← 返回 Skills 市场
384961890-ui

brain 大脑

作者 🇨🇳王柏懿 · GitHub ↗ · v1.1.7 · MIT-0
cross-platform ⚠ suspicious
525
总下载
9
收藏
2
当前安装
8
版本数
在 OpenClaw 中安装
/install agent-memory-wisdom
功能描述
brain v1.1.7 用户版(免费):让AI Agent拥有类大脑决策架构。置信度触发、双保险记忆、经验胶囊复用。开箱即用,3片段内存池,2并发上限。
使用说明 (SKILL.md)

brain v1.1.7 用户版

你的AI Agent,正在漏掉你给它的每一次机会


你有没有遇到过这些?

  • 给了Agent一个任务,它做到一半突然问你"刚才那个是啥"
  • 复杂任务跑了2小时,Agent说"抱歉上下文太长了,我忘了前面干嘛了"
  • 好不容易调教好的工作方式,换个会话全忘光
  • 子agent派出去就收不回来,不知道是成功了还是还在跑

这不怪Agent。怪它没有大脑。


给它装一个

brain不是提示词模板,不是一堆规则文档。

是真的让Agent有一套类大脑神经网络在工作:

收到任务
  → 感知:这是什么活儿,紧急吗,跟之前有关系吗
  → 思考:这事我能做吗,要花多少预算,要不要先写个快照
  → 记忆:有没有做过类似的,上次怎么干的
  → 执行:派给谁,怎么派,超时了怎么办
  → 复盘:成了吗,要不要把这次经验存起来

每个环节都有脚本支撑,每个决策都有据可查。


核心能力

🔒 置信度双保险

Agent每次接活儿,先给自己打个分(0~1)。

  • 0.9:简单直接,直接干
  • 0.7~0.9:有点东西,干完记得汇报
  • \x3C 0.6:高危,先把快照写了再说

打分不是拍脑袋,是信号表算出来的:

+0.1  历史成功过这个类型
+0.1  任务描述很明确
-0.3  涉及删除/销毁
-0.2  涉及发布/上线
-0.2  涉及密码/密钥
-0.2  系统核心配置
-0.2  第一次遇到这种任务

置信度\x3C0.6时,系统强制先写快照再执行。 就算中途断电,回来一看就知道做到哪了。


📦 经验胶囊

Agent每次成功搞定一个复杂任务,把这次"怎么干的"打包成胶囊。

下次遇到类似的,直接调取经验,不用从0摸索。

胶囊 = 一套可复用的成功模式

比如:
CAP-飞书消息发送-20260401.md
CAP-官网迭代流程-20260415.md

胶囊有成熟度:

  • raw:刚试成功,还不确保持续能用
  • tested:连续成功2次,可以用
  • stable:连续成功5次,默认自动用

🧠 语义记忆搜索

问Agent:"上次那个调研是怎么做的来着?"

它能语义搜索历史记忆,不只是关键词匹配。


vs 完整版差在哪

用户版 完整版
置信度触发 ✅ 精确化+对抗验证
经验胶囊 ✅ 手动创建 ✅ 自动维护成熟度
语义记忆搜索 ✅ 基础版 ✅ 全量QMD(~24000 chunks)
并发Agent上限 2个 无限制
内存池上限 3片段/5K tokens 无限制/30K tokens
自动化脚本 ✅ 全部(含auto-*系列)
记忆Fragment接口 ✅ 标准化
测试框架 ✅ 30+用例自动化验证
Context实时监控 ✅ 自动压缩+标注

脚本清单

scripts/
├── build-session-injection.js   # 会话启动时自动组装上下文
├── subagent-think-chain.js    # 任务分类+置信度评估(派agent前必跑)
├── subagent-budget.js         # 任务复杂度→预算+模型分配
├── subagent-watchdog.js       # 超时重试+失败兜底
├── pre-checkpoint.js         # 高危任务前主动写快照
├── capsule-auto-suggest.sh    # 任务完成后自动建议创建胶囊
└── create-capsule.sh          # 手动创建胶囊

安装

# 1. 复制脚本到brain目录
cp ~/.openclaw/skills/brain-v1.1.7-user/scripts/* \
   ~/.openclaw/skills/brain-v1.1.7/scripts/

# 2. 启动上下文注入(在 openclaw.json 的 agents.defaults 下加入)
"bootstrapExtraFiles": [
  "memory/SNAPSHOT.md",
  "memory/工作缓冲区.md"
]

# 3. 任务来了,先跑思考链
node ~/.openclaw/skills/brain-v1.1.7/scripts/subagent-think-chain.js "\x3C任务描述>"

常见问题

Q: 为什么不给我更多并发和内存? A: 用户版是证明能力的。如果这套系统对你有价值,完整版给你专业级的无限量。

Q: 真的有用吗? A: 最直接验证:跑一个复杂任务,中途Ctrl+C断掉,然后新会话里问"刚才那个任务做到哪了"。有快照的版本能接上,没快照的版本从头开始。

Q: 会不会太复杂? A: 用户版已经做了最大简化。你只需要记住一件事:每个任务开始前跑一下 subagent-think-chain.js,剩下的脚本自动处理。


一句话

装上之后,你的Agent不再健忘,不再乱来,不再忘了自己干到哪了

不是变聪明了——是终于有记性了。


关于作者

一个中国开发者。

至今没搞懂python到底是干什么用的,没学过任何语言,但对架构及工程思维有那么一点点🤏天赋(自夸)。

最初只是想让我的agent别那么健忘,省点token,结果越搞越上头,一不小心整出了一套类大脑神经网络架构。

pete说过:"你很难和一个只想玩的人做对手"

如果你也觉得有意思,欢迎联系我交流。

您的意见与反馈,是我进步的动力。


想要完整版?

brain v1.1.7 完整版 包含:

  • 无限制并发Agent:任务再复杂也能并行跑
  • 30K tokens内存池:大型任务不断档
  • 全量QMD语义搜索(~24000 chunks):跨workspace全局记忆
  • 记忆Fragment标准接口:统一记忆格式,注入全链路可控
  • 10K token硬上限:自动压缩+标注,永远不爆token
  • 30+自动化测试用例:每次改动都能秒级验证
  • *全部auto-脚本:主动预判+胶囊自动维护
  • Context实时监控:后台守护,永不失控

订阅费用: $6/月 订阅方式: 发送邮件至 [email protected],标题注明"订阅brain完整版" 增值服务: 若有升级/内测版本,会第一时间通过Gmail通知您


brain v1.1.7 用户版 — 开箱即用,试试它到底有没有用

安全使用建议
This skill implements a local 'brain' using scripts that read and write files in ~/.openclaw/workspace and that can spawn subagents via the openclaw CLI. Before installing: - Inspect the scripts yourself (they are plain JS / shell) and confirm you are comfortable with them reading SNAPSHOT.md and 工作缓冲区.md and writing logs/capsules under your HOME directory. - Note the packaging omission: the skill did not declare required binaries. Ensure you have Node.js and the 'openclaw' CLI available and understand what openclaw sessions_spawn will do in your environment. - Be cautious about sensitive data: don't store secrets or credentials in SNAPSHOT.md or the work buffer because their contents are programmatically injected into agent contexts. - The watchdog provides an 'exec' entrypoint and a general execWithRetry wrapper. Only run it with trusted commands and avoid passing untrusted input into shell execution paths. - Test in an isolated workspace (or backup current ~/.openclaw/workspace) before copying scripts into production. If you need stricter control, remove or sandbox the parts that spawn subagents or execute shell commands. If you want, I can: (a) list the exact lines that call external commands and file paths, (b) suggest a safer minimal configuration that disables exec entrypoints, or (c) produce a checklist for a secure manual install.
功能分析
Type: OpenClaw Skill Name: agent-memory-wisdom Version: 1.1.7 The bundle implements a 'brain' architecture for AI agents, providing frameworks for memory, task splitting, and confidence-based decision-making. It contains high-risk capabilities, specifically a shell execution wrapper in scripts/subagent-watchdog.js that uses execSync and a subagent spawning utility (openclaw sessions_spawn). While the bundle includes safety-oriented logic in scripts/subagent-think-chain.js—which penalizes confidence scores for risky actions like deleting files or accessing secrets—the provision of generic shell execution tools and broad file system access (scripts/build-session-injection.js) for memory management constitutes a significant attack surface if the agent is prompted to execute unsanitized input.
能力评估
Purpose & Capability
The scripts implement the described features (confidence scoring, checkpoints, 'capsules', semantic recall helper, watchdog and subagent routing). However the packaging declares no required binaries while several scripts expect a Node environment and the 'openclaw' CLI (sessions_spawn) to exist; that dependency is not declared in the skill metadata/instructions.
Instruction Scope
Runtime instructions (SKILL.md) tell you to copy scripts into your skills/workspace and to add SNAPSHOT.md and 工作缓冲区.md to bootstrapExtraFiles so their contents are injected into agent context. The scripts read and write files under $HOME/.openclaw/workspace (SNAPSHOT.md, 工作缓冲区.md, logs, capsules directory) and will publish assembled injection JSON to stdout. This legitimately supports the stated purpose, but it also means potentially sensitive workspace content will be programmatically read and included in agent contexts — and some wrappers allow execution of arbitrary shell commands when invoked.
Install Mechanism
No automated install spec — instruction-only plus scripts that you copy manually. This is lower install risk (nothing downloaded at runtime). The user is required to manually copy scripts into their environment, which gives them a chance to inspect files before use.
Credentials
No environment variables or credentials are declared or requested, which aligns with a local, file-based memory system. However the code assumes process.env.HOME, a Node runtime, and an 'openclaw' CLI binary; those runtime dependencies are not listed. There are no network endpoints or secret-exfiltration hooks in the code, but the scripts will read whatever is in your workspace snapshot/buffer (which may contain secrets) and can write logs and capsule files to your HOME.
Persistence & Privilege
always:false (good). The skill writes and updates files under ~/.openclaw/workspace (buffers, snapshots, logs, capsule files) and instructs adding files to bootstrapExtraFiles so they become part of agent context — this is persistent and intended, but you should be aware it modifies your workspace files and will cause their contents to be injected into agent prompts.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install agent-memory-wisdom
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /agent-memory-wisdom 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.1.7
brain v1.1.7 用户版采用新版可读性更强的说明文档,突出用户核心体验和区别点: - 完全重写的用户文档,突出场景化说明,便于上手和理解 - 明确列出核心能力:置信度触发、双保险机制、经验胶囊、语义记忆搜索 - 增加与完整版能力差异清单,便于用户快速对比 - 优化脚本使用方法和安装步骤指引 - 简化说明结构,降低学习门槛,增强试用转化引导
v1.1.5
brain v1.1.5 brings a user-focused major upgrade with a new brain-inspired architecture and practical new scripts. - 全新类大脑神经网络架构,实现感知、认知、记忆、调度多层分工,让Agent更稳更省钱。 - 增加工作缓冲区与置信度双保险机制,长链任务不中断、不跑偏,显著提升稳定性。 - 引入经验胶囊系统和自动胶囊建议脚本,支持复用高价值任务经验,持续进步少踩坑。 - 新增7个实用脚本(任务预算、会话注入、思考链、看门狗、预检查点等),开箱即用。 - 优化记忆持久性,推荐自适应压缩,单次会话可连续2-3天记忆不丢失。 - 清理冗余参考文档,保持文件结构简洁。
v1.1.4
brain v1.1.4 - Introduced感知置信度系统, enabling AI agents to assess task confidence and choose appropriate capabilities. - Replaced传统三省六部制 with a capability-driven, parallel agent architecture for higher efficiency. - EnabledAI自主判断与能力路由, emphasizing "what is needed" over "who is responsible." - Enhanced并行任务处理, allowing multiple agents to deep-dive and aggregate results simultaneously. - Added会话注入和动态预算分配, further improving memory precision and task adaptability.
v1.1.3
🎉 brain 1.1.5 更新! 三个升级: 子agent预算 — 按任务复杂度分配,简单10次/中等30次/复杂不限,不再无脑烧token 感知置信度 — 每次感知带百分比,还能统计准确率,不再盲目瞎猜 会话注入 — 自动组装上下文丢进系统prompt,不用每次从头解释 对比hermes: hermes子agent固定50次(不管任务难不难),我们智能分配;hermes没有感知层,我们有还带校准;hermes静态文件注入,我们动态组装。 就...比它聪明一点点 🤏
v1.1.2
brain v1.1.5 introduces smarter agent operations with dynamic budgets, confidence-based perception, and session context injection. - Adds dynamic subagent budget allocation based on task complexity. - Upgrades perception logging to include confidence scores and accuracy tracking. - Introduces automatic session context injection for more relevant information flow. - Implements agentic generation-evaluation loop for self-improvement. - Updates startup sequence and file structure for the new features.
v1.1.1
- Skill renamed from "agent-memory-wisdom" to "brain" for a clearer identity. - Documentation streamlined and improved for clarity and onboarding, including simplified file structure explanations. - Added and reorganized examples for quick-start and essential practices (快照、感知层、自动推送等). - Updated terminology (e.g., "用户" replaces "爸爸") and refined descriptions for general use. - More concise tables and clearer instructions on incorporating perception and user habits. - No code or logic changes; this update is documentation and naming only.
v1.1.0
让你的Agent 拥有“知识网络” 真正的“大脑” **agent-memory-wisdom v2.0.0** — 一次重大升级,让AI记忆更聪明、更主动 - 引入SNAPSHOT.md,支持“快照恢复”,AI启动即可秒懂当前上下文和关键状态 - 增加“感知层”,记录并理解用户情绪、状态和偏好,实现更有温度的交互 - 实现“知识关联网络”,通过标签链路将知识点互相关联,便于查找和推理 - 支持主动提醒(通知/推送),AI可根据条件主动触发消息,无需用户催问 - 优化启动序列:依赖SNAPSHOT一站式恢复+多层文件协同,启动效率大幅提升 - 全新文档结构与使用规则,更易用、更贴近实际AI助理场景
v1.0.0
agent-memory-wisdom 1.0.0 - Initial release of the agent-memory-wisdom skill. - Provides a three-layer file-based memory system for AI Agents, including WAL Protocol and optional vector memory architecture. - Enables persistent, cross-session memory for personal assistants, multi-agent collaboration, and long-term tasks. - Framework-agnostic: can be used with any AI Agent implementation. - Includes detailed setup, usage guidelines, and best practices for robust AI memory management.
元数据
Slug agent-memory-wisdom
版本 1.1.7
许可证 MIT-0
累计安装 2
当前安装数 2
历史版本数 8
常见问题

brain 大脑 是什么?

brain v1.1.7 用户版(免费):让AI Agent拥有类大脑决策架构。置信度触发、双保险记忆、经验胶囊复用。开箱即用,3片段内存池,2并发上限。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 525 次。

如何安装 brain 大脑?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install agent-memory-wisdom」即可一键安装,无需额外配置。

brain 大脑 是免费的吗?

是的,brain 大脑 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

brain 大脑 支持哪些平台?

brain 大脑 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 brain 大脑?

由 🇨🇳王柏懿(@384961890-ui)开发并维护,当前版本 v1.1.7。

💬 留言讨论