Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
wihy avatar

Hermes Agent

  • 552 installs
  • Updated April 12, 2026
  • wihy/hermes-agent-skill

hermes-agent is a Claude Code skill that invokes NousResearch Hermes Agent for developers who need delegated sub-agents with durable memory, MCP access, browser work, and sandboxed code execution.

About

hermes-agent is a Claude Code skill that bridges your primary coding agent to NousResearch Hermes Agent for task delegation beyond a single session's limits. Hermes Agent adds durable memory, dynamic skill creation, MCP server integration, browser automation, and sandboxed code runs as a secondary agent layer. Developers reach for hermes-agent when a coding task needs persistent context across delegations, external tool access through MCP, or isolated execution environments the parent agent cannot safely provide. The skill packages Hermes invocation patterns so the main agent can offload research, browsing, and multi-step workflows without reimplementing agent orchestration.

  • CLI integration for hermes run, delegate, memory, and skills workflows
  • Self-improving skill system that materializes reusable skills from completed tasks
  • Persistent memory with FTS5 search plus LLM summarization
  • Sub-agent delegation for isolated or parallel tasks
  • Bidirectional MCP, browser automation, code execution, and web research in one runtime

Hermes Agent by the numbers

  • 552 all-time installs (skills.sh)
  • +4 installs in the week ending Jul 27, 2026 (Skillselion tracking)
  • Ranked #1,669 of 16,546 AI & Agent Building skills by installs in the Skillselion catalog
  • Security screen: MEDIUM risk (skills.sh audit)
  • Data as of Jul 27, 2026 (Skillselion catalog sync)
npx skills add https://github.com/wihy/hermes-agent-skill --skill hermes-agent

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
Installs552
Security audit1 / 3 scanners passed
Last updatedApril 12, 2026
Repositorywihy/hermes-agent-skill

How do you delegate tasks to Hermes Agent?

Invoke NousResearch Hermes Agent from your coding agent for delegation, durable memory, skill creation, MCP, browser work, and sandboxed code runs.

Who is it for?

Claude Code users who need NousResearch Hermes Agent for delegation, durable memory, MCP tools, browser work, or sandboxed runs.

Skip if: Simple single-shot coding edits that the primary agent can complete without sub-agent delegation or persistent memory.

When should I use this skill?

The user needs Hermes Agent for delegation, durable memory, skill creation, MCP integration, browser automation, or sandboxed execution.

What you get

Hermes Agent delegations, durable memory updates, MCP-connected tool runs, browser session results, and sandboxed execution outputs.

  • Delegated agent task results
  • Persistent memory artifacts
  • Browser and sandbox execution outputs

Files

SKILL.mdMarkdownGitHub ↗

Hermes Agent Skill v2.0

概述

本 Skill 封装了 NousResearch Hermes Agent 的 CLI 调用能力,让 WorkBuddy/Claw 可以通过 Shell 命令利用 Hermes 的核心功能。

v2.0 改进:完全可移植,无硬编码路径,支持任意实例一键安装。

---

首次安装

一键安装(推荐)

当检测到 Hermes 未安装时,运行:

# 安装 Hermes Agent(自动克隆、创建虚拟环境、创建 CLI 入口)
bash ~/.workbuddy/skills/hermes-agent/scripts/install_hermes.sh

# 或自定义安装目录
bash ~/.workbuddy/skills/hermes-agent/scripts/install_hermes.sh --prefix ~/custom/path

安装脚本会自动: 1. ✅ 检测 Python 3.11+ 环境 2. ✅ 克隆 Hermes Agent 源码 3. ✅ 创建 Python 虚拟环境并安装依赖 4. ✅ 创建 ~/.local/bin/hermes CLI 入口 5. ✅ 初始化 ~/.hermes/ 配置目录 6. ✅ 生成默认 .env 配置模板

安装后配置 API Key

# 编辑配置文件,填入你的 API Key
nano ~/.hermes/.env

可选提供商(任选其一):

# 智谱 AI(推荐国内用户)
GLM_API_KEY=your-key-here

# OpenRouter(支持多种模型)
OPENROUTER_API_KEY=sk-or-v1-your-key-here

# Anthropic
ANTHROPIC_API_KEY=sk-ant-your-key-here

# OpenAI
OPENAI_API_KEY=sk-your-key-here

验证安装

# 确认 PATH 包含 hermes
export PATH="$HOME/.local/bin:$PATH"
hermes --version

# 运行诊断
hermes doctor

---

迁移到其他 Claw 实例

将整个 Skill 目录复制到目标实例即可:

# 在目标实例上执行:
cp -r /path/to/hermes-agent ~/.workbuddy/skills/hermes-agent
bash ~/.workbuddy/skills/hermes-agent/scripts/install_hermes.sh
# 然后配置 API Key

---

核心工作流

1. 调用模式速查

场景命令说明
快速问答hermes run "问题" --non-interactive --no-stream最简调用
带上下文hermes run "问题" --context-file ./ctx.md --non-interactive注入项目上下文
子代理委托使用 scripts/hermes_delegate.sh复杂任务分解
技能查询hermes skills list查看已学技能
记忆搜索hermes memory search "关键词"检索历史知识
状态检查hermes statushermes doctor诊断安装状态

2. CLI 命令完整参考

基础命令
# 启动交互式对话
hermes

# 单轮执行(WorkBuddy 集成首选)
hermes run "prompt" [选项]

# 非交互模式选项
--non-interactive    # 关闭交互式 TUI(必需)
--no-stream          # 禁用流式输出,返回完整结果
--context-file PATH  # 注入上下文文件
--toolset NAME       # 限制使用的工具集
--model MODEL        # 指定模型
--timeout SECONDS    # 超时时间(默认300秒)
子代理委托
# 通过 wrapper 脚本调用(推荐)
./scripts/hermes_delegate.sh \
  --task "分析竞品A和B的产品特性" \
  --tools "web_search,browser,file_write" \
  --timeout 300 \
  --output ./result.md

# 直接在 hermes run 中使用 delegate_task 工具
hermes run '使用delegate_task工具,任务是:分析XXX,工具限制:web_search,browser' \
  --non-interactive --no-stream
记忆管理
# 搜索历史记忆
hermes memory search "关键词"

# 查看所有笔记
hermes memory notes list

# 添加手动笔记
hermes memory notes add "重要发现:..."

# 导出/导入记忆
hermes memory export ./backup/
hermes memory import ./backup/
技能管理
hermes skills list                           # 列出所有技能
hermes skills create my-skill --description "描述"  # 创建新技能
hermes skills edit my-skill                  # 编辑技能
hermes skills remove my-skill                # 删除技能
插件管理
hermes plugins list                    # 列出插件
hermes plugins install user/repo       # 安装插件
hermes plugins enable/disable/update/remove plugin-name
定时任务 (Cron)
hermes cron list                       # 列出定时任务
hermes cron add --name "日报" --cron "0 9 * * *" --message "生成总结"
hermes cron pause/resume/remove TASK_ID
MCP 集成
hermes mcp serve --port 8080           # 启动 MCP Server
hermes mcp connect <server-config>     # 连接外部 MCP 服务

---

Wrapper 脚本

scripts/hermes_wrapper.sh

统一的 CLI 封装脚本,提供 JSON 格式化输出和错误处理:

./scripts/hermes_wrapper.sh [命令] [参数...]

# 示例
./scripts/hermes_wrapper.sh run "分析内容" --timeout 60
./scripts/hermes_wrapper.sh memory search "关键词"
./scripts/hermes_wrapper.sh status

输出格式:JSON(包含 success, output, error, duration_ms 字段)

scripts/hermes_delegate.sh

子代理委托专用脚本:

./scripts/hermes_delegate.sh --task "任务描述" [选项]

# 可选选项
--tools "tool1,tool2"      # 限制可用工具集
--timeout 300              # 超时时间(秒)
--output ./result.md        # 输出文件路径
--max-concurrent 3         # 最大并发数(默认3)
--context-file ./ctx.md     # 额外上下文文件
-v                         # 详细输出

scripts/install_hermes.sh

一键安装脚本(详见上方「首次安装」章节):

bash scripts/install_hermes.sh [--skip-deps] [--prefix DIR]

---

模型配置

运行交互式配置向导:

hermes model

或直接编辑 ~/.hermes/config.yaml

model:
  provider: zai           # 可选: openrouter, anthropic, openai, zai, gemini 等
  default: "glm-5"        # 默认模型
  base_url: "https://api.z.ai/api/paas/v4"  # 自定义 API 地址

支持的提供商:openrouter, anthropic, openai, gemini, zai, kimi-coding, nous, custom

---

最佳实践

✅ 推荐做法

1. 始终使用 `--non-interactive --no-stream`:避免 TUI 阻塞 2. 设置合理的超时时间:简单任务 60s,复杂任务 300s 3. 限制工具集:用 --toolset 减少 Token 消耗 4. 使用上下文文件:将大段背景信息放入文件,而非 prompt 中 5. 错误重试机制:网络问题时自动重试 1-2 次

⚠️ 注意事项

1. Token 成本:每次调用都有成本 2. 并发限制:最多 3 个并发子代理 3. 超时保护:长时间运行的任务必须设置 timeout 4. API Key 安全:不要在 Skill 文件中硬编码密钥 5. Python 版本:确保使用 Python 3.11+

---

故障排除

问题解决方案
command not found: hermes运行 export PATH="$HOME/.local/bin:$PATH" 或重新执行 install_hermes.sh
TypeError: unsupported operand确保 Python 3.11+
API Key 错误检查 ~/.hermes/.env 配置
连接超时检查网络,或更换 LLM 提供商
子代理失败减少 --max-concurrent 或增加 --timeout
安装脚本失败运行 hermes doctor 诊断

---

文件结构

hermes-agent/
├── SKILL.md                    # 本文件(Skill 说明文档)
├── _meta.json                  # Skill 元数据(可移植性声明)
├── scripts/
│   ├── install_hermes.sh       # 一键安装脚本(通用)
│   ├── hermes_wrapper.sh       # 统一 CLI 封装(动态路径检测)
│   └── hermes_delegate.sh      # 子代理委托脚本(动态路径检测)
└── references/                 # 参考文档

---

更新日志

  • v2.0.0 (2026-04-12): 完全可移植版 — 移除所有硬编码路径,添加一键安装脚本,支持任意 Claw 实例迁移
  • v1.0.0 (2026-04-11): 初始版本,支持基础 CLI 调用、子代理委托、记忆/技能管理

Related skills

How it compares

Choose hermes-agent when sub-agent delegation with durable memory beats doing everything in the primary Claude Code session.

FAQ

What is Hermes Agent in hermes-agent skill?

Hermes Agent is NousResearch's agent runtime that hermes-agent invokes from Claude Code. It handles delegation, durable memory, MCP connections, browser work, and sandboxed code runs as a secondary agent layer.

When should developers use hermes-agent?

hermes-agent fits multi-step tasks needing persistent memory, external MCP tools, browser automation, or isolated code execution that exceed a single Claude Code session's scope.

Is Hermes Agent safe to install?

skills.sh reports 1 of 3 security scanners passed. Review the Security Audits panel on this page before installing in production.

AI & Agent Buildingagentsautomation

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.