claude-community
Use it for: adding real integrations to Claude Code in one command · /eli5 first, it earns the hype. One honest flag: the scanning is automated, no human reviews submissions, and security researchers keep finding issues across the wider plugin ecosystem. Read a plugin before you install it.
