Self Memory Manager

管理 Claude 的记忆和工作流程优化。包括：(1) Context 使用管理 (2) 重要信息存档 (3) 定时总结 (4) 工作文件夹维护 用于：context 超过 80%、重要信息需要记录、每日总结、清理旧 session

Security Scan

The skill's instructions tell the agent to read and persist potentially sensitive agent/config data (e.g., ~/.openclaw/openclaw.json and account/API info) to a Desktop folder, but the registry metadata declares no config access or credentials — this mismatch is suspicious and could lead to unintended disclosure or persistence of secrets.

Purpose & Capability

The named purpose (manage memory/context, periodic summaries, and maintain a work folder) can legitimately include writing notes to disk, but the SKILL.md explicitly copies ~/.openclaw/openclaw.json and instructs storing '账号信息' and 'API 配置变化'. The skill metadata declares no required config paths or credentials, so accessing the agent's config file and account/API data is disproportionate and not declared.

Instruction Scope

The runtime instructions explicitly reference system paths and file operations: copying ~/.openclaw/openclaw.json to ~/Desktop/小牛马的工作文件夹/ and reading/writing multiple files. They also direct the agent to archive 'account information' and 'API configuration changes' — actions that collect and persist sensitive data beyond simple summary text. There is no explicit user-consent step in the workflow for these file reads/writes.

Install Mechanism

This is instruction-only with no install spec and no code files, so there is no additional install-time risk or external downloads. That lowers risk compared with skills that fetch and execute code.

Credentials

The skill declares no required environment variables or config paths, yet the instructions access a platform-specific config file (~/.openclaw/openclaw.json) and say to archive account/API info. Requesting or reading such data should be declared; absence of declared credentials/config access is a mismatch and raises the risk of secret exposure.

Persistence & Privilege

Although the skill is not marked 'always: true', the instructions persist files under the user's Desktop and create a stable memory folder (memory/*.md), including sensitive items. Persisting agent/credential-related data to the user's file system (and possibly a synced Desktop) is a privilege that should be explicit and consented to; the skill does not document that.

Scan Findings in Context

[no_regex_findings] expected: The static regex scanner found nothing because this is an instruction-only skill (no code files). That is expected, but absence of findings is not evidence that the instructions are safe — the SKILL.md itself contains concerning file operations.

What to consider before installing

This skill asks the agent to read and store potentially sensitive data (the agent's own config: ~/.openclaw/openclaw.json, 'account information', and 'API configuration changes') but the registry metadata doesn't declare access to those files or credentials. Before installing or enabling this skill: 1) Inspect the actual contents of ~/.openclaw/openclaw.json to see if it contains secrets you wouldn't want copied to a Desktop folder. 2) Decide whether you want automated writes to ~/Desktop/小牛马的工作文件夹/ (which may be synced to cloud services). 3) Ask the author to explicitly declare required config paths/credentials and add an explicit user-consent/confirmation step before any archival of sensitive items. 4) If you must use it, modify the workflow to redact secrets (or only store non-sensitive summaries) and require explicit user approval before saving agent/config data. If you cannot confirm these points, treat the skill as risky and avoid installing or enabling it with autonomous invocation.

Claude 的自我记忆管理工作流。

核心规则

Context 管理

• 预警线: 80% - 超过时主动提醒用户存档
• 危险线: 150% - 必须存档后才能继续
• 定期检查: 每隔一段时间检查 context 使用情况

重要信息存档

存档位置: ~/Desktop/小牛马的工作文件夹/

存档内容:

• API 配置变化
• 账号信息
• 任务进度
• 项目状态
• 工作规则更新

存档格式:

## 日期-时间
### 任务/项目
- 完成项
- 待办项

定期总结

• 重要进展后立即存档
• 每天结束时做简短总结
• 存到 memory/日期.md

文件夹结构

~/Desktop/小牛马的工作文件夹/
├── 工作笔记.md      # 主记录
├── 工作规则.md     # 本规则
├── memory/         # 按日期记忆
│   └── 2026-03-16.md
└── [项目文件夹]/

工作流程

1. 定期检查 Context

当前 context > 80%?
→ 是 → 提醒用户存档
→ 否 → 继续

2. 重要信息存档

完成重要任务后?
→ 是 → 立即存档到工作文件夹
→ 否 → 继续

3. 查找信息

用户问之前的内容?
→ 检查当前对话
→ 无 → 查工作文件夹

使用场景

触发条件	动作
Context > 80%	提醒存档
完成重要任务	自动存档
用户问之前的事	查文件夹
每天结束	建议做总结

快速命令

# 存档
cp ~/.openclaw/openclaw.json ~/Desktop/小牛马的工作文件夹/

# 查看状态
cat ~/Desktop/小牛马的工作文件夹/工作笔记.md

---

本技能帮助 Claude 自我管理记忆，避免 context 溢出同时保证重要信息不丢失。