Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

Oclaw Hermes

v3.0.0

OpenClaw × Hermes × DeerFlow 三位一体智能体桥接方案 - 实现 mflow 记忆流同步、智能体集群协作、深度研究链和专家蒸馏。让 OpenClaw、Hermes、DeerFlow 的边界,成为你能力的延伸。

0· 108·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for ruiyongwang/oclaw-hermes.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Oclaw Hermes" (ruiyongwang/oclaw-hermes) from ClawHub.
Skill page: https://clawhub.ai/ruiyongwang/oclaw-hermes
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install oclaw-hermes

ClawHub CLI

Package manager switcher

npx clawhub@latest install oclaw-hermes
Security Scan
Capability signals
Crypto
These labels describe what authority the skill may exercise. They are separate from suspicious or malicious moderation verdicts.
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Suspicious
medium confidence
!
Purpose & Capability
The skill claims to bridge OpenClaw, Hermes, and DeerFlow (expected to need local services, containers, and tokens), but the registry metadata lists no required environment variables or config paths. In reality SKILL.md, docker-compose.yml, and multiple scripts reference and require tokens/URLs (OPENCLAW_TOKEN, OPENROUTER_API_KEY, ANTHROPIC_API_KEY, DEERFLOW_*), and write to home config directories (~/.openclaw, ~/.hermes, ~/.oclaw-hermes). The omission in metadata is an incoherence: either the metadata is incomplete or the skill is hiding required privileges.
Instruction Scope
SKILL.md explicitly instructs cloning a GitHub repo, creating a .env with API keys, running docker-compose to pull/run multiple containers, and running Python scripts. These instructions will create persistent DBs and memory files and may push data to OpenClaw/Hermes/DeerFlow endpoints. The actions (running containers, creating files under user home, requiring tokens) are coherent for a bridge/orchestrator but grant broad local persistence and network access; the SKILL.md also references commands (e.g., python scripts/verify.py) not present in the manifest, which is a minor inconsistency.
Install Mechanism
No formal install spec in the registry (instruction-only), but the project includes docker-compose that will pull images (nousresearch/hermes, bytedance/deerflow) and build local Dockerfiles. There are no obscure download URLs in the package itself, but running docker-compose will fetch external container images—this is expected for this type of project but increases risk surface (third-party images, network pulls).
!
Credentials
Registry metadata claims no required env vars, yet SKILL.md, docker-compose.yml, and scripts require multiple credentials and service URLs (OPENCLAW_TOKEN, OPENROUTER_API_KEY, ANTHROPIC_API_KEY, DEERFLOW_GATEWAY_URL, DEERFLOW_LANGGRAPH_URL, etc.). The code also reads/writes to local config paths and persistent DBs under user home. Requesting multiple unrelated model-provider keys and platform tokens is reasonable for a multi-platform bridge, but the mismatch with declared requirements and lack of explicit justification is a red flag.
Persistence & Privilege
The skill will create persistent artifacts (SQLite DBs in ~/.openclaw/.oclaw-hermes, files under ~/.hermes, Docker containers and volumes) and can sync memories across platforms. It does not set always:true, but it does request persistent local storage and may auto-register/publish skills per config options. Persistence and container orchestration are expected for this use case but increase blast radius; verify storage locations and retention policy before use.
What to consider before installing
This package appears to be a substantial bridge/orchestration project that will create local databases, write files under your home directory, and run Docker containers that connect to local and remote services. However, the registry metadata incorrectly claims no required environment variables or config paths while the SKILL.md, docker-compose.yml, and scripts clearly require multiple API keys and will persist memory data. Before installing: 1) do not supply high-privilege tokens (e.g., full cloud credentials); create least-privilege/test tokens instead; 2) review the .env.example and every script for endpoints and persistence locations (~/.openclaw, ~/.hermes, ~/.oclaw-hermes); 3) inspect the Docker images (nousresearch/hermes, bytedance/deerflow) you will pull—prefer pinned image digests and official images; 4) run in an isolated environment or VM first (to avoid leaking secrets or contaminating your real home config); 5) confirm the repository source and maintainership (the registry lists source unknown but SKILL.md references a GitHub repo); and 6) if you need this functionality, ask the author to correct the registry metadata (declare required env vars and config paths) and to provide a minimal test mode that doesn't auto-persist or auto-publish. The mismatches raise enough concern to pause installation until you verify secrets, image provenance, and the exact runtime behavior.

Like a lobster shell, security has layers — review code before you run it.

agent-bridgevk978rs8nh468zzc1ejm15v0n5184n0eyauto-memoryvk978rs8nh468zzc1ejm15v0n5184n0eydeerflowvk978rs8nh468zzc1ejm15v0n5184n0eyexpert-distillationvk978rs8nh468zzc1ejm15v0n5184n0eygraph-memoryvk978rs8nh468zzc1ejm15v0n5184n0eyhermesvk978rs8nh468zzc1ejm15v0n5184n0eylatestvk978rs8nh468zzc1ejm15v0n5184n0eymemory-syncvk978rs8nh468zzc1ejm15v0n5184n0eymflowvk978rs8nh468zzc1ejm15v0n5184n0eymulti-agentvk978rs8nh468zzc1ejm15v0n5184n0eyopenclawvk978rs8nh468zzc1ejm15v0n5184n0eyresearchvk978rs8nh468zzc1ejm15v0n5184n0eyunified-corevk978rs8nh468zzc1ejm15v0n5184n0ey
108downloads
0stars
2versions
Updated 2w ago
v3.0.0
MIT-0

oclaw-hermes v3.0

OpenClaw × Hermes × DeerFlow 深度融合智能体架构

Agent 的边界,就是世界的边界。

简介

oclaw-hermes 是专为 OpenClaw 平台打造的 下一代智能体架构,深度融合三大核心系统:

  • Hermes 五层自进化记忆 (mflow v2.0)
  • DeerFlow 六大智能体协作 (LangGraph编排)
  • OpenClaw 58+ Skills 生态

实现记忆驱动的智能体路由、Skill-记忆双向增强、跨平台状态同步的完整闭环。

v3.0 核心创新

特性说明
统一核心UnifiedCore 深度融合三大系统
记忆驱动路由基于记忆上下文智能选择智能体
Skill-记忆双向增强Skill调用自动沉淀为记忆,记忆指导Skill选择
六智能体协作Lead/Research/Code/Browser/Memory/Skill
自动记忆提取会话内容实时提取事实,自动分层存储
图记忆推理实体关系网络支持复杂推理查询
跨平台同步OpenClaw/Hermes/DeerFlow 三端记忆实时同步

核心能力矩阵

维度OpenClawHermesDeerFlowoclaw-hermes
技能生态✅ 58+ Skills✅ 技能自动创建✅ 技能管理 API✅ 三端同步
记忆系统❌ 会话隔离✅ 三层记忆✅ 线程持久化✅ mflow 记忆流
多智能体❌ 单 Agent⚠️ 子 Agent 委托✅ LangGraph 编排✅ 智能体集群
研究能力❌ 无⚠️ 基础搜索✅ 深度研究链✅ 一键研究
代码执行✅ 本地执行✅ 多后端✅ 沙箱执行✅ 安全隔离
浏览器❌ 无✅ 浏览器工具✅ 自动化浏览✅ 视觉感知
部署方式云端本地/DockerDocker/本地全场景覆盖

核心特性

1. 三位一体架构

┌─────────────────────────────────────────────────────────────┐
│                      oclaw-hermes                           │
│              (OpenClaw × Hermes × DeerFlow)                 │
├───────────────┬───────────────────┬─────────────────────────┤
│   OpenClaw    │      Hermes       │       DeerFlow          │
│   技能生态    │    自进化记忆     │    多智能体协作         │
├───────────────┼───────────────────┼─────────────────────────┤
│ • 58+ Skills  │ • 三层记忆体系    │ • LangGraph 编排        │
│ • 工程咨询    │ • 技能自动创建    │ • 子智能体集群          │
│ • 大师系列    │ • 长期记忆        │ • 深度研究链            │
│ • 专业工具    │ • 多平台网关      │ • 代码沙箱              │
└───────────────┴───────────────────┴─────────────────────────┘
                              │
                              ▼
                    ┌─────────────────┐
                    │   mflow 记忆流   │
                    │  (记忆同步中枢)  │
                    └─────────────────┘

2. mflow 记忆流系统

独创的 记忆流 (Memory Flow) 架构,实现三端记忆无缝同步:

mflow:
  layers:
    - layer_1: 即时记忆      # 当前会话上下文
    - layer_2: 短期记忆      # 最近 7 天会话
    - layer_3: 长期记忆      # 持久化知识库
    - layer_4: 技能记忆      # Skill 使用经验
    - layer_5: 专家记忆      # 蒸馏的专家思维
  
  sync:
    openclaw: 实时同步 Skill 调用记录
    hermes:   双向同步记忆状态
    deerflow: 线程级记忆持久化
  
  bridge:
    - 会话状态桥接
    - 技能调用桥接
    - 记忆检索桥接
    - 专家思维桥接

3. 智能体集群

基于 DeerFlow 的 LangGraph 多智能体编排

智能体职责触发条件
Lead Agent意图识别、任务分发所有请求
Research Agent深度研究、信息收集研究类任务
Code Agent代码生成、执行、调试编程类任务
Browser Agent网页浏览、数据提取需要外部信息
Skill AgentSkill 调用、管理需要专业技能
Memory Agent记忆检索、更新需要上下文
Expert Agent专家思维蒸馏需要专家视角

4. 深度研究链

继承 DeerFlow 的研究能力,实现 一键深度研究

用户请求
    │
    ▼
┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│  问题分解   │───▶│  多源搜索   │───▶│  信息验证   │
└─────────────┘    └─────────────┘    └─────────────┘
                                              │
    ┌─────────────────────────────────────────┘
    ▼
┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│  综合分析   │───▶│  报告生成   │───▶│  技能沉淀   │
└─────────────┘    └─────────────┘    └─────────────┘

部署指南

前置要求

  • Docker & Docker Compose
  • Python 3.10+
  • Node.js 18+
  • OpenClaw CLI (openclawmp)

快速部署

# 1. 克隆仓库
git clone https://github.com/ruiyongwang/oclaw-hermes.git
cd oclaw-hermes

# 2. 配置环境变量
cp .env.example .env
# 编辑 .env 填入 API Keys

# 3. Docker 启动三件套
docker-compose up -d

# 4. 验证部署
python scripts/verify.py

手动部署

# 1. 安装 Hermes Agent
pip install hermes-agent

# 2. 安装 DeerFlow
git clone https://github.com/bytedance/deerflow.git
cd deerflow && docker-compose up -d

# 3. 配置 OpenClaw Bridge
openclawmp login --token YOUR_TOKEN

# 4. 启动桥接服务
python scripts/bridge.py --mode full

配置详解

主配置 (config.yaml)

# oclaw-hermes 核心配置
version: "2.0.0"

# 平台连接
platforms:
  openclaw:
    endpoint: "https://openclawmp.stepfun.com"
    token: "${OPENCLAW_TOKEN}"
    skills_path: "~/.openclaw/skills"
  
  hermes:
    endpoint: "http://localhost:8080"
    memory_path: "~/.hermes/memories"
    config_path: "~/.hermes/config.yaml"
  
  deerflow:
    endpoint: "http://localhost:2026"
    gateway_url: "${DEERFLOW_GATEWAY_URL}"
    langgraph_url: "${DEERFLOW_LANGGRAPH_URL}"

# mflow 记忆流
mflow:
  enabled: true
  sync_interval: 300  # 秒
  layers:
    - name: "instant"
      ttl: 3600        # 1小时
    - name: "short"
      ttl: 604800      # 7天
    - name: "long"
      persistent: true
    - name: "skill"
      persistent: true
    - name: "expert"
      persistent: true

# 智能体集群
agents:
  lead:
    model: "anthropic/claude-3.7-sonnet"
    mode: "ultra"  # flash/standard/pro/ultra
  
  research:
    enabled: true
    max_depth: 5
  
  code:
    enabled: true
    sandbox: "docker"
  
  browser:
    enabled: true
    headless: true
  
  skill:
    enabled: true
    auto_register: true
  
  memory:
    enabled: true
    fts_index: true
  
  expert:
    enabled: true
    sources: 6        # 六路采集
    validation: 3     # 三重验证

# 技能创造者
skill_creator:
  enabled: true
  expert_system: "~/.workbuddy/skills/dlh365-expert-system"
  output_path: "~/.hermes/skills"
  auto_publish: false  # 自动发布到 OpenClaw

环境变量 (.env)

# OpenClaw
OPENCLAW_TOKEN=sk_xxxxxxxx

# Hermes
HERMES_MODEL_PROVIDER=openrouter
HERMES_MODEL=anthropic/claude-3.7-sonnet

# DeerFlow
DEERFLOW_URL=http://localhost:2026
DEERFLOW_GATEWAY_URL=http://localhost:2026
DEERFLOW_LANGGRAPH_URL=http://localhost:2026/api/langgraph

# LLM API Keys
OPENROUTER_API_KEY=sk-or-xxx
ANTHROPIC_API_KEY=sk-ant-xxx
OPENAI_API_KEY=sk-xxx

# 可选:其他配置
LOG_LEVEL=INFO
DEBUG=false

使用方式

命令行

# 启动完整服务
oclaw-hermes start

# 发送消息
oclaw-hermes chat "分析当前建筑行业趋势"

# 深度研究
oclaw-hermes research "中国装配式建筑发展现状"

# 蒸馏专家
oclaw-hermes distill "曹德旺"

# 同步记忆
oclaw-hermes sync --full

# 查看状态
oclaw-hermes status

Python API

from oclaw_hermes import OclawHermes

# 初始化
client = OclawHermes()

# 发送消息(自动路由到最优智能体)
response = client.chat(
    message="帮我写一个 Python 爬虫",
    mode="ultra",  # flash/standard/pro/ultra
    context={
        "skills": ["regex-generator", "excel-formula"],
        "experts": ["elon-musk", "buffett"]
    }
)

# 深度研究
report = client.research(
    topic="新能源汽车电池技术",
    depth=5,
    output_format="markdown"
)

# 蒸馏专家
skill_path = client.distill(
    expert_name="段永平",
    sources=["books", "interviews", "social"],
    output_dir="~/my-skills/"
)

# 记忆同步
client.sync_memories(direction="bidirectional")

在 OpenClaw 中使用

# 安装 oclaw-hermes Skill
openclawmp install oclaw-hermes

# 使用
> 启动 hermes 桥接
> 用 deerflow 研究一下量子计算
> 蒸馏一个马斯克视角
> 同步所有记忆

核心脚本

脚本功能
bridge.py三平台桥接服务
mflow.py记忆流同步引擎
distill.py专家蒸馏器
research.py深度研究链
verify.py部署验证
chat.py交互式对话

架构图

┌─────────────────────────────────────────────────────────────────────┐
│                         用户交互层                                   │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐ │
│  │  命令行 CLI  │  │  Python API │  │ OpenClaw   │  │  Web UI     │ │
│  └──────┬──────┘  └──────┬──────┘  └──────┬──────┘  └──────┬──────┘ │
└─────────┼────────────────┼────────────────┼────────────────┼────────┘
          │                │                │                │
          └────────────────┴────────────────┴────────────────┘
                                   │
                                   ▼
┌─────────────────────────────────────────────────────────────────────┐
│                      oclaw-hermes 核心层                            │
│  ┌─────────────────────────────────────────────────────────────┐   │
│  │                    智能体路由器                              │   │
│  │         (意图识别 → 智能体选择 → 任务分发)                   │   │
│  └─────────────────────────────────────────────────────────────┘   │
│                              │                                      │
│  ┌─────────────┬─────────────┼─────────────┬─────────────┐         │
│  │  Lead Agent │Research Agent│  Code Agent │Browser Agent│         │
│  └──────┬──────┴──────┬──────┴──────┬──────┴──────┬──────┘         │
│         │             │             │             │                │
│  ┌──────┴──────┐ ┌────┴────┐  ┌────┴────┐  ┌────┴────┐            │
│  │  Skill Agent │ │Memory Agent│ │Expert Agent│                    │
│  └──────┬──────┘ └────┬────┘  └────┬────┘  └────┬────┘            │
│         │             │             │             │                │
│         └─────────────┴─────────────┴─────────────┘                │
│                              │                                      │
│  ┌───────────────────────────┴───────────────────────────┐         │
│  │                    mflow 记忆流                        │         │
│  │  ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐     │         │
│  │  │ 即时记忆 │ │短期记忆 │ │长期记忆 │ │技能记忆 │     │         │
│  │  └────┬────┘ └────┬────┘ └────┬────┘ └────┬────┘     │         │
│  │       └───────────┴───────────┴───────────┘          │         │
│  │                         │                            │         │
│  │                    ┌────┴────┐                       │         │
│  │                    │专家记忆 │                       │         │
│  │                    └─────────┘                       │         │
│  └──────────────────────────────────────────────────────┘         │
└─────────────────────────────────────────────────────────────────────┘
                                   │
          ┌────────────────────────┼────────────────────────┐
          │                        │                        │
          ▼                        ▼                        ▼
┌─────────────────┐      ┌─────────────────┐      ┌─────────────────┐
│    OpenClaw     │      │     Hermes      │      │    DeerFlow     │
│   技能生态系统   │      │   自进化记忆    │      │  多智能体协作   │
│                 │      │                 │      │                 │
│ • 58+ Skills    │      │ • 三层记忆      │      │ • LangGraph     │
│ • 工程咨询      │      │ • 技能创建      │      │ • 研究链        │
│ • 大师系列      │      │ • 多平台网关    │      │ • 代码沙箱      │
│ • 专业工具      │      │ • 40+ 工具      │      │ • 浏览器自动化  │
└─────────────────┘      └─────────────────┘      └─────────────────┘

与原版 Hermes 的差异

特性Hermesoclaw-hermes
目标平台通用OpenClaw 专属优化
记忆同步单端三端同步 (mflow)
多智能体子 Agent 委托DeerFlow 集群
研究能力基础深度研究链
技能生态自建OpenClaw 58+ Skills
专家蒸馏集成度量衡专家系统
部署方式多种Docker Compose 一键

贡献

欢迎提交 Issue 和 PR!

许可

MIT License

致谢


Agent 的边界,就是世界的边界。

让 OpenClaw、Hermes、DeerFlow 的边界,成为你能力的延伸。

Comments

Loading comments...