Claudart Agent工作流 是 AI Skill Hub 本期精选Agent工作流之一。综合评分 7.5 分,整体质量较高。我们推荐使用将其纳入你的 AI 工具库,帮助提升工作效率。
Claudart Agent工作流 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
Claudart Agent工作流 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 克隆仓库 git clone https://github.com/vankhaivn/Claudart cd Claudart # 查看安装说明 cat README.md # 按 README 完成环境依赖安装后即可使用
# 查看帮助 claudart --help # 基本运行 claudart [options] <input> # 详细使用说明请查阅文档 # https://github.com/vankhaivn/Claudart
# claudart 配置说明 # 查看配置选项 claudart --config-example > config.yml # 常见配置项 # output_dir: ./output # log_level: info # workers: 4 # 环境变量(覆盖配置文件) export CLAUDART_CONFIG="/path/to/config.yml"
The markdown operating layer for Claude Code & Codex CLI — memory, plans, and review, all in git.
<p> <a href="https://github.com/vankhaivn/Claudart/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/github/license/vankhaivn/Claudart?style=for-the-badge&color=orange"></a> <img alt="Pure Markdown" src="https://img.shields.io/badge/memory-pure_markdown-blue?style=for-the-badge"> <img alt="Offline-friendly" src="https://img.shields.io/badge/works-offline-green?style=for-the-badge"> <a href="https://github.com/vankhaivn/Claudart/issues"><img alt="Issues" src="https://img.shields.io/github/issues/vankhaivn/Claudart?style=for-the-badge&color=blue"></a> </p> </div>
---
The problem. Working with AI coding agents leaks state in every direction: every session starts blind, plans die in chat, the same decisions are re-discovered every week, andCLAUDE.md/AGENTS.mdaccumulates until it's a token sink that nobody trusts.
CLAUDART is a small set of slash commands and a layered markdown memory model that turns ad-hoc agent use into a reproducible workflow. Everything is plain markdown under .claude/ and .codex/ — versioned in git, reviewable in PRs, readable offline. No vector DB. No daemon. No cloud account.
```bash
install.sh does a fresh copy — great for a clean project, but it will clobber an existing setup and can't tell you what changed on an upgrade. If you already run your own agents/workflow, or you're upgrading from an older CLAUDART, paste this into your Claude Code or Codex session instead. Your agent reads the repo, diffs it against your project, and merges — asking before it changes anything you've customized:
Read https://raw.githubusercontent.com/vankhaivn/Claudart/main/INTEGRATE.md and follow it to integrate CLAUDART into this project. Ask me before touching anything I've customized.
/start # orient the session /plan add JWT middleware # write a persistent task doc — agent waits for your approval before coding /checkpoint # rebuild CONTEXT.md and sync state at end of session /learn # graduate recurring decisions into durable rules /doctor # health check whenever the setup feels off ```
Codex CLI: same flow, swap / for $codex- (e.g. $codex-start, $codex-plan).
Codex subagents are supported as an opt-in workflow. Tell Codex explicitly to use subagents, delegation, or parallel agents; CLAUDART then guides it to split critical-path work from sidecar explorers/workers, assign non-overlapping ownership, and persist durable results in task files.
```bash
| CLAUDART | Mem0 | Zep | LangMem | Understand-Anything | MemPalace | |
|---|---|---|---|---|---|---|
| **Setup** | curl \| bash | vector DB + Docker + OpenAI key | Neo4j + managed cloud | PostgreSQL + pgvector | curl \| bash or plugin | pip install + 300 MB model |
| **Human-readable** | ✅ | ❌ | ❌ | ❌ | ⚠️ JSON + dashboard | ⚠️ verbatim text, binary DB |
| **Works offline / air-gapped** | ✅ | ❌ | ❌ | ❌ | ❌ LLM required | ✅ |
| **PR-reviewable memory** | ✅ | ❌ | ❌ | ❌ | ✅ JSON committed to git | ❌ ChromaDB + SQLite binary |
| **Tool support** | Claude Code, Codex CLI, Cursor, Windsurf | API only | API only | LangGraph only | Claude, Codex, Cursor, Copilot, Gemini + 6 more | Claude Code, Codex CLI, Gemini CLI, MCP-compatible |
Every major coding tool converged on plain markdown in the repo — AGENTS.md already appears in ~20k public GitHub repos. CLAUDART makes the convention structured and adds the workflow pieces around it: orientation, planning, learning, hygiene checks, and code-review safety nets.
高质量的AI工作流项目
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
✅ MIT 协议 — 最宽松的开源协议之一,可自由商用、修改、分发,仅需保留版权声明。
经综合评估,Claudart Agent工作流 在Agent工作流赛道中表现稳健,质量良好。如果你已有明确的使用需求,可以直接上手体验;如果还在评估阶段,建议对比同类工具后再做决策。
| 原始名称 | Claudart |
| 原始描述 | 开源AI工作流:A modular and self-evolving rules framework to optimize context windows for Clau。⭐9 · Shell |
| Topics | AI工作流Shell |
| GitHub | https://github.com/vankhaivn/Claudart |
| License | MIT |
| 语言 | Shell |
收录时间:2026-05-30 · 更新时间:2026-05-31 · License:MIT · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端