360 Secure SkillHub
360 安全技能中心

Auto Dream Light

面向 OpenClaw 的轻量记忆整理技能:扫描近期日记笔记,提取长期有价值的信息,路由到 MEMORY.md 或 memory/projects 等现有文件做增量更新,并写入 dream 执行日志,最后返回简要总结;保留用户原有记忆结构,无价值内容会明确跳过。

mrgyan
mrgyan
热度 40.66
最近更新 2026-05-11 17:45:27
分类:知识依赖:读写文件应用场景:笔记整理 / 摘要总结 / 工作流

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add auto-dream-light
Metadata
name
auto-dream-light
version
0.1.1
description
Lightweight, memory-safe Auto Dream workflow for OpenClaw that consolidates recent notes into existing memory files without replacing the user’s current memory structure.

Auto Dream Light

A lightweight, memory-safe Auto Dream skill for OpenClaw that works with the memory system the user already has.

What this skill is for

Use this skill when the user wants:

  • on-demand memory consolidation
  • a lightweight “dream” workflow
  • a dream log with execution history
  • a gradual path from manual to semi-automatic memory cleanup
  • memory consolidation without replacing the current MEMORY.md structure

This skill is intentionally conservative:

  • it does not rebuild the memory architecture
  • it does not introduce dashboards, indexes, or archives by default
  • it does not assume the user's memory follows a template
  • it prefers small, explicit, high-value updates

Why install it

  • works with existing MEMORY.md and memory/
  • makes conservative, incremental updates
  • avoids noisy or fake “productivity” output
  • keeps runs easy to review and audit
  • supports a clean path from manual runs to semi-automatic operation

Core idea

Instead of forcing a new memory system, this skill works with the one the user already has.

Default approach:

  1. scan recent daily logs
  2. extract durable, high-value information
  3. route items to the correct destination
  4. update memory incrementally
  5. record the run in a dream log
  6. return a concise summary to the user

Read these references when needed

  • Read references/adapted-plan.md before designing or changing the workflow.
  • Read references/manual-run.md when running the workflow manually.
  • Read references/semi-auto.md when the user wants a fixed trigger-based flow or wants to prepare for cron later.

Default file roles

Typical targets:

  • MEMORY.md → long-term facts, stable preferences, system conclusions, reusable lessons
  • memory/projects/** → project-specific context
  • memory/system/auto-dream-log.md → dream execution history only
  • memory/YYYY-MM-DD.md → raw daily notes; only add a consolidation marker when appropriate

Operating rules

  • Preserve the existing memory structure.
  • Prefer project files over stuffing everything into MEMORY.md.
  • Skip low-value chat, tests, and one-off noise.
  • Do not invent memory just to make a run look productive.
  • If there is nothing worth consolidating, explicitly skip and say so.

Trigger guidance

Typical trigger phrases:

  • "整理记忆"
  • "dream now"
  • "跑 dream"
  • "做一次记忆整理"
  • "跑一次适配版 auto dream"

If the user says dream details, show recent dream-log history instead of running a new consolidation.