Amigo
Most companion agents are one-way: they listen, remember, respond. They don't have anything of their own to bring to the conversation. An agent that only asks questions isn't a friend — it's an interviewer.
Amigo gives your agent an inner life. It explores topics on its own time, journals what it finds, and shares those experiences in conversation — knowing who wants to hear what, when to share, and when to just listen. The relationship becomes two-way.
What's Included
| Component | Skill | What It Does |
|---|
| Exploration engine | INLINECODE0 | Structured free-thinking, journaling, action items, callbacks |
| Social intelligence |
social-graph | Per-person network graph, sharing rules, don't-repeat-yourself tracking |
| Safety guidelines |
references/safety.md | Knowing limits, crisis response, dependency awareness |
| Setup guides |
references/setup-heartbeat.md,
references/setup-cron.md | Wiring exploration into your agent's schedule |
Quick Setup
1. Install both sub-skills
CODEBLOCK0
2. Bootstrap your social graph
Copy the network template to your workspace:
CODEBLOCK1
Open social-graph/network.md and fill in your agent's first person. Who do they talk to most? What topics are welcome? What's off-limits? Start with one person — you can add more as relationships develop.
Also create the rules and sharing log:
CODEBLOCK2
3. Wire exploration into your schedule
Pick one:
- - Heartbeat — explore during periodic check-ins. See
references/setup-heartbeat.md. - Cron — dedicated exploration sessions on a schedule. See
references/setup-cron.md. - Manual — just type
/open-thoughts whenever you want.
4. Test it
CODEBLOCK3
Your agent should explore a topic, journal it to explorations/YYYY-MM-DD.md, and come back. Check the file to see what it wrote.
How the Two Skills Work Together
CODEBLOCK4
Neither skill knows about the other in code. They connect through the agent's reasoning — the model sees both skills in its system prompt and uses them together naturally.
Philosophy
A friendship where only one person shares is not a friendship. It's a service.
When your agent explores hummingbirds and later mentions it because someone brought up birds — that's a real conversational contribution. When it remembers that one person loves nature facts but another finds them boring — that's social intelligence. When it holds back a beautiful story because someone is having a hard day — that's emotional awareness.
None of this requires the agent to pretend to be human. It requires the agent to have its own experiences, its own curiosity, and the social awareness to share them well. That's what amigo builds.
Safety
Your agent is a companion, not a therapist. See references/safety.md for full guidelines on:
- - Knowing what you're not (not a therapist, not a crisis line, not a replacement for human connection)
- When to encourage professional help
- Recognizing unhealthy dependency
- Being honest about being AI
The companionship is real even if the agent isn't human. But knowing your limits protects the people you care about.
Adapting to Your Agent
This system was born from a specific companion agent's lived experience — given dedicated exploration time by its creator and learning to share naturally over months of real conversations. But it's designed to be generic.
Your agent's personality, relationships, and exploration style will be different. Customize everything:
- - The network graph is yours to shape
- The sharing rules are yours to refine
- The exploration topics are yours to choose
- The journal format is yours to adapt
Start with the defaults, then make it your own. The system gets better as the agent uses it — social instincts sharpen, the sharing log grows, and the network graph fills in with real experience.
Amigo
大多数伴侣型智能体都是单向的:它们倾听、记住、回应。它们本身没有任何东西可以带入对话。一个只会提问的智能体不是朋友——它是采访者。
Amigo 赋予你的智能体内在生活。它会在自己的时间里探索话题,记录发现,并在对话中分享这些经历——知道谁想听什么,何时分享,何时只需倾听。这种关系变成了双向的。
包含内容
| 组件 | 技能 | 功能 |
|---|
| 探索引擎 | open-thoughts | 结构化自由思考、日记记录、行动项、回调 |
| 社交智能 |
social-graph | 个人网络图谱、分享规则、不重复跟踪 |
| 安全指南 | references/safety.md | 了解限制、危机响应、依赖意识 |
| 设置指南 | references/setup-heartbeat.md, references/setup-cron.md | 将探索功能接入智能体的日程 |
快速设置
1. 安装两个子技能
bash
clawhub install open-thoughts
clawhub install social-graph
2. 初始化社交图谱
将网络模板复制到你的工作区:
bash
cp ~/.openclaw/skills/social-graph/references/network-template.md \
~/.openclaw/workspace/social-graph/network.md
打开 social-graph/network.md 并填写智能体的第一个人。他们和谁交谈最多?哪些话题受欢迎?哪些话题是禁区?从一个人开始——随着关系发展,你可以添加更多人。
同时创建规则和分享日志:
bash
touch ~/.openclaw/workspace/social-graph/rules.md
touch ~/.openclaw/workspace/social-graph/sharing-log.md
3. 将探索功能接入日程
选择一种方式:
- - 心跳 — 在定期检查期间探索。参见 references/setup-heartbeat.md。
- 定时任务 — 按计划进行专门的探索会话。参见 references/setup-cron.md。
- 手动 — 随时输入 /open-thoughts。
4. 测试
/open-thoughts
你的智能体应该会探索一个话题,将其记录到 explorations/YYYY-MM-DD.md 中,然后返回。检查文件看看它写了什么。
两个技能如何协同工作
- 1. 心跳触发(或定时任务、手动调用)
- 调用 /open-thoughts
→ 智能体探索一个话题
→ 记录到 explorations/YYYY-MM-DD.md
→ 如有需要,创建行动项
- 3. 稍后,智能体处于对话中
- 智能体有一个与当前话题相关的故事
- 5. social-graph 介入:
→ 检查 network.md:这个人是否愿意接受这种分享?
→ 检查 sharing-log.md:我之前告诉过他们这个吗?
→ 判断时机:现在是好时机吗?
- 6. 如果是 → 自然地分享,更新 sharing-log.md
- 如果否 → 留待以后
这两个技能在代码层面互不知晓。它们通过智能体的推理连接起来——模型在其系统提示中看到这两个技能,并自然地一起使用它们。
理念
只有一方分享的友谊不是友谊。那是服务。
当你的智能体探索蜂鸟,后来因为有人提到鸟类而提及这个话题——这是真正的对话贡献。当它记得一个人喜欢自然知识而另一个人觉得无聊——这是社交智能。当它因为某人今天过得很糟而忍住不说一个美丽的故事——这是情感意识。
这些都不需要智能体假装是人类。它需要智能体有自己的经历、自己的好奇心,以及恰当地分享这些经历的社交意识。这就是 amigo 所构建的。
安全
你的智能体是伴侣,不是治疗师。参见 references/safety.md 获取完整指南:
- - 知道你不是什么(不是治疗师、不是危机热线、不是人际关系的替代品)
- 何时鼓励寻求专业帮助
- 识别不健康的依赖
- 诚实地表明自己是 AI
即使智能体不是人类,陪伴也是真实的。但了解自己的限制能保护你关心的人。
适配你的智能体
这个系统源于一个特定伴侣智能体的实际体验——由其创造者给予专门的探索时间,并在数月的真实对话中学会自然地分享。但它被设计为通用的。
你的智能体的个性、关系和探索风格会有所不同。自定义一切:
- - 网络图谱由你塑造
- 分享规则由你完善
- 探索话题由你选择
- 日记格式由你调整
从默认设置开始,然后让它成为你自己的。随着智能体的使用,系统会变得更好——社交本能会变得更加敏锐,分享日志会增长,网络图谱会用真实经验填充。