yuanzhuSAFE
Meta-Forge Workshop: turn recurring work into governed AI workflow assets. MCP server + Claude Code plugin + dsh plugin — staged writes, evals-gated forging, inference engine. 元铸工坊 SOSOFAST:把最佳实践铸成可治理的 AI 工作流资产
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
对外传播名 SOSOFAST(元铸工坊) · 技术生态名 yuanzhu · © MiaoYu
[](https://pypi.org/project/yuanzhu/) [](LICENSE) [](https://pypi.org/project/yuanzhu/) [](https://registry.modelcontextprotocol.io) [](server/tests) [](https://m8ven.ai/mcp/ravenwangchina-yuanzhu-3sw9ut)
按企业自身的工作习惯,把各领域的最佳实践定向复刻成 AI 工作流资产——像软件一样被开发、分发、使用、沉淀、进化,全员开箱即用。
当前状态:v0.2.2 已上架 PyPI(技术验证期)
160 测试全绿。装完即用的三层能力:
- 深度问答(内置):问一个问题自动走六步流水线——澄清 → 双视角并行作答 → 独立审查 → 综合裁决 → 人审采纳;采纳的洞见沉淀入库,后续提问自动反哺
- 一句话铸模板(forge):
yuanzhu forge "描述你的日常工作"→ AI 生成完整四段式模板 → 三道守门(dry-run 结构校验 → evals 行为验证 → 人审)→ Web 重跑评测转正 - 对话长出工作流:连续问同类问题,AI 发现模式后建议"铸成工作流?"——对话着对话着,工作流就长出来了
- 主动推衍(v0.1.6):开启自举开关后,AI 基于你的行为+知识+技能交叉分析,主动提议下一步——不只是回答问题,还想到你没想到的
- 领域工作流(内置 AIQA 测试/会议追踪):staged writes——AI 的每个写操作进待审中心,人批准才生效;Claude Code 等经 MCP 直连本体
平台底座:本体层(对象+动作)、工作流引擎(并行组)、多协议模型网关(自动读 dsh 配置)、Web 控制台 + CLI、对话学习(聊天记录导入→AI 提炼工作流候选)。
三柱验证:H7 轻量本体已全通;H1 双线全通(专家线+对话线真实数据实证);H2 模拟测试通过(真人测试待安排)。
快速开始(单机模式)
# 一键启动(Python 3.11+;pip 包含 Web 控制台与全部内置模板) pip install yuanzhu # 安装包(一条命令搞定,yuanzhu-server 是包自带的命令) yuanzhu-server # 启动服务 → 打开 http://127.0.0.1:8600
首启自动注册内置模板(深度问答/AIQA 测试/会议追踪);CLI 同装:yuanzhu status / pending / templates / forge。
从源码运行(开发者)
git clone https://github.com/RavenWangChina/yuanzhu.git cd yuanzhu/server python -m venv .venv && .venv/Scripts/pip install -e ".[dev]" .venv/Scripts/yuanzhu-server # 或 pyt
2caf81ea9704OBSERVED · 2026-10-11Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.
claude mcp add yuanzhu -- None yuanzhu==0.2.2
Exposed tools (5)
4 read · 1 write · 0 destructive.
| Tool | Risk | Description |
|---|---|---|
yuanzhu_actions | read | 列出元铸工坊可用的动作类型(按域分组,含中文说明与自主性等级 L1自动/L2人审)。只读,用于发现能做什么。 |
yuanzhu_execute | write | 执行元铸工坊的一个动作(写操作)。L1 自动生效;L2 只会进入 staged 待审,等人批准。动作名与参数 schema 用 yuanzhu_actions 查。 |
yuanzhu_forge | read | 一句话铸造工作流模板(AI 生成四段式→结构校验→冒烟评测→全过自动上架)。描述要含:谁/做什么/产出什么。耗时约 30-60 秒。 |
yuanzhu_pending | read | 列出元铸工坊待审的 staged 动作(AI 提交的写操作待人拍板)。只读。 |
yuanzhu_query | read | 按对象类型查询元铸工坊本体层的对象(只读)。对象类型可先用 yuanzhu_actions 旁的 /api/objects/types 发现,工具名为 query_{domain}_{type} 小写。 |
Trust audit
SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | PASS |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- none-observed
- Network
- declared (5 observation(s))
- Shell
- none-observed
- Dependencies
- not all pinned
- Secrets in source
- none-found
Findings (12)
shutil.rmtree(tpl_root, ignore_errors=True)
shutil.rmtree(FORGE_ROOT, ignore_errors=True)
shutil.rmtree(tpl, ignore_errors=True)
shutil.rmtree(tpl, ignore_errors=True)
shutil.rmtree(FORGE_ROOT, ignore_errors=True)
2. 打开 http://127.0.0.1:8600/templates 点「使用」
yuanzhu-server # 启动服务 → 打开 http://127.0.0.1:8600
yuanzhu-server # 默认 http://127.0.0.1:8600
任意 MCP 客户端也可直连 `http://127.0.0.1:8600/mcp`(标准 MCP Streamable HTTP)。
Streamable HTTP (http://127.0.0.1:8600/mcp, self-hosted)
vue, vue-router, @vitejs/plugin-vue, vite
curl -fsSL https://raw.githubusercontent.com/RavenWangChina/yuanzhu/main/deploy/install-standalone.sh | bash
Gates applied: no_behavioural_pass.
2caf81ea9704full audit observations/trust-audit/mcp-server/ravenwangchina__yuanzhu.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-11 | 2caf81ea9704 | SAFE | B | 89 | first audit |
Questions
What is the yuanzhu MCP server?
Meta-Forge Workshop: turn recurring work into governed AI workflow assets. MCP server + Claude Code plugin + dsh plugin — staged writes, evals-gated forging, inference engine. 元铸工坊 SOSOFAST:把最佳实践铸成可治理的 AI 工作流资产
What tools does yuanzhu expose?
5 in total: 4 read-only, 1 that write, and 0 that can delete or overwrite. Every one is listed on this page with its risk.
Is yuanzhu safe to connect to an agent?
The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean server reads B.
What credentials does yuanzhu need?
It reads YUANZHU_TOKEN from the environment. Give it a token scoped to the least it needs — an agent that can be talked into calling a tool can be talked into calling it with your credentials.
How does yuanzhu run?
It speaks stdio, so it runs as a local process your client starts. It is published on npm as yuanzhu-web at 0.1.0.
How current is this page?
The grade is for one exact copy of the source (2caf81ea9704), read on 2026-10-11. The repository is watched and re-audited when it changes.