wiki-ingest: 4 Agent Use Cases (autonomous PM, content factory, product factory, knowledge base RAG) - 2026-04-15 evening batch
This commit is contained in:
48
wiki/sources/Autonomous-Project-Management.md
Normal file
48
wiki/sources/Autonomous-Project-Management.md
Normal file
@@ -0,0 +1,48 @@
|
||||
---
|
||||
title: "Autonomous Project Management with Subagents"
|
||||
type: source
|
||||
tags: [agent, project-management, subagent]
|
||||
date: 2026-04-15
|
||||
---
|
||||
|
||||
## Source File
|
||||
- [[raw/Agent/usecases/autonomous-project-management.md]]
|
||||
|
||||
## Summary
|
||||
- 核心主题:去中心化项目管理模式,多 subagent 通过共享 STATE.yaml 协调而非中央 orchestrator
|
||||
- 问题域:传统 orchestrator 模式造成主 agent 瓶颈,多 repo 重构/研究冲刺/内容管线等复杂项目需要并行执行
|
||||
- 方法/机制:STATE.yaml 作为单一事实来源 → 各 agent 自主认领任务 → 状态更新触发其他 agent 接力
|
||||
- 结论/价值:文件协调优于消息传递;Git 作为审计日志;薄主会话原则(CEO 模式)
|
||||
|
||||
## Key Claims
|
||||
- 传统 orchestrator 模式产生瓶颈——主 agent 成为交通指挥
|
||||
- STATE.yaml > orchestrator:文件协调比消息传递更具扩展性
|
||||
- Git 作为审计日志:提交 STATE.yaml 变更获取完整历史
|
||||
- Label 约定很重要:用 `pm-{project}-{scope}` 格式便于追踪
|
||||
- 薄主会话原则:主 agent 做得越少,响应越快
|
||||
|
||||
## Key Quotes
|
||||
> "Managing complex projects with multiple parallel workstreams is exhausting. You end up context-switching constantly." — 痛点陈述
|
||||
> "Let agents self-organize rather than micromanaging them." — [[Nicholas Carlini]] 方法论核心
|
||||
|
||||
## Key Concepts
|
||||
- [[STATE.yaml]]:项目协调文件,YAML 格式定义任务状态、owner、blocked_by 依赖关系
|
||||
- [[去中心化协调]]:无中央 orchestrator,各 agent 自主读写共享状态文件
|
||||
- [[薄主会话]]:主会话仅做策略/调度,所有执行下沉 subagent
|
||||
- [[CEO 模式]]:主 agent = 协调者,subagent = 执行者
|
||||
|
||||
## Key Entities
|
||||
- [[Nicholas Carlini]]:自主编码 agent 方法论提出者,STATE.yaml 协调模式灵感来源
|
||||
- [[Anthropic]]:Building Effective Agents 论文发布方
|
||||
|
||||
## Connections
|
||||
- [[Multi-Agent Hierarchy]] ← 架构基础 ← [[Autonomous Project Management]]
|
||||
- [[sessions_spawn]] ← 核心能力 ← [[Autonomous Project Management]]
|
||||
- [[sessions_send]] ← 核心能力 ← [[Autonomous Project Management]]
|
||||
- [[GitOps]] ← 审计日志机制 ← [[Autonomous Project Management]]
|
||||
|
||||
## Contradictions
|
||||
- 与中央 orchestrator 模式冲突:
|
||||
- 当前观点:去中心化文件协调,无单点瓶颈
|
||||
- 对方观点:中央 orchestrator 便于全局控制
|
||||
- 适用场景:复杂多任务 > 简单顺序任务
|
||||
46
wiki/sources/Content-Factory.md
Normal file
46
wiki/sources/Content-Factory.md
Normal file
@@ -0,0 +1,46 @@
|
||||
---
|
||||
title: "Multi-Agent Content Factory"
|
||||
type: source
|
||||
tags: [agent, content, multi-agent, discord]
|
||||
date: 2026-04-15
|
||||
---
|
||||
|
||||
## Source File
|
||||
- [[raw/Agent/usecases/content-factory.md]]
|
||||
|
||||
## Summary
|
||||
- 核心主题:Discord 内多 agent 内容工厂,研究→写作→设计链式协作
|
||||
- 问题域:内容创作三阶段(研究/写作/设计)全靠人工,AI 工具也需逐个 prompt
|
||||
- 方法/机制:Research Agent 扫描趋势 → Writing Agent 生成脚本 → Thumbnail Agent 生成配图,全流程定时自动执行
|
||||
- 结论/价值:链式 agent 威力在于前序输出喂给后序;Discord channel 便于分阶段 review
|
||||
|
||||
## Key Claims
|
||||
- 链式 agent 是核心:research → writing → thumbnails,每步输出自动喂给下一步
|
||||
- Discord channel 分离各 agent 工作便于 review 和反馈
|
||||
- 本地模型(如 Mac Studio 上的 Nano Banana)降低成本增加控制
|
||||
- 可适配任何内容格式:tweets/newsletter/LinkedIn/posts/podcast outlines/blog
|
||||
|
||||
## Key Quotes
|
||||
> "Content creation has three phases — research, writing, and design — and most creators are doing all three manually." — 痛点陈述
|
||||
> "The power is in the chained agents — research feeds writing, writing feeds thumbnails." — 核心价值
|
||||
|
||||
## Key Concepts
|
||||
- [[内容工厂]]:研究+写作+设计三阶段链式 agent 协作管线
|
||||
- [[链式 Agent]]:前序 agent 输出自动作为后序 agent 输入
|
||||
- [[Discord 集成]]:多 channel 分离各 agent 工作流
|
||||
|
||||
## Key Entities
|
||||
- [[Alex Finn]]:Life-changing OpenClaw use cases 视频作者,本内容工厂灵感来源
|
||||
|
||||
## Connections
|
||||
- [[Market Research Product Factory]] ← 共享研究 agent 基础设施
|
||||
- [[Last30Days]] ← Research Agent 数据来源
|
||||
- [[baoyu-imagine]] ← Thumbnail Agent 图像生成能力
|
||||
- [[Agentic AI]] ← 自主执行多步骤工作流
|
||||
- [[multi-agent-team]] ← 多 agent 协作模式
|
||||
|
||||
## Contradictions
|
||||
- 与单人内容创作流程对比:
|
||||
- 当前观点:多 agent 链式协作,人工降至最低
|
||||
- 对方观点:单人全流程控制质量更一致
|
||||
- 结论:质量控制需人工 review 环节,不可完全自动化
|
||||
48
wiki/sources/Market-Research-Product-Factory.md
Normal file
48
wiki/sources/Market-Research-Product-Factory.md
Normal file
@@ -0,0 +1,48 @@
|
||||
---
|
||||
title: "Market Research & Product Factory"
|
||||
type: source
|
||||
tags: [agent, market-research, product, last30days]
|
||||
date: 2026-04-15
|
||||
---
|
||||
|
||||
## Source File
|
||||
- [[raw/Agent/usecases/market-research-product-factory.md]]
|
||||
|
||||
## Summary
|
||||
- 核心主题:Last30Days 挖掘 Reddit/X 真实痛点 → OpenClaw 构建 MVP 解决方案的自动化创业管线
|
||||
- 问题域: aspiring entrepreneur 面临"做什么产品"困境;传统市场研究耗时耗力
|
||||
- 方法/机制:Last30Days 研究痛点 → ranked pain points → 选择痛点 → OpenClaw 构建 MVP
|
||||
- 结论/价值:创业自动化:找问题→验证需求→构建方案,全流程通过文本对话完成
|
||||
|
||||
## Key Claims
|
||||
- Last30Days 给出真实、未经过滤的用户情绪,而非经过滤的调查数据
|
||||
- 不需要技术背景,OpenClaw 做研究和构建两部分
|
||||
- 每周定时研究保持对市场痛点演变的持续追踪
|
||||
- 研究→产品完整管线零编码
|
||||
|
||||
## Key Quotes
|
||||
> "Most aspiring entrepreneurs struggle with the 'what to build' problem." — 核心痛点
|
||||
> "This is entrepreneurship on autopilot: find problems → validate demand → build solutions, all through text messages." — 价值主张
|
||||
|
||||
## Key Concepts
|
||||
- [[产品工厂]]:市场研究到 MVP 构建的自动化管线
|
||||
- [[Last30Days]]:多平台热点聚合工具(Reddit/X/YouTube/Polymarket 等)
|
||||
- [[创业自动化]]:问题发现→需求验证→方案构建全链路
|
||||
- [[MVP 构建]]:最小可行产品快速原型
|
||||
|
||||
## Key Entities
|
||||
- [[Matt Van Horne]]:Last30Days skill 作者
|
||||
- [[Alex Finn]]:Life-changing OpenClaw use cases 视频作者
|
||||
|
||||
## Connections
|
||||
- [[Last30Days]] ← 核心技术能力
|
||||
- [[Content Factory]] ← 共享研究基础设施
|
||||
- [[AI产品经理]] ← 产品创意与需求分析能力
|
||||
- [[多平台热点聚合]] ← 研究方法论
|
||||
- [[社交信号权重]] ← 痛点评级框架
|
||||
|
||||
## Contradictions
|
||||
- 与传统市场研究方法冲突:
|
||||
- 当前观点:Last30Days 实时抓取真实用户帖子反映即时痛点
|
||||
- 对方观点:传统调研方法(访谈/问卷)结构化程度更高
|
||||
- 结论:两者互补,定量+定性结合最优
|
||||
@@ -0,0 +1,44 @@
|
||||
---
|
||||
title: "Multi-Agent Specialized Team (Solo Founder Setup)"
|
||||
type: source
|
||||
tags: [openclaw, multi-agent, telegram, solo-founder, workflow]
|
||||
date: 2026-04-15
|
||||
---
|
||||
|
||||
## Source File
|
||||
- [[raw/Agent/usecases/multi-agent-team.md]]
|
||||
|
||||
## Summary
|
||||
- 核心主题:Solo founder 通过多 Agent 虚拟团队实现 24/7 全天候工作能力
|
||||
- 问题域:单一 Agent 无法高效处理多领域工作;Context 切换破坏深度工作;知识孤岛导致洞察无法跨 Agent 流动
|
||||
- 方法/机制:专业化 Agent(各角色独立模型/人格)+ 共享记忆(GOALS.md/DECISIONS.md)+ 私有上下文 + Telegram 统一入口 + 定时主动任务
|
||||
- 结论/价值:从 2 Agent 开始按瓶颈扩展;定时任务是价值飞轮;团队协作产生真正价值
|
||||
|
||||
## Key Claims
|
||||
- 单一 Agent 无法高效处理战略/开发/营销/分析多领域,context window 快速填满
|
||||
- 共享记忆(GOALS.md/DECISIONS.md)+ 私有上下文是多 Agent 协作核心
|
||||
- 所有 Agent 通过同一 Telegram 群聊控制,各自只响应被 @ 的消息
|
||||
- 定时主动任务(早会摘要/指标推送/内容创意)是价值飞轮
|
||||
- 从 2 Agent 开始,不是一上来建 4 个团队
|
||||
- [[Trebuh]] 的实践:4 个 Agent(Milo/Josh/Marketing/Dev)+ Telegram + VPS,描述为"真正的 24/7 小团队"
|
||||
|
||||
## Key Quotes
|
||||
> "Start with 2, not 4: Begin with a lead + one specialist, then add agents as you identify bottlenecks" — 实践总结
|
||||
> "The real value emerges when agents proactively surface insights, not just when you ask" — 定时任务价值
|
||||
> "Personality matters more than you'd think: Giving agents distinct names and communication styles makes it natural to talk to your team" — Trebuh
|
||||
|
||||
## Key Concepts
|
||||
- [[共享记忆模式]]:GOALS.md(OKR与优先级,所有Agent可读)+ DECISIONS.md(关键决策日志)+ PROJECT_STATUS.md(当前项目状态)
|
||||
- [[定时主动任务]]:Agent 主动在后台工作并推送结果,而非等待用户请求
|
||||
- [[Multi-Agent Hierarchy]]:团队层级架构,Lead Agent 协调 + Specialist Agent 执行
|
||||
- [[Telegram路由]]:单群聊入口 + @AgentName 路由 + 无@默认 Lead Agent
|
||||
|
||||
## Key Entities
|
||||
- [[Trebuh]]:Solo founder,4 Agent 团队实践者,通过 X 分享案例
|
||||
- [[OpenClaw]]:多 Agent 管理平台,支持 sessions_spawn/sessions_send/Telegram skill
|
||||
|
||||
## Connections
|
||||
- [[Trebuh]] ← 实践者 ← [[Multi-Agent Specialized Team]]
|
||||
- [[OpenClaw]] ← 平台 ← [[Multi-Agent Specialized Team]]
|
||||
- [[共享记忆模式]] ← 核心机制 ← [[Multi-Agent Specialized Team]]
|
||||
- [[定时主动任务]] ← 价值飞轮 ← [[Multi-Agent Specialized Team]]
|
||||
45
wiki/sources/Personal-Knowledge-Base-RAG.md
Normal file
45
wiki/sources/Personal-Knowledge-Base-RAG.md
Normal file
@@ -0,0 +1,45 @@
|
||||
---
|
||||
title: "Personal Knowledge Base (RAG)"
|
||||
type: source
|
||||
tags: [agent, rag, knowledge-base, memory]
|
||||
date: 2026-04-15
|
||||
---
|
||||
|
||||
## Source File
|
||||
- [[raw/Agent/usecases/knowledge-base-rag.md]]
|
||||
|
||||
## Summary
|
||||
- 核心主题:语义可搜索个人知识库,自动从任意 URL(文章/tweets/YouTube/PDF)摄取内容
|
||||
- 问题域:书签堆积无法检索,看过的内容找不到
|
||||
- 方法/机制:Drop URL 到 Telegram/Slack → 自动抓取内容 → 向量嵌入 → 语义搜索返回 ranked 结果+来源
|
||||
- 结论/价值:RAG 驱动的个人第二大脑;其他工作流可查询知识库获取相关已存内容
|
||||
|
||||
## Key Claims
|
||||
- Drop any URL 自动摄取:文章/tweets/YouTube transcripts/PDFs
|
||||
- 语义搜索:"What did I save about agent memory?" 返回 ranked 结果+来源引文
|
||||
- 喂入其他工作流:视频创意管线构建 research cards 时自动查询知识库
|
||||
- Zero maintenance:URL 即摄入触发器
|
||||
|
||||
## Key Quotes
|
||||
> "You read articles, tweets, and watch videos all day but can never find that one thing you saw last week." — 核心痛点
|
||||
|
||||
## Key Concepts
|
||||
- [[RAG]]:Retrieval-Augmented Generation,基于向量嵌入的语义检索
|
||||
- [[个人知识库]]:第二大脑,内容自动积累+语义检索
|
||||
- [[语义搜索]]:自然语言查询,返回 ranked 相关结果
|
||||
- [[内容摄取]]:URL → 结构化文本 → 向量嵌入全流程
|
||||
|
||||
## Key Entities
|
||||
|
||||
## Connections
|
||||
- [[NotebookLM]] ← 共享 source-grounding 理念
|
||||
- [[向量数据库]] ← 底层存储检索基础设施
|
||||
- [[Embedding]] ← 语义表示核心机制
|
||||
- [[Agentic AI]] ← 自主触发知识库查询
|
||||
- [[Content Factory]] ← 可查询知识库获取背景信息
|
||||
|
||||
## Contradictions
|
||||
- 与传统书签/笔记工具冲突:
|
||||
- 当前观点:语义搜索+自动摄取优于手动书签整理
|
||||
- 对方观点:手动整理确保质量和结构
|
||||
- 结论:摄取自动化+人工审核结合最优
|
||||
42
wiki/sources/Vibe-Kanban-OpenCode-Ubuntu-Server安装管理指南.md
Normal file
42
wiki/sources/Vibe-Kanban-OpenCode-Ubuntu-Server安装管理指南.md
Normal file
@@ -0,0 +1,42 @@
|
||||
---
|
||||
title: "Vibe-Kanban + OpenCode 在 Ubuntu Server 上安装与管理指南"
|
||||
type: source
|
||||
tags: [ubuntu, vibe-coding, vibe-kanban, opencode, pm2, node, nvm]
|
||||
date: 2026-04-15
|
||||
---
|
||||
|
||||
## Source File
|
||||
- [[raw/Vibe Coding/Vibe-Kanban + OpenCode 在 Ubuntu Server 上安装与管理指南.md]]
|
||||
|
||||
## Summary
|
||||
- 核心主题:在 Ubuntu Server 上以非 root 用户(shenwei)安装 Node 20、Vibe-Kanban 与 OpenCode,并通过 pm2 管理进程
|
||||
- 问题域:权限冲突、端口占用、executor 启动失败、I/O error
|
||||
- 方法/机制:nvm 管理 Node 版本 → npm 全局安装 vibe-kanban + opencode → pm2 管理进程 → RUST_LOG=debug 调试
|
||||
- 结论/价值:不要用 root 启动;vibe-kanban 自动 spawn executor;pm2 只管理 vibe-kanban,executor 随进程管理
|
||||
|
||||
## Key Claims
|
||||
- nvm 管理 Node 版本(v0.39.7),安装 Node 20,nvm alias default 20
|
||||
- 不要用 root 启动 OpenCode serve,vibe-kanban 会自动 spawn executor
|
||||
- RUST_LOG=debug HOST=0.0.0.0 PORT=9999 npx vibe-kanban 启动,executor 在随机端口
|
||||
- pm2 start "RUST_LOG=debug HOST=0.0.0.0 PORT=9999 npx vibe-kanban" --name vibe-kanban
|
||||
- pm2 startup systemd -u shenwei --hp /home/shenwei 生成启动脚本
|
||||
- I/O error 通常是 executor 没启动或端口被占用
|
||||
- 清理旧工作树:rm -rf /var/tmp/vibe-kanban/worktrees/* 和 ~/.vibe-kanban
|
||||
|
||||
## Key Quotes
|
||||
> "不要用 root 启动 OpenCode serve,vibe-kanban 会自动 spawn executor" — 操作规范
|
||||
> "pm2 只管理 vibe-kanban,executor 随进程一起管理" — 进程管理原则
|
||||
|
||||
## Key Concepts
|
||||
- [[nvm]]:Node Version Manager,通过 curl -fsSL 安装,管理多个 Node 版本
|
||||
- [[pm2]]:进程管理器,pm2 start/pm2 logs/pm2 save/pm2 startup systemd
|
||||
- [[Vibe-Kanban]]:AI 结对编程任务看板,Web UI(PORT 9999)+ executor 随机端口
|
||||
- [[OpenCode Executor]]:vibe-kanban spawn 的 AI 编程执行器,随机端口运行
|
||||
|
||||
## Key Entities
|
||||
- [[shenwei]]:Ubuntu 服务器非 root 用户,Vibe-Kanban + OpenCode 安装操作者
|
||||
|
||||
## Connections
|
||||
- [[nvm]] ← 安装工具 ← [[Node 20]]
|
||||
- [[Vibe-Kanban]] ← spawns ← [[OpenCode Executor]]
|
||||
- [[pm2]] ← 管理 ← [[Vibe-Kanban]]
|
||||
48
wiki/sources/baoyu-skills-claude-code-技能集.md
Normal file
48
wiki/sources/baoyu-skills-claude-code-技能集.md
Normal file
@@ -0,0 +1,48 @@
|
||||
---
|
||||
title: "baoyu-skills Claude Code 技能集"
|
||||
type: source
|
||||
tags: [claude-code, skills, baoyu, openclaw, 内容生成, AI图像]
|
||||
date: 2026-04-15
|
||||
---
|
||||
|
||||
## Source File
|
||||
- [[raw/Skills/baoyu-skills-claude-code-技能集.md]]
|
||||
|
||||
## Summary
|
||||
- 核心主题:宝玉(JimLiu)发布的 Claude Code 技能集,通过 ClawHub 协议按单个 skill 安装
|
||||
- 问题域:内容创作者和开发者需要多平台内容发布、多服务商图像生成、日常效率工具
|
||||
- 方法/机制:技能分内容技能(内容发布)、AI生成技能(图像/文本)、工具技能(处理工具)三大类;ClawHub 按单个 skill 安装(clawhub install baoyu-imagine)
|
||||
- 结论/价值:覆盖小红书/X/微信/微博全平台,9家图像服务商自动选择,翻译/抓取/压缩等日常工具
|
||||
|
||||
## Key Claims
|
||||
- ClawHub 按单个 skill 安装,不是 marketplace 批量装(clawhub install baoyu-imagine)
|
||||
- baoyu-imagine 支持 9 家服务商(OpenAI/Google/Azure/OpenRouter/DashScope/MiniMax/即梦/豆包/Replicate),自动选择最优
|
||||
- baoyu-translate 三模式(快速/标准/精翻)覆盖从短文本到出版级文档
|
||||
- baoyu-comic 5种画风 × 8种基调,漫画生成支持预设(ohmsha/wuxia/shoujo)
|
||||
- baoyu-post-to-wechat 支持 API 方式和浏览器方式发布公众号文章
|
||||
- EXTEND.md 支持用户级/项目级自定义,覆盖样式/配置/个人预设
|
||||
|
||||
## Key Quotes
|
||||
> "ClawHub 按单个 skill 安装,不是把整个 marketplace 一次性装进去" — 宝玉
|
||||
> "最好的 Skill 是工具箱,而非写好的提示词" — Anthropic
|
||||
> "精翻模式完成后,可回复「继续润色」或「refine」继续审校润色步骤" — baoyu-translate
|
||||
|
||||
## Key Concepts
|
||||
- [[内容技能]]:内容生成与发布类技能子集,baoyu-xhs-images/infographic/cover-image/slide-deck/comic/article-illustrator
|
||||
- [[baoyu-imagine]]:9家服务商自动选择的AI图像生成,支持文生图/参考图/自定义尺寸/批量生成
|
||||
- [[baoyu-infographic]]:20种布局×17种视觉风格的专业信息图生成
|
||||
- [[baoyu-xhs-images]]:小红书信息图,9种风格×6种布局的二维系统
|
||||
- [[baoyu-translate]]:三模式翻译(快速/标准/精翻),支持受众预设和风格预设
|
||||
- [[baoyu-comic]]:知识漫画创作,5种画风×8种基调,支持预设(ohmsha/wuxia/shoujo)
|
||||
- [[工具技能]]:baoyu-youtube-transcript/url-to-markdown/x-to-markdown/compress-image/format-markdown/markdown-to-html/translate
|
||||
|
||||
## Key Entities
|
||||
- [[宝玉]](JimLiu):baoyu-skills 项目作者,Claude Code 技能集开发者
|
||||
- [[JimLiu]]:GitHub 账号,baoyu-skills 仓库维护方
|
||||
- [[ClawHub]]:按单个 skill 安装的插件市场协议
|
||||
|
||||
## Connections
|
||||
- [[宝玉]] ← 发布 ← [[baoyu-skills]]
|
||||
- [[baoyu-imagine]] ← 依赖 ← [[ClawHub]]
|
||||
- [[baoyu-xhs-images]] ← 属于 ← [[内容技能]]
|
||||
- [[baoyu-translate]] ← 属于 ← [[工具技能]]
|
||||
46
wiki/sources/养虾日记2-OpenClaw-Self-Improving复盘实战.md
Normal file
46
wiki/sources/养虾日记2-OpenClaw-Self-Improving复盘实战.md
Normal file
@@ -0,0 +1,46 @@
|
||||
---
|
||||
title: "养虾日记2:让Agent更懂你:OpenClaw + Self-Improving 复盘实战案例分享"
|
||||
type: source
|
||||
tags: [openclaw, self-improving, memory, agent, 复盘, 经验积累]
|
||||
date: 2026-04-15
|
||||
---
|
||||
|
||||
## Source File
|
||||
- [[raw/微信公众号/养虾日记2:让Agent更懂你:OpenClaw + Self-Improving 复盘实战案例分享.md]]
|
||||
|
||||
## Summary
|
||||
- 核心主题:通过 self-improving skill + 双层记忆架构 + 每日定时复盘,实现 Agent 在错误中学习、持续进化
|
||||
- 问题域:AI 每次对话都是白纸,没有记忆;同一个错误反复出现;无对话日出现记忆断层
|
||||
- 方法/机制:短期记忆(memory/YYYY-MM-DD.md)+ 长期记忆(memory-lancedb-pro)+ self-improving 复盘(LEARNINGS.md,Pattern-Key 追踪)
|
||||
- 结论/价值:错误只犯一次;Pattern-Key 重复是系统性问题的信号;Recurrence-Count 区分偶发错误与系统性问题
|
||||
|
||||
## Key Claims
|
||||
- 双层记忆架构:短期(每日文件)+ 长期(LanceDB 向量数据库)+ self-improving(成长追踪)
|
||||
- 每日 23:00 定时复盘,cron job 触发,agent 独立运行各自复盘流程
|
||||
- LEARNINGS.md 固定格式:Summary/Details/Suggested Action/Metadata(Pattern-Key/Recurrence-Count)
|
||||
- Pattern-Key 重复是信号:第一次记了,第二次就该解决了
|
||||
- Recurrence-Count 区分偶发一次性错误与需系统性解决的重复问题
|
||||
- 3月27日发现无对话日记忆文件缺失 → 推动流程优化:每次 Session 启动都检查并创建当天文件
|
||||
|
||||
## Key Quotes
|
||||
> "AI每次对话都是一张白纸。昨天说过不要用A方法,今天它照常用。" — 核心痛点
|
||||
> "错误只犯一次,第二次就知道怎么做对。" — self-improving 核心价值
|
||||
> "Pattern-Key 重复本身就是一个信号——第一次记了,第二次就该解决了。" — Pattern-Key 机制
|
||||
|
||||
## Key Concepts
|
||||
- [[Self-Improving Skill]]:结构化经验记录系统,self_improvement_log 工具写入 LEARNINGS.md/ERRORS.md
|
||||
- [[双层记忆架构]]:短期记忆(每日文件)+ 长期记忆(memory-lancedb-pro)+ self-improving(成长追踪)
|
||||
- [[Pattern-Key]]:经验记录的唯一标识键,格式如 cron.telegram-delivery,用于发现重复踩坑
|
||||
- [[Recurrence-Count]]:重复次数计数器,区分偶发错误与系统性问题
|
||||
- [[每日复盘]]:23:00 定时任务,读取当天 memory → self_improvement_log → 检查 Pattern-Key 重复 → 同步到长期记忆
|
||||
- [[记忆断层]]:无对话日不生成 memory 文件的问题,通过 Session 启动时强制检查解决
|
||||
|
||||
## Key Entities
|
||||
- [[OpenClaw]]:多 Agent 管理平台,支持 cron job、memory-lancedb-pro、self-improving skill
|
||||
|
||||
## Connections
|
||||
- [[OpenClaw]] ← 平台 ← [[self-improving skill]]
|
||||
- [[双层记忆架构]] ← 包含 ← [[Self-Improving Skill]]
|
||||
- [[双层记忆架构]] ← 包含 ← [[memory-lancedb-pro]]
|
||||
- [[Pattern-Key]] ← 核心机制 ← [[Self-Improving Skill]]
|
||||
- [[每日复盘]] ← 触发 ← [[OpenClaw cron]]
|
||||
Reference in New Issue
Block a user