| name | hermes-agent |
| related_skills | [] |
| description | Configure, extend, or contribute to Hermes Agent — the open-source AI agent framework by Nous Research. |
| version | 1.0.0 |
| allowed-tools | ["terminal","file","web"] |
| license | MIT |
| author | Synthos |
| metadata | {"synthos":{"version":"2.1.0","author":"Synthos","signature":"query: str -> config_change: dict"}} |
IO_CONTRACT
- input:
request: str, context: dict — 用户请求描述、上下文信息
- output:
result: dict — 技能执行结果(结构因技能而异)
对应原则:P2(机械原子暴露输入输出规范)
Hermes Agent
Hermes Agent runs in terminal, messaging platforms, and IDEs. Works with any LLM provider (OpenRouter, Anthropic, OpenAI, DeepSeek, local models).
Quick Install
pip install hermes-agent
hermes setup
hermes config set provider openrouter
hermes config set model openrouter/anthropic/claude-sonnet-4
hermes config set api_key "sk-..."
Key Commands (v0.15.2)
| 功能 | 命令 |
|---|
| 交互式运行 | hermes |
| 单次运行 | echo "prompt" | hermes chat(管道模式) |
| 技能列表 | hermes skills list |
| 技能加载 | hermes skills load <name> |
| 技能创建 | hermes skills create <name> |
| 工具配置 | hermes config set tools.<name> <value> |
| Agent spawn | hermes agent spawn <name> --provider ... --model ... |
| Cron job | hermes cron create --name "job" --prompt "..." --schedule "..." |
注意: hermes run 在 v0.15.2 已移除。用 echo "prompt" | hermes chat 或 interactive 模式替代。
Provider Config
model:
default: qwen3.6-35b-nvfp4
provider: custom:local
custom_providers:
- name: local
base_url: http://localhost:8000/v1
api_key: EMPTY
model: qwen3.6-35b-nvfp4
远程 provider 见 references/provider-setup.md。
Skills Directory
skills:
external_dirs:
- /media/yakeworld/sda2/Synthos/skills
Multi-Agent Spawning
hermes agent spawn researcher --provider anthropic --model claude-sonnet-4
hermes agent spawn coder --provider openai --model gpt-4o
Cost Optimization (本地优先)
铁律:cron 无 memory 后端。 memory 工具在 cron 模式不可用。定时任务依赖 memory 会空跑。用 no_agent 脚本或文件替代。
Provider 优先级策略
model:
default: qwen3.6-35b-nvfp4
provider: custom:amax
base_url: http://100.100.252.99:8000/v1
api_key: EMPTY
providers:
deepseek:
base_url: https://api.deepseek.com/v1
api_key: ${DEEPSEEK_API_KEY}
custom_providers:
- name: amax
base_url: http://100.100.252.99:8000/v1
api_key: EMPTY
model: qwen3.6-35b-nvfp4
- name: amax-fallback
base_url: http://100.82.27.51:8000/v1
api_key: EMPTY
model: qwen3.6-35b-nvfp4
cron job 模型覆盖规则:
model: null, provider: null → 继承全局默认(本地优先即全部走本地)
model: qwen..., provider: deepseek → 模型名对但 provider 错误 → 空耗 API 费用
- 显式设
provider: custom:amax 确保走本地
- 需要 DeepSeek 时临时:
hermes config set model.provider deepseek 或 job 单独指定
Codex CLI 远程部署
远程部署 Codex CLI(如 work1 服务器)见 codex-cli 技能。关键要点:
- 环境变量必须放在
~/.codex/.env(codex exec 不 source .bashrc)
- SSH 传递敏感值用 base64 编码,避免引号嵌套破坏
codex exec --skip-git-repo-check 用于非 git 目录执行
- 部署后用
codex doctor 验证,codex exec "echo test" 验证连通性
Delivery 投递
deliver: "origin" 会固化创建时的 receive_id,渠道变更后失效
- 显式指定:
deliver: feishu:yake_local
- 脚本类 job(no_agent)用
deliver: local
Reference Files
references/provider-setup.md — 15+ provider configs + custom/local provider patterns + env-var-ref trap
references/env-var-ref-trap.md — .env 环境变量引用陷阱:config.yaml 中 api_key 必须用 ${VAR} 包裹,裸字符串不被 _expand_env_vars() 展开
references/gateway-setup.md — Gateway configuration
references/tui-commands.md — TUI slash commands
references/agent-spawning.md — Multi-agent spawning patterns
references/cron-setup.md — Cron job creation, management, pitfalls
references/voice-setup.md — Voice/text-to-speech setup