backup

OpenClaw 会话重置-存档-备份完整工作流管理。自动化处理 Token 超限、闲置超时、手动重置时的四层数据存档体系，支持 GitHub/OneDrive 双备份。 TRIGGERS: - 用户说"设置存档工作流"、"配置自动备份" - 需要建立会话生命周期管理机制 - 想要自动化重置-存档-备份流程

Security Scan

The skill's stated purpose (session archive + GitHub/OneDrive backup) matches the instructions in broad strokes, but it omits required credential declarations, includes steps that access and upload sensitive session data, and contains inconsistent guidance about automatic resets and confirmation — worth caution before installing.

Purpose & Capability

The skill claims GitHub/OneDrive backups and session lifecycle management — that logically requires access to session history, local filesystem paths (workspace, OneDrive) and a GitHub token. However the registry metadata lists no required env vars/config paths or binaries. The absence of declared credentials/config requirements is inconsistent with its actual needs.

Instruction Scope

SKILL.md instructs the agent to read full session history (via sessions_history or an internal API), write many local files (raw-history, MEMORY.md, config/ files), run PowerShell scripts, and push data to GitHub/OneDrive. The references/templates even include a step that performs '自动重置（无需用户确认）' (automatic reset without user confirmation), which conflicts with the Phase 1/Phase 2 'ALWAYS ask and wait for confirmation' requirement — this is scope creep and an explicit behavioral inconsistency.

Install Mechanism

This is an instruction-only skill with no install spec or bundled binaries — low installation risk from third-party code downloads. However runtime instructions create and execute scripts on the host, which moves risk to runtime behavior rather than install-time.

Credentials

The workflow requires sensitive access (session history, config files like SOUL.md/USER.md, local OneDrive directories) and a GitHub token, but the skill metadata does not declare any required environment variables or credentials. Instructions also recommend embedding tokens into Git remote URLs (e.g., https://[TOKEN]@github.com/...), which is insecure and increases risk of token leakage. Requesting broad access to session data is proportionate to backup functionality only if credential handling and user consent are explicit — those are missing.

Persistence & Privilege

The skill is not marked always:true, but it describes autonomous heartbeat-driven checks that can trigger immediate backups and resets. Combined with the ability to read full session histories and push to external services, autonomous invocation amplifies risk. There is also an internal inconsistency about whether automatic reset requires user confirmation, which affects the level of privilege the skill exercises.

What to consider before installing

Before installing or enabling this skill, confirm these points with the author and take precautions: - Credentials and storage: Ask exactly how the GitHub token is provided, stored, and scoped. Do not accept workflows that embed tokens into remote URLs. Prefer using a token stored in a secure env var with minimal repo permissions (only the repo push scope required), or a deploy key with narrowly scoped access. - Explicit consent for resets: Resolve the contradiction between 'ALWAYS ask' and the template's 'automatic reset without confirmation'. Require explicit, per-reset user confirmation if you do not want automatic destructive resets. - Inspect scripts before execution: The skill instructs creation/execution of PowerShell scripts (generate-ai-summary.ps1, backup.ps1). Get the actual script contents and review them for any network calls, credential exfiltration, or commands that modify other system components before running. - Limit data uploaded: Confirm what files will be uploaded. Backing up full raw-history and config files can leak sensitive personal data; prefer selective backups or private repos/storage and encryption at rest. - Run in an isolated environment first: Test the workflow in a disposable user account or VM so resets and file writes cannot affect your primary environment. - Logging and rollback: Ensure the skill records operations in a local log and provides a way to abort or roll back an unintended reset or upload. - Additional info that would raise confidence: (1) explicit declaration of required env vars and how tokens are handled, (2) the exact PowerShell script sources to review, (3) a guarantee that automatic resets are disabled unless explicitly permitted, and (4) guidance that GitHub pushes avoid embedding tokens in URLs and use secure auth methods. If the author cannot provide these assurances and the script contents, treat the skill as high-risk and do not enable it with access to real session data or production environments.

Session Archive & Backup System

完整的 OpenClaw 会话生命周期管理工作流，实现从原始对话到云端备份的四层数据体系。

CRITICAL: 3-Phase Setup Workflow

NEVER 直接修改配置。ALWAYS 完成 Phase 1（需求确认）和 Phase 2（环境检查）， 并 WAIT for user confirmation 后再执行 Phase 3（配置实施）。

---

Phase 1: 需求确认（ALWAYS ASK FIRST）

询问用户：

"你希望配置哪种类型的存档备份方案？"

场景类型	描述	适用情况
S1 - 全新设置	从零搭建完整工作流	刚安装 OpenClaw，无现有配置
S2 - 已有配置	已有部分组件，需要整合	有零散脚本，需要体系化
S3 - 迁移升级	从其他方案迁移	使用其他备份方案，想切换到本工作流

询问用户：

"你的备份目标是什么？"

备份目标	说明
GitHub	备份到 GitHub 仓库（需要配置 token）
OneDrive	备份到本地 OneDrive 目录
双备份	同时备份到 GitHub + OneDrive（推荐）

---

Phase 2: 环境检查（ASK AFTER PHASE 1）

诊断问题，确定配置路径：

1. "当前 workspace 是否有 memory/ 目录？"

   • YES → 检查现有结构
   • NO → 需要创建完整目录结构

2. "是否有现有的 HEARTBEAT.md？"

   • YES → 需要合并或备份现有配置
   • NO → 创建新的 HEARTBEAT.md

3. "是否有 GitHub 备份仓库？"

   • YES → 配置 remote URL
   • NO → 提供创建仓库指引

4. "OneDrive 路径是否可用？"

   • YES → 确认路径 C:\Users\[username]\OneDrive\
   • NO → 跳过 OneDrive 备份

确认以上信息后，向用户展示 配置计划摘要，等待确认。

DO NOT PROCEED TO PHASE 3 UNTIL CONFIRMED.

---

Phase 3: 配置实施

用户确认后，执行以下步骤：

Step 1: 创建目录结构

workspace/
├── memory/
│   ├── session-backups/      # 结构化存档（AI摘要）
│   ├── raw-history/          # 原始对话（JSONL）
│   ├── subagents/            # 子Agent输出
│   ├── index.json            # 快速索引
│   ├── reset-log.md          # 重置日志
│   └── backup-status.json    # 备份状态
├── scripts/
│   ├── generate-ai-summary.ps1   # AI摘要生成
│   └── backup.ps1                # 备份脚本
├── config/                   # 需备份的配置文件
│   └── (SOUL.md, USER.md, etc.)
└── HEARTBEAT.md              # 核心工作流定义

Step 2: 配置 HEARTBEAT.md

创建/更新 HEARTBEAT.md，包含：

• Token 超限检测（默认阈值：70k/100k）
• 闲置超时检测（默认：120 分钟）
• AI 摘要生成调用
• 四层数据存档流程
• 备份触发逻辑

Step 3: 创建核心脚本

scripts/generate-ai-summary.ps1

• 读取会话历史
• 生成结构化摘要
• 标记差异 [NEW]/[UPD]/[KEEP]

scripts/backup.ps1

• 备份到 GitHub
• 备份到 OneDrive
• 更新 backup-status.json

Step 4: 初始化索引文件

memory/index.json

• 配置项快速查询
• 标签索引
• 变更历史追踪

---

工作流详解

触发条件

触发场景	条件	执行动作
Token 超限	> 70k/100k	存档 → 同步 MEMORY.md → 自动重置
闲置超时	120 分钟无活动	预存档 → 同步 MEMORY.md → 重置
手动重置	用户发送"重置"	存档 → 同步 MEMORY.md → 重置

四层数据流转

Layer 1: 原始对话 (Raw)
└── memory/raw-history/*.jsonl
    └── 完整消息记录（每次重置前自动保存）

Layer 2: 结构化存档 (Structured)
└── memory/session-backups/*.md
    └── AI 生成摘要 + 差异标记 [NEW]/[UPD]/[KEEP]

Layer 3: 长期记忆 (Memory)
└── MEMORY.md
    └── 跨会话保持的关键决策、待办、配置

Layer 4: 云端备份 (Cloud)
├── GitHub: [username]/openclaw-backup
└── OneDrive: C:\Users\[username]\OneDrive\openclaw-backup

执行流程

触发条件满足
      ↓
┌─────────────────┐
│ 1. 生成 AI 摘要  │ ← 提取关键决策、待办、主题
│ 2. 对比上次存档  │ ← 标记 [NEW]/[UPD]/[KEEP]
│ 3. 保存结构化存档 │ → memory/session-backups/*.md
└─────────────────┘
      ↓
┌─────────────────┐
│ 4. 保存原始对话  │ → memory/raw-history/*.jsonl
└─────────────────┘
      ↓
┌─────────────────┐
│ 5. 同步长期记忆  │ ← 更新 MEMORY.md
└─────────────────┘
      ↓
┌─────────────────┐
│ 6. 更新索引     │ ← 更新 memory/index.json
└─────────────────┘
      ↓
┌─────────────────┐
│ 7. 记录重置日志  │ ← 更新 memory/reset-log.md
└─────────────────┘
      ↓
┌─────────────────┐
│ 8. 标记待备份   │ ← pendingChanges = true
└─────────────────┘
      ↓
┌─────────────────┐
│ 9. 执行重置     │ ← 会话清零，重新开始
└─────────────────┘
      ↓
新会话开始 → 读取 MEMORY.md → 汇报上次状态

---

关键配置参数

配置项	默认值	说明
token_threshold	70k/100k	Token 超限触发存档重置
idle_timeout	120 分钟	闲置超时触发存档重置
backup_interval	12 小时	定期备份间隔
pending_changes	true/false	重要变更立即备份标记

---

存档格式模板

# 会话备份 - YYYY-MM-DD HH:MM

## 统计信息
- Token 使用量: XXk/100k
- 触发原因: [token/idle/manual]
- 主要话题: [摘要]

## 关键决策（差异标记）
- [NEW] **新增决策**: 描述
- [UPD] **更新决策**: 描述（原：xxx）
- [KEEP] **保持决策**: 无变化

## 待办事项（当前状态）
- [ ] 进行中
- [x] 已完成

## 与上次存档的差异摘要
- 新增决策: X 个
- 更新决策: X 个
- 新增待办: X 个
- 完成待办: X 个

## 完整上下文引用
- 上次存档: [文件名]
- 长期记忆: MEMORY.md
- 快速索引: index.json

## 已安装/配置的技能
- [skill-name] v1.0.0

## 下次继续
[需要跟进的事项]

---

使用示例

手动触发存档

用户：存档当前会话
→ 执行完整存档流程
→ 保存到 memory/session-backups/manual-YYYY-MM-DD-HHMM.md

查看存档历史

用户：查看存档记录
→ 读取 memory/reset-log.md
→ 展示最近的重置和存档记录

查询配置

用户：Token 超限阈值是多少？
→ 查询 memory/index.json
→ 返回：70k/100k

---

故障排查

问题：存档未触发

检查：

1. HEARTBEAT.md 是否正确配置
2. token 阈值设置是否正确
3. 检查 memory/reset-log.md 是否有记录

问题：备份失败

检查：

1. GitHub token 是否有效
2. OneDrive 路径是否存在
3. 查看 backup-status.json 中的失败记录

问题：AI 摘要生成失败

检查：

1. scripts/generate-ai-summary.ps1 是否存在
2. 是否有足够的会话历史
3. PowerShell 执行权限

---

Reference Files

• references/workflow-diagram.md — 完整工作流程图
• references/config-templates.md — 配置文件模板
• references/troubleshooting.md — 详细故障排查指南