Many people might glance over it, thinking it's just another collection of Claude's prompts.
However, if you clone it and take a closer look, you'll find that this repository thoroughly and practically addresses the issue of 'turning Claude Code
from casual vibe coding into a true engineering productivity tool.'
It’s not just a simple pile of prompts; rather, it packages the entire system of agentic engineering.
Sub-agents, skill packs, commands, hooks, MCP tool integration, cross-model collaboration... all are modules that are directly reusable and version-controlled.
The warehouse maintainer has almost fully incorporated Boris Cherny (the author of Claude Code) and the team's practical experience.
Continuously updated (with commits today), akin to a living Claude Code engineering manual.
Key highlights worth focusing on
Voice input: /voice command supports Push-to-talk, just press the key to speak and feed prompts.
Supports 20 languages, paired with macOS and other tools, input speed is several times faster than typing. When coding, the mouth is quicker than the hands.
Boris Cherny: compiled multiple tips (13, 10, 12, 15, etc.), including:
Complex tasks must be planned in Plan Mode first, then let Claude execute with one click.
Use git worktrees to run multiple Claude sessions in parallel (team members can open 5-10 at once).
Lets Claude maintain CLAUDE.md as a living document of rules, updating them after each error to avoid repeating mistakes.
Claude can fix most bugs itself; just throw a description and say 'fix'.
Voice + mobile + teleport seamless cross-device conversation continuation.
Subagents + skills + hooks
Subagents: create independent agents for different functions, carrying exclusive memory and tools without polluting the main context.
Skills: turn repetitive tasks into reusable modules, auto-load.
Hooks: auto-formatting, security checks, permission routing, event triggers, etc.
MCP / LSP: directly connect to external tools, browser automation, databases, Chrome control, and even desktop operations (macOS Computer Use).
Supports Cross-Model: Claude Code handles implementation, Codex manages planning and review, enhancing efficiency.
Other productivity features:
/rewind for quick rollback, checkpointing for automatic tracking of edits.
Scheduled tasks /loop and scheduling.
Automated code review, PR analysis, CI/CD integration.
Status bar shows real-time context consumption, costs, and Git branches.
Agent SDK (npm/pip) helps package these capabilities into production-grade agents.
Safe mode: sandbox + auto-mode classifier to reduce risks.
Supports macOS, Windows (WSL), and Linux, usable in terminal, Cursor, VS Code, claude.ai/code, and mobile.
The repository itself is maintained using Claude Code.
Summary: Many Claude projects are closed after the demo, this feels more like a ready-to-use engineering toolbox + practical guide. If you want to truly leverage Claude Code for productivity, not just generate a few lines of code and manually fix them, this repository is worth cloning and digesting.
Project address:
https://github.com/shanraisshan/claude-code-best-practice