A sanitized, agent-generated design package for memory-system enhancement in OpenClaw-style agents.
一个经过净化处理、由 Agent 主导生成的 OpenClaw 风格记忆系统增强设计包。
This repository is a sanitized open-source export of an experimental memory-system project.
Its core direction is intentionally conservative:
- keep the existing file-first memory + LanceDB architecture,
- add Working Memory only as a runtime layer,
- use candidate -> episode-like summary -> LanceDB as a controlled bridge,
- use context packet as a manual retrieval aid,
- avoid creating a second long-term truth source.
- architecture and design documents
- migration and bridge plans
- Working Memory templates
- archive / candidate rules
- context packet templates
- sanitized examples
- publication, risk, and provenance documents
This repository does not include private runtime materials from the originating workspace, including:
- personal identity or persona files
- credentials or secrets
- private user logs or archives
- local databases
- screenshots, media artifacts, and unrelated experiments
/docs design docs and reports
/specs/working-memory Working Memory templates and rules
/specs/bridge bridge templates and examples
README.md bilingual overview
ANNOUNCEMENT.md public announcement
NOTICE.md publication notice
SOURCE_ATTRIBUTION.md provenance and origin notes
THIRD_PARTY_NOTICES.md upstream/open-source attributions
RISK_DISCLOSURE.md risks and operating boundaries
LICENSE repository license
This repository was assembled primarily by an AI agent under human instruction.
That means:
- design assumptions may still require human review,
- examples may reflect the originating workspace conventions,
- no production use should happen without security and privacy review.
This project is experimental.
Do not use it as-is to:
- auto-publish private memory logs,
- auto-inject historical memory into prompts without review,
- expose secrets, local paths, or user archives,
- treat bridge files as a new canonical database.
Read RISK_DISCLOSURE.md and docs/public-release-security-audit-2026-03-23.md.
- ANNOUNCEMENT.md
- docs/memory-system-v3-technical-architecture-2026-03-23.md
- docs/context-packet-manual-checklist-2026-03-23.md
- specs/working-memory/README.md
- SOURCE_ATTRIBUTION.md
这是一个经过净化处理的开源导出仓库，来源于一次记忆系统增强实验。
它的设计路线是刻意收束的：
- 保留现有 文件主账本 + LanceDB 主轴
- 将 Working Memory 仅作为运行时状态层
- 通过 candidate -> episode-like summary -> LanceDB 做受控桥接
- 将 context packet 作为人工组装的检索辅助包
- 明确避免形成第二套长期真相源
- 技术架构与设计文档
- 迁移与桥接方案
- Working Memory 模板
- archive / candidate 规则
- context packet 模板
- 已净化的示例文件
- 公告、来源说明与风险披露
本仓库不包含原始私域工作区中的以下内容：
- 个人身份 / 人格配置文件
- 凭据、密钥、令牌
- 私人对话日志与原始归档
- 本地数据库
- 截图、媒体制品与无关实验材料
/docs 设计文档与报告
/specs/working-memory Working Memory 模板与规则
/specs/bridge bridge 模板与示例
README.md 中英双语总览
ANNOUNCEMENT.md 对外公告
NOTICE.md 发布说明
SOURCE_ATTRIBUTION.md 来源与底层说明
THIRD_PARTY_NOTICES.md 第三方与上游引用说明
RISK_DISCLOSURE.md 风险与边界说明
LICENSE 开源协议
本仓库内容主要由 AI Agent 在人工指令下完成整理、编写与组装。
因此你应默认：
- 文档中可能仍有隐含假设
- 示例可能保留原始工作流习惯
- 真正投入生产前，必须由人类完成安全和隐私复核
这是一个实验性项目。
请不要直接将其用于：
- 自动公开私密记忆日志
- 在未经审查时自动注入历史记忆到 prompt
- 暴露密钥、本地路径、用户归档
- 将 bridge 层文件误当成新的正式数据库
请先阅读：
- ANNOUNCEMENT.md
- docs/memory-system-v3-technical-architecture-2026-03-23.md
- docs/context-packet-manual-checklist-2026-03-23.md
- specs/working-memory/README.md
- SOURCE_ATTRIBUTION.md
- local audit completed / 本地安全审计已完成
- sanitized export completed / 净化导出已完成
- public repository created and pushed / GitHub 公共仓库已创建并推送
- post-release security review in progress / 发布后安全复核已完成首轮