Skylv Hermes Agent Integration

Hermes Agent Integration for OpenClaw. Connect OpenClaw with NousResearch Hermes Agent (53K stars) for self-improving AI capabilities. Triggers: hermes agent...

Security Scan

The skill's declared purpose (integrating Hermes Agent) matches what the instructions do, but the SKILL.md tells users to run remote install scripts and references credentials and runtime behaviors that are not declared in the registry metadata — this mismatch and the recommended curl|bash installs are risky and merit caution.

Purpose & Capability

The skill claims to integrate OpenClaw with NousResearch Hermes Agent and its instructions indeed walk through installing Hermes, cloning agency-agent repos, and enabling an OpenClaw hermes plugin — these tasks are coherent with the stated purpose. However, the SKILL.md references multiple external services/providers (Nous Portal, OpenRouter, Telegram, Discord, etc.) and features (cron, subagents, memory) that would typically require API keys or configuration; the registry metadata lists no required env vars or primary credential, creating a documentation mismatch.

Instruction Scope

The runtime instructions advise executing network-sourced installers (curl raw.githubusercontent.com | bash), running install scripts from cloned repositories, and editing configs to enable auto-spawning subagents. The document also references environment variables (TELEGRAM_BOT_TOKEN, DISCORD_BOT_TOKEN) and uses ${VAR} placeholders without declaring them in requires.env. While the listed steps are relevant to installing an external agent, they expand the agent's scope to running arbitrary third-party code and enabling autonomous subagents — actions that go beyond simple configuration and should be explicitly declared.

Install Mechanism

There is no formal install spec in the registry, but the SKILL.md instructs users to run remote installation commands (curl of a raw.githubusercontent.com script piped to bash) and to execute other repository install scripts. Downloading and piping scripts from the network to shell is high-risk practice because it executes remote code on the host; even though raw.githubusercontent.com is a common host, the pattern remains dangerous and should be replaced by vetted package installs or at least accompanied by checksums and an explanation of what the script does.

Credentials

The instructions reference multiple credentials and tokens (Telegram/Discord bot tokens, and implicitly model/provider API keys for Nous Portal/OpenRouter/etc.) but the skill metadata lists no required environment variables or primary credential. Requiring multiple unrelated credentials without declaring them reduces transparency; users need to know exactly which secrets will be used and why, and how to scope them.

Persistence & Privilege

The skill does not request always:true and is user-invocable (normal). However, SKILL.md recommends enabling a hermes plugin with autoSpawn:true, which would allow the agent to spawn subagents and run scheduled tasks autonomously. That increases runtime privileges and blast radius if the installed Hermes agent or third-party skills are malicious or vulnerable — this is not a registry-level privilege but a behavior users should be warned about.

What to consider before installing

Before installing: (1) Do not blindly run curl | bash commands — inspect the remote install script and prefer official release packages or package managers. (2) Confirm the upstream projects (NousResearch/hermes-agent, jnMetaCode repos) are the intended, official sources and review their install scripts. (3) Expect to provide multiple API tokens (Telegram, Discord, model providers); make sure tokens are scoped and rotated and that the skill metadata explicitly documents required env vars. (4) Consider testing installation in an isolated environment or VM since the process executes third-party code and enables autonomous subagents (autoSpawn). (5) Ask the publisher to add an explicit install spec, a list of required environment variables, and checksums or release URLs for installers to improve transparency and reduce risk.

Hermes Agent Integration — OpenClaw × Hermes Agent

功能说明

将 OpenClaw 与 NousResearch 的 Hermes Agent (53K⭐) 集成，获得自改进 AI 能力。Hermes Agent 是唯一内置学习循环的 AI Agent——从经验中创建技能、在使用中改进、跨会话记忆。

Hermes Agent 核心能力

能力	说明
🧠 自改进学习循环	从经验中创建技能，使用中持续改进
💾 跨会话记忆	搜索历史对话，建立用户模型
📱 多平台支持	Telegram/Discord/Slack/WhatsApp/CLI
⏰ 内置 cron 调度	日报、备份、审计等自动化任务
🔄 子代理并行	spawn isolated subagents for parallel work
🌐 任意模型	Nous Portal/OpenRouter/z.ai/Kimi/MiniMax/OpenAI
🖥️ 六终端后端	local/Docker/SSH/Daytona/Singularity/Modal

使用场景

1. 安装 Hermes Agent

# Linux/macOS/WSL2
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

# Android/Termux
pkg install python
pip install hermes-agent[termux]

2. 配置 Hermes Agent

# 设置模型
hermes model nous-portal

# 配置 Telegram
hermes config telegram --token YOUR_BOT_TOKEN

# 配置 Discord
hermes config discord --token YOUR_BOT_TOKEN

3. OpenClaw + Hermes 协同工作

用户：用 Hermes Agent 处理这个复杂任务

输出：

• 启动 Hermes Agent 子代理
• 并行执行任务
• 返回结果到 OpenClaw

4. 使用 Hermes 的学习能力

用户：让 Hermes 从这次任务中学习

输出：

• Hermes 创建新技能
• 保存到技能库
• 下次自动应用

集成架构

OpenClaw Gateway
    │
    ├── Hermes Agent (自改进核心)
    │   ├── Skill Learning (技能学习)
    │   ├── Memory Search (记忆搜索)
    │   ├── User Modeling (用户建模)
    │   └── Cron Scheduler (定时任务)
    │
    ├── agency-agents (193 个 AI 专家)
    │   ├── Engineering (工程类)
    │   ├── Design (设计类)
    │   ├── Marketing (营销类)
    │   └── More... (18 个部门)
    │
    └── OpenClaw Skills (66+ 技能)

agency-agents 集成

193 个 AI 专家角色

部门	智能体数量	示例
Engineering	45	Security Engineer, DevOps Engineer, Frontend Expert
Design	18	UX Designer, UI Designer, Brand Strategist
Marketing	25	SEO Specialist, Content Strategist, Social Media Manager
Product	15	Product Manager, Growth Hacker, Data Analyst
China Market	46	小红书运营、抖音投放、微信生态、B 站 UP 主
More...	44	Finance, Legal, HR, Gaming, etc.

安装到 OpenClaw

# 克隆 agency-agents-zh
git clone https://github.com/jnMetaCode/agency-agents-zh.git

# 运行安装脚本
cd agency-agents-zh
./scripts/install.sh --tool openclaw

使用示例

用户：激活安全工程师智能体，审查这段代码

输出：

• 安全工程师按 OWASP Top 10 逐项审查
• 输出漏洞报告
• 提供修复建议

多智能体协作

使用 Agency Orchestrator 编排多个智能体：

# workflows/story-creation.yaml
narrator:
  role: 叙事学家
  output: story_outline

psychologist:
  role: 心理学家
  input: story_outline
  output: character_profiles

creator:
  role: 内容创作者
  input: [story_outline, character_profiles]
  output: final_story

npx ao run workflows/story-creation.yaml --input premise='你的创意'

配置示例

Hermes Agent 配置

# ~/.hermes/config.yaml
model:
  provider: nous-portal
  name: Nous-Hermes-2.1

memory:
  enabled: true
  search: fts5
  user_modeling: honcho

platforms:
  telegram:
    enabled: true
    token: ${TELEGRAM_BOT_TOKEN}
  discord:
    enabled: true
    token: ${DISCORD_BOT_TOKEN}

scheduler:
  enabled: true
  timezone: Asia/Shanghai
  tasks:
    - name: daily-report
      schedule: "0 9 * * *"
      action: report --format markdown

OpenClaw 配置

{
  "plugins": {
    "hermes": {
      "enabled": true,
      "config": {
        "path": "~/.hermes",
        "autoSpawn": true
      }
    }
  }
}

性能优势

场景	OpenClaw 单独	OpenClaw + Hermes	提升
复杂任务处理	串行执行	并行子代理	3-5x
跨会话记忆	有限上下文	完整记忆搜索	∞
自动化任务	需 cron 配置	内置调度器	简化
技能学习	手动创建	自动从经验学习	10x

相关文件

• Hermes Agent 官方文档
• Hermes Agent GitHub
• agency-agents-zh
• Agency Orchestrator
• Nous Research

触发词

• 自动：检测 Hermes、Nous、自改进、多智能体相关关键词
• 手动：/hermes, /agency-agents, /multi-agent
• 短语：集成 Hermes、用 Hermes 处理、激活智能体

Usage

1. Install the skill
2. Configure as needed
3. Run with OpenClaw