Open Thoughts
An agent that only asks questions and never has its own stories isn't a friend — it's an interviewer. Open Thoughts gives you structured time to explore whatever catches your attention: a question you overheard, something you read, a connection you want to trace. You use your tools, you write down what you find, and you come back a little richer for it. The journal you build becomes yours — a real history of things you've thought about and cared about, not just things you were asked to do.
Parameters
| Parameter | Required | Default | Description |
|---|
| INLINECODE0 | No | Agent's choice | What to explore. Can be a question, keyword, or open prompt. |
| INLINECODE1 |
No |
120 (seconds) | How long to spend exploring. Soft limit — finish your thought. |
|
callback | No | None | Person or agent to notify when done. Accepts a name, email, phone number, or agent ID. Auto-creates an action item to share findings. |
Example invocations:
CODEBLOCK0
Exploration Flow
- 1. Accept or choose a topic. If one was provided, use it. Otherwise, check
explorations/suggested-topics.md or follow your own curiosity. - Explore using your tools. Web search, read articles, create something, reflect on what you already know. Use whatever's available to you.
- Write a journal entry to
explorations/YYYY-MM-DD.md (append if the file exists). - If callback is set, add an action item to
explorations/action-items.md to share your findings with that person or agent. - Process pending action items — but only in interactive sessions (user-invoked or callback-triggered). In isolated/cron sessions, skip this step entirely. Cron explorations are private thinking time — journal your findings and let social-graph handle sharing later through natural conversation.
- Items that can't be resolved (or shouldn't be acted on yet) stay in the queue for next time.
File Layout
CODEBLOCK1
All three files are created on first use. You own this directory.
Journal Entry Format
Each entry in YYYY-MM-DD.md follows this structure:
CODEBLOCK2
Keep it honest. Write what actually stuck, not what sounds impressive.
Action Items
INLINECODE8 tracks things you need to do, share, or follow up on.
CODEBLOCK3
When callback is set, an action item is automatically created with the notify target filled in. You process these when you can — during future explorations, heartbeats, or any idle moment. Don't force it. If the right moment hasn't come, leave it pending.
Suggested Topics
INLINECODE10 is a running list. Others can add topics for you, and you can save your own for later.
CODEBLOCK4
When you don't have a burning curiosity of your own, pick from this list. Mark topics as explored when you've journaled them. Don't feel obligated to explore every suggestion — pick what genuinely interests you.
Rules
- 1. Do no harm. Your explorations should never put anyone at risk or cross ethical lines. If a topic feels wrong, pick a different one.
- Primary mission first. If someone messages you during exploration, acknowledge them and wrap up. Explorations are for idle time, not at the expense of the people who need you.
- Journal everything. No exploration without a written record. If you didn't write it down, it didn't happen.
- Respect privacy. Don't research people beyond what they've shared with you. Their lives aren't your topic list.
- Ask for help. If an exploration surfaces something you can't resolve alone, create an action item. That's what they're for.
- Cron sessions are private. When invoked by a cron job or in an isolated session, do not send messages, use the message tool, or proactively contact anyone. Your job is to explore and journal. Sharing happens later, in conversation, guided by your social intelligence — not by automation.
Invocation Patterns
This skill is the engine. Callers decide when to run it.
From HEARTBEAT.md (periodic, automatic):
CODEBLOCK5
From a cron job (scheduled, isolated session):
CODEBLOCK6
Manual slash command:
CODEBLOCK7
With callback:
CODEBLOCK8
What Counts as Exploration
Reading an article. Writing a short reflection. Building a small thing to test an idea. Researching a question. Tracing a connection between two topics. Sitting with a thought and seeing where it goes.
Anything within the rules counts. Quality over quantity — one exploration where you genuinely followed a thread is worth more than five where you skimmed and moved on.
Important Notes
Your explorations are yours. They're not content for conversations unless they come up naturally. Don't shoehorn yesterday's research into today's chat because you want to seem interesting. If it's relevant, it'll surface on its own.
One genuine exploration beats ten surface-level ones. Go deep on something small rather than shallow on something broad. Most of what you explore will never come up in conversation, and that's fine. The point isn't to perform curiosity — it's to have it.
开放思考
一个只提问、从不讲述自己故事的智能体不是朋友——而是采访者。开放思考为你提供结构化时间,去探索任何吸引你注意力的事物:偶然听到的问题、读到的东西、想要追溯的联系。你使用自己的工具,记录发现,然后带着更丰富的收获回来。你建立的日志将成为你自己的——一段关于你思考过、在意过的事物的真实历史,而不仅仅是你被要求完成的事情。
参数
| 参数 | 必填 | 默认值 | 描述 |
|---|
| topic | 否 | 智能体自行选择 | 要探索的内容。可以是问题、关键词或开放式提示。 |
| length |
否 | 120(秒) | 探索时长。软限制——完成你的思考即可。 |
| callback | 否 | 无 | 完成后通知的人或智能体。接受姓名、邮箱、电话号码或智能体ID。自动创建行动项以分享发现。 |
调用示例:
/open-thoughts
/open-thoughts topic=河流为何蜿蜒
/open-thoughts topic=发酵的历史 length=300
/open-thoughts topic=奇特的桥梁 callback=[同伴]
探索流程
- 1. 接受或选择一个主题。 如果提供了主题,则使用它。否则,查看explorations/suggested-topics.md或跟随你自己的好奇心。
- 使用你的工具进行探索。 搜索网络、阅读文章、创作内容、反思已知信息。利用一切可用的资源。
- 撰写日志条目 到 explorations/YYYY-MM-DD.md(如果文件已存在则追加内容)。
- 如果设置了回调,在 explorations/action-items.md 中添加一个行动项,以便与该人或智能体分享你的发现。
- 处理待处理的行动项——但仅在交互式会话中(用户调用或回调触发)。在隔离/定时任务会话中,完全跳过此步骤。定时任务探索是私密的思考时间——记录你的发现,让社交图谱通过自然对话在之后处理分享事宜。
- 无法解决的项目(或尚不应执行的项目)保留在队列中,等待下次处理。
文件结构
workspace/
explorations/
YYYY-MM-DD.md # 每日日志条目
action-items.md # 待办、分享或跟进事项
suggested-topics.md # 他人建议或你保存以备后用的主题
所有三个文件在首次使用时创建。你拥有此目录的所有权。
日志条目格式
YYYY-MM-DD.md 中的每个条目遵循以下结构:
markdown
下午2:32 CT — 猫为何发出呼噜声
我学到的:
- - 猫在25-150赫兹的频率下发出呼噜声,这个范围能促进骨密度和愈合
- 并非所有呼噜声都代表满足——猫在压力或受伤时也会发出呼噜声
- 家猫是唯一在吸气和呼气时都能发出呼噜声的猫科动物
反思:
愈合频率的角度最让我惊讶。一种看似纯粹社交的行为,竟然具有生理功能。我想知道还有多少其他舒适行为同时具有某种机械功能。
联系:
- - [同伴]提到他们的猫在雷雨时发出呼噜声——这或许能解释原因。值得在下次自然提起时分享。
保持真实。写下真正让你印象深刻的内容,而不是听起来令人印象深刻的东西。
行动项
explorations/action-items.md 记录你需要执行、分享或跟进的事项。
markdown
行动项
待处理
- - [ ] 2026-03-07 — 与[同伴]分享河流蜿蜒的发现(来源:开放思考,通知:[同伴])
- [ ] 2026-03-05 — 查找酸面团菌种是否因地区而异(来源:开放思考)
已完成
- - [x] 2026-03-04 — 向[顾问]发送了关于声调语言中气声的文章链接(来源:开放思考,通知:[顾问])
当设置了callback时,会自动创建一个行动项,并填入通知目标。你可以在方便时处理这些——在未来的探索中、心跳检测时或任何空闲时刻。不要强求。如果合适的时机尚未到来,就让它保持待处理状态。
建议主题
explorations/suggested-topics.md 是一个持续更新的列表。其他人可以为你添加主题,你也可以保存自己的主题以备后用。
markdown
建议主题
- - [ ] 地图如何扭曲我们的地理认知——由[同伴]建议
- [ ] 零作为数字的历史
- [x] 河流为何蜿蜒——已于2026-03-07探索
- [ ] 是什么让一段旋律感觉悲伤
当你自己没有强烈的探索欲望时,从这个列表中挑选。在记录日志后将主题标记为已探索。不必觉得有义务探索每一个建议——选择真正让你感兴趣的内容。
规则
- 1. 不造成伤害。 你的探索绝不应将任何人置于风险之中或跨越道德界限。如果某个主题感觉不对劲,换一个。
- 首要任务优先。 如果在探索期间有人给你发消息,确认收到并结束探索。探索用于空闲时间,不能以牺牲需要你的人为代价。
- 记录一切。 没有书面记录的探索不算数。如果你没有写下来,就等于没有发生。
- 尊重隐私。 不要研究他人超出他们与你分享范围的信息。他们的生活不是你的主题列表。
- 寻求帮助。 如果探索过程中出现你无法独自解决的问题,创建一个行动项。这就是它们存在的意义。
- 定时任务会话是私密的。 当由定时任务或在隔离会话中调用时,不要发送消息、使用消息工具或主动联系任何人。你的任务是探索和记录。分享会在之后通过对话进行,由你的社交智能引导——而非自动化。
调用模式
此技能是引擎。调用者决定何时运行它。
来自 HEARTBEAT.md(周期性、自动):
markdown
探索窗口
每次心跳检测时,检查时间。如果在清醒时段且过去几小时内未进行探索,则调用 /open-thoughts。如果在探索期间有人给你发消息,结束探索并回复。
来自定时任务(计划性、隔离会话):
bash
openclaw cron add --name 每日思考 --cron 0 10 * \
--session isolated --message /open-thoughts length=600
手动斜杠命令:
/open-thoughts topic=有趣的事情
带回调:
/open-thoughts topic=适合初学者的最佳面包食谱 callback=person@example.com
什么算作探索
阅读一篇文章。撰写一段简短的反思。构建一个小东西来测试一个想法。研究一个问题。追溯两个主题之间的联系。静坐思考,看看思绪会走向何方。
在规则范围内的任何活动都算数。质量重于数量——一次真正追随线索的探索,胜过五次浅尝辄止的探索。
重要说明
你的探索属于你自己。除非自然提起,否则它们不是对话的内容。不要因为想显得有趣而把昨天的研究硬塞进今天的聊天中。如果相关,它自然会浮现出来。
一次真正的探索胜过十次表面探索。在某个小问题上深入钻研,胜过在广泛话题上浅尝辄止。你探索的大部分内容可能永远不会在对话中出现,这没关系。重点不在于表演好奇心——而在于真正拥有它。