Weekly GitHub Stars #1 — Claude Code Ecosystem, Voice AI, Browser Automation

A weekly curated selection of noteworthy open-source projects from my GitHub Stars, organized by theme. This is the first issue.

Overview

This week I starred 8 projects across 5 domains:

Language distribution: Python dominates (5/8), with TypeScript, C, and HTML each appearing once.

Stars ranking:

Claude Code Ecosystem

Claude Code is rapidly forming its own tooling ecosystem. The three projects I starred this week cover best practices, multi-agent orchestration, and structured tutorials — complementing each other nicely.

shanraisshan/claude-code-best-practice

Once the #1 GitHub Trending repo, this is the most comprehensive reference for Claude Code. It systematically organizes 86 tips by category:

• Prompting & Planning: Writing effective CLAUDE.md files for project context
• Skills & Hooks: Best practices for custom skills and automation hooks
• Agents & Workflows: The Command → Agent → Skill orchestration pattern
• Git & PR: Advanced version control integration

The Orchestration Workflow diagram is particularly valuable — it clearly shows Claude Code’s internal Command → Agent → Skill call chain.

Best for: Developers already using Claude Code who want to level up. Great as a daily cheatsheet.

Yeachan-Heo/oh-my-claudecode

If claude-code-best-practice is the “knowledge base”, then oh-my-claudecode is the “execution engine”. It brings multi-agent orchestration to Claude Code:

• Team Mode: Staged pipeline (plan → PRD → exec → verify → fix loop) with parallel agents
• 32 specialized agents: Architect, researcher, designer, test engineer, etc.
• Smart model routing: Haiku for simple tasks, Opus for complex ones — 30-50% token cost savings
• tmux workspaces: Claude, Codex, and Gemini CLIs collaborating in split panes

Named after oh-my-zsh, it positions itself as the “plugin manager + orchestrator” for Claude Code. Multiple execution modes include Autopilot, Ultrawork, and Ralph (persistent verify/fix loops).

Best for: Teams managing complex projects, or developers orchestrating multiple AI agents.

luongnv89/claude-howto

Unlike the previous two, claude-howto is a structured tutorial with a progressive learning path:

• 10 tutorial modules: slash commands → memory → skills → hooks → MCP → subagents → plugins…
• 15-minute quickstart or 11-13 hours complete learning
• Mermaid diagrams, copy-paste config templates, and self-assessment quizzes per module
• EPUB export for offline reading

It bridges the gap between Claude Code’s official docs (reference-style) and what developers actually need (tutorial-style with learning paths). 690+ forks indicate active community adaptation.

Best for: Claude Code beginners, or experienced users wanting to systematically review their knowledge.

The three together: claude-howto (learn) → claude-code-best-practice (reference) → oh-my-claudecode (orchestrate)

AI Agent & Browser Automation

browser-use/browser-use

At 85k stars, the most popular project I starred this week. browser-use lets AI agents operate browsers like humans:

• Multi-LLM support: OpenAI, Anthropic, Google Gemini
• CLI tools: browser-use open, click, type, screenshot
• Project scaffolding: uvx browser-use init --template default
• Open-source benchmark: 100 real-world browser tasks for accuracy comparison

from browser_use import Agent
from langchain_openai import ChatOpenAI

agent = Agent(
    task="Go to Reddit, search for 'browser-use' and return the first post's title",
    llm=ChatOpenAI(model="gpt-4o"),
)
result = await agent.run()

Also offers a cloud version with stealth browsing, proxy rotation, and CAPTCHA solving.

Best for: Developers needing AI-driven browser automation, RPA engineers.

Voice AI

microsoft/VibeVoice

From Microsoft Research, a family of open-source voice AI models:

Uses continuous speech tokenizers at an ultra-low 7.5 Hz frame rate with a next-token diffusion framework. The ASR model is now integrated into Hugging Face Transformers v5.3.0.

Notably, the TTS code was removed due to misuse concerns — demonstrating responsible AI release practices.

Best for: Voice AI researchers, product developers needing high-quality ASR/TTS.

Developer Tools

sherlock-project/sherlock

A classic OSINT tool — one command to search usernames across 400+ social networks:

sherlock username

Supports Tor/proxy anonymization, CSV/XLSX output, Docker deployment, and Apify cloud execution. Install via pipx, Docker, Homebrew, or dnf.

Best for: Security researchers, OSINT analysts, or anyone curious about their digital footprint.

fastfetch-cli/fastfetch

The spiritual successor to the discontinued neofetch, rewritten in C for significant performance gains:

• Cross-platform: Linux, macOS, Windows, Android, FreeBSD… near-universal coverage
• JSONC config with JSON Schema for IDE autocompletion
• Highly customizable output modules and logos
• Proper Wayland support (which neofetch lacked)

Best for: Terminal aesthetics enthusiasts, sysadmins needing quick system info.

CAD / 3D Modeling

gumyr/build123d

Low star count but exceptionally well-engineered Python CAD library. Built on the Open Cascade geometric kernel, it offers two modeling modes:

• Algebra Mode (stateless): Pure operator-driven, obj += sub_obj
• Builder Mode (context managers): Interactive modeling with design history

from build123d import *

# Algebra Mode: create a filleted box
box = Box(10, 20, 5)
box = fillet(box.edges(), radius=1)

Supports SVG and STEP import/export, interoperable with FreeCAD and SolidWorks. For 3D printing, CNC, and laser cutting, programmatic CAD is more efficient and reproducible than GUI modeling.

The successor to CadQuery with a cleaner, more Pythonic API. Full type annotations and test coverage. Academic users can cite via Zenodo DOI.

Best for: Robotics researchers (URDF generation), 3D printing enthusiasts, engineers needing parametric modeling.

Weekly Takeaways

1. Claude Code’s ecosystem is maturing — from “single tool” to “platform + ecosystem”. The best-practice → howto → oh-my-claudecode trio forms a complete learn → reference → orchestrate pipeline.
2. browser-use at 85k stars shows AI Agent browser control is a genuine need.
3. Microsoft open-sourcing VibeVoice under MIT accelerates voice AI democratization.
4. build123d represents Python’s continued penetration into CAD — particularly useful for robotics researchers.

See you next week!

References

1. browser-use/browser-use — AI Agent browser automation
2. sherlock-project/sherlock — Social media username search
3. microsoft/VibeVoice — Open-source Voice AI
4. shanraisshan/claude-code-best-practice — Claude Code best practices
5. fastfetch-cli/fastfetch — System information tool
6. Yeachan-Heo/oh-my-claudecode — Claude Code multi-agent orchestration
7. luongnv89/claude-howto — Claude Code visual guide
8. gumyr/build123d — Python CAD modeling library