Main English | Main 中文 | Codex English | Codex 中文
Production-ready configuration for Claude Code: global instructions, multi-language coding rules, plugins, MCP integration, and a self-improvement loop.
.
├── CLAUDE.md # Global instructions
├── settings.json # Settings (permissions, plugins, hooks, model)
├── lessons.md # Self-correction log template (auto-loaded via hook)
├── rules/ # Multi-language coding standards (common + python/typescript/golang)
├── hooks/ # Statusline with gradient progress bars (context + 5h usage)
├── mcp/ # MCP server config (Lark-MCP, Codex CLI)
├── plugins/ # Plugin installation guide (19 plugins, 5 marketplaces)
├── skills/ # Custom skills (paper-reading)
├── VERSION # Semantic version number
└── install.sh # One-command installer
One-line remote install (no clone needed):
bash <(curl -fsSL https://raw.githubusercontent.com/Mizoreww/awesome-claude-code-config/main/install.sh)Or install a specific version:
VERSION=v1.0.0 bash <(curl -fsSL https://raw.githubusercontent.com/Mizoreww/awesome-claude-code-config/main/install.sh)Local install (from clone):
git clone https://github.com/Mizoreww/awesome-claude-code-config.git
cd awesome-claude-code-config
./install.sh # Install everything
./install.sh --dry-run # Preview changes| Component | Included | Flag to install separately |
|---|---|---|
| CLAUDE.md | Yes | --claude-md |
| settings.json | Yes (smart merge) | --settings |
| Rules (all languages) | Yes | --rules [LANG...] |
| Skills | Yes | --skills |
| lessons.md | Yes (skip if exists) | --lessons |
| Plugins (core, 14) | Yes | --plugins |
| Plugins (ai-research, 5) | No | --plugins ai-research |
| MCP (Lark, Codex CLI) | No | --mcp |
./install.sh --rules python typescript # Rules only
./install.sh --plugins # Core plugins only (14)
./install.sh --plugins all # All plugins (19)
./install.sh --plugins ai-research # AI research plugins only (5)
./install.sh --mcp # MCP (Lark, Codex CLI)./install.sh --uninstall # Remove everything
./install.sh --uninstall --rules # Remove rules only
./install.sh --uninstall --force # Skip confirmation (CI/non-interactive)./install.sh --version # Show source / installed / remote versions- User corrects Claude → auto-saved to
~/.claude/lessons.md - Next session →
SessionStarthook auto-injects lessons into context - Pattern confirmed → rule promoted to
CLAUDE.md
settings.json includes two SessionStart hooks:
- startup: Injects lessons.md when a new session starts
- compact: Re-injects lessons.md after context compaction
Replaces the previous approach of requiring manual Read lessons.md in CLAUDE.md (more reliable).
A single-line status bar with gradient progress bars, powered by hooks/statusline.sh:
- Model + directory + git branch
- Context window: gradient bar (green → yellow → red) with percentage and size
- 5-hour usage: pulled from
api.anthropic.com/api/oauth/usage(cached 60s), shows reset countdown - Progress bars are fixed-width (20 chars) with 16-step color gradients
Configured via statusLine in settings.json:
"statusLine": {
"type": "command",
"command": "bash $HOME/.claude/hooks/statusline.sh"
}When settings.json already exists, the installer performs a smart merge (requires jq):
- env: Incoming values as defaults, existing values take priority
- permissions.allow: Union of both arrays (deduped)
- enabledPlugins: Merged, existing keys take priority
- hooks.SessionStart: Deduplicated by
matcherfield - statusLine: Incoming config takes priority
Without jq, a manual merge warning is shown instead.
CLAUDE.md includes a Code Review rule: whenever a code review is needed — whether requested by the user or triggered by a skill (e.g., code-reviewer, simplify) — Claude must invoke the mcp__codex-cli__review tool (powered by codex-mcp-server) instead of providing text-only feedback. This enables cross-model review: Claude orchestrates while Codex performs the actual diff analysis.
Install the Codex CLI MCP server with --mcp to enable this feature.
common/ → Universal principles (always loaded)
↓ extended by
python/ → PEP 8, pytest, black, bandit
typescript/ → Zod, Playwright, Prettier
golang/ → gofmt, table-driven tests, gosec
19 plugins across 5 marketplaces, organized into two groups:
Core plugins (14) — installed by default:
| Plugin | Marketplace | What It Does |
|---|---|---|
| superpowers | claude-plugins-official | Brainstorming, debugging, code review, git worktrees, plan writing |
| everything-claude-code | everything-claude-code | TDD, security review, database patterns, Go/Python/Spring Boot |
| document-skills | anthropic-agent-skills | PDF, DOCX, PPTX, XLSX creation and manipulation |
| example-skills | anthropic-agent-skills | Frontend design, MCP builder, canvas design, algorithmic art |
| claude-mem | thedotmack | Persistent memory with smart search, timeline, AST-aware code search |
| frontend-design | claude-plugins-official | Production-grade frontend interfaces |
| context7 | claude-plugins-official | Up-to-date library documentation lookup |
| code-review | claude-plugins-official | Confidence-based code review |
| github | claude-plugins-official | GitHub integration (issues, PRs, workflows) |
| playwright | claude-plugins-official | Browser automation, E2E testing, screenshots |
| feature-dev | claude-plugins-official | Guided feature development |
| code-simplifier | claude-plugins-official | Code simplification and refactoring |
| ralph-loop | claude-plugins-official | Session-aware AI assistant REPL |
| commit-commands | claude-plugins-official | Git commit, clean branches, commit-push-PR |
AI Research plugins (5) — install with --plugins ai-research or --plugins all:
| Plugin | Marketplace | What It Does |
|---|---|---|
| fine-tuning | ai-research-skills | Axolotl, LLaMA-Factory, PEFT, Unsloth |
| post-training | ai-research-skills | GRPO, RLHF, DPO, SimPO |
| inference-serving | ai-research-skills | vLLM, SGLang, TensorRT-LLM, llama.cpp |
| distributed-training | ai-research-skills | DeepSpeed, FSDP, Megatron-Core, Ray Train |
| optimization | ai-research-skills | AWQ, GPTQ, GGUF, Flash Attention, bitsandbytes |
See plugins/README.md for installation details.
CLAUDE.md includes a Version Changelog rule: when making version-level changes (new features, major refactors, breaking changes), Claude proactively maintains a CHANGELOG.md in the project root with structured entries covering features, design rationale, and caveats. This keeps design decisions traceable alongside the code.
| Skill | Description |
|---|---|
| paper-reading | Structured research paper summarization with auto-screenshot of key figures. Uses ar5iv HTML for arXiv papers, Playwright for figure capture, outputs standardized markdown (problem, method, experiments, insights). |
Place custom skills in skills/<name>/SKILL.md.
settings.json ships with bypassPermissions mode for power users. If you prefer safer defaults, change defaultMode to "default" and set skipDangerousModePermissionPrompt to false.
- Add a language: Create
rules/<lang>/extending common rules - Add a skill: Place in
skills/<name>/SKILL.md - Adapt CLAUDE.md: Customize for your shell, package manager, and project context
- Working for 10 Claude Codes by Hu Yuanming — Practical experience running multiple Claude Code instances in parallel
- Harness Engineering by OpenAI — "Harness Engineering": engineers shift from writing code to designing systems, using agents to generate a million lines of code
- Claude Code in Action by Anthropic Academy — Official course covering Claude Code integration, MCP servers, GitHub automation, and dev workflows
MIT