Token导航 LogoToken导航TokenDH.com
研究检索只读clawhub未标认证来源可访问clear审计提醒

memory-dreaming记忆做梦

Agent Skill

memory-dreaming 用于记录任务执行中的错误、用户纠正、经验和能力缺口,适合在 OpenClaw 中希望让 Agent 持续沉淀问题、修正和最佳实践时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

5,777

周安装

236

GitHub Stars

公开资料未说明

下载量

1,850
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:memory-dreaming(记忆做梦)
来源仓库:https://github.com/oryanmoshe/memory-dreaming
安装命令:
openclaw skills install memory-dreaming
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install memory-dreaming

简介

memory-dreaming 模拟代理自主记忆整合过程,定期从日志与学习中提炼信号并归档。

  • 适用于 OpenClaw 中实现类生物记忆巩固机制的长期代理进化场景。
  • 能从日常交互中提取模式、纠正偏差并形成稳定心智表征。
  • 安装命令为 openclaw skills install memory-dreaming,需持续访问会话记录权限。
  • 使用前应设置归档频率与容量上限,防止无限增长导致性能下降。

SKILL.md

name
memory-dreaming
description
Autonomous memory consolidation for OpenClaw agents — like REM sleep. Periodically gathers signal from daily logs, session transcripts, and learnings; consolidates into MEMORY.md; syncs structured knowledge to an Obsidian vault (or any markdown knowledge base); tracks plans; prunes stale entries. Use when: (1) setting up periodic memory maintenance, (2) manually triggering a dream cycle, (3) configuring Obsidian vault sync, (4) agent memory is getting noisy/contradictory and needs consolidation.

Memory Dreaming

Autonomous memory consolidation ("dreaming") for OpenClaw agents. Runs as a cron job, consolidates scattered daily notes into curated long-term memory, and syncs structured knowledge to an Obsidian vault.

Quick Start

  1. Install: clawhub install oryanmoshe/memory-dreaming
  2. Configure your vault path (optional): edit dreaming-config.json in your workspace
  3. Set up the cron: run scripts/setup-cron.sh
  4. Done — the agent dreams automatically every 8 hours

To trigger a dream manually, tell the agent: "Run a dream cycle now."

How It Works

The dream cycle has 4 phases, inspired by biological REM sleep and Claude Code's AutoDream:

Phase 1: Orient

Read current memory state — MEMORY.md, recent daily logs, learnings, dreaming log. Build a map of what exists and when it was last touched.

Phase 2: Gather Signal

Search for high-value information added since the last dream:

  • Daily logs (memory/YYYY-MM-DD.md) since last dream
  • Learnings (.learnings/*.md) — pending corrections, errors, best practices
  • Session transcripts — grep for corrections ("actually...", "no that's wrong"), decisions ("let's do X"), proper nouns, preferences
  • Plan files — scan workspace for task_plan.md files

Key: grep narrowly for high-signal patterns. Don't read full transcripts — that burns tokens for marginal value.

Phase 3: Consolidate

Update MEMORY.md with gathered signal:

  • Merge duplicate entries (same fact from 3 sessions → one entry)
  • Absolute dates — convert "yesterday" → "2026-03-25"
  • Delete contradicted facts (if preference changed, remove old one)
  • Remove stale entries (references to deleted files, completed tasks)
  • Promote high-priority learnings from .learnings/ to MEMORY.md

Phase 4: Sync

Push consolidated knowledge to external targets:

  • Obsidian vault (opt-in) — create/update notes with tags, wikilinks, full depth
  • Plan tracking — ensure every task_plan.md has a corresponding Plans/<name>.md in the vault
  • Dreaming log — write what changed, tokens used, duration

Gate

The cron fires on schedule but the dream cycle only executes if ≥6 hours have passed since the last dream (checked via dreaming-log.md timestamp). This prevents wasted runs when nothing has changed.

Configuration

Create dreaming-config.json in your workspace root to customize. All fields are optional — sensible defaults are used.

See assets/dreaming-config.json for the full schema with defaults.

Key options:

  • schedule — cron expression (default: "0 */8 * * *")
  • model — which model runs the dream (default: "anthropic/claude-sonnet-4-6")
  • gate.minHours — minimum hours between dreams (default: 6)
  • obsidian.enabled — enable vault sync (default: false)
  • obsidian.vaultPath — absolute path to Obsidian vault
  • delivery.mode"none" or "announce" changes to a channel

Obsidian Sync Details

When enabled, the sync phase:

  1. Compares MEMORY.md sections against existing vault notes
  2. Creates new notes in configured subfolders (People/, Projects/, Plans/, Tools/)
  3. Updates existing notes with new information (appends, doesn't overwrite)
  4. Follows formatting rules: tags on first line, [[wikilinks]] throughout, full depth content
  5. Tracks plans: scans for task_plan.md files → creates/updates Plans/<name>.md

For detailed sync behavior, see references/obsidian-sync.md.

Manual Dream

Tell the agent any of these:

  • "Run a dream cycle"
  • "Consolidate memory"
  • "Dream now"
  • "Sync to obsidian"

The agent reads this skill and executes the 4-phase cycle immediately, ignoring the gate.

Setup Script

# Creates the cron job in OpenClaw
bash scripts/setup-cron.sh

The script reads dreaming-config.json (or uses defaults) and creates an isolated agentTurn cron job. See scripts/setup-cron.sh for details.

Architecture

For the detailed 4-phase architecture, design decisions, and how this compares to Claude Code AutoDream, see references/architecture.md.

What This Skill Does NOT Do

  • Does not replace QMD indexing (QMD handles search, this handles consolidation)
  • Does not delete source files (daily logs are never removed, only consolidated from)
  • Does not modify source code or project files
  • Does not run without explicit setup (cron must be created via setup script)

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

73.02%
按下载量换算1,351

安全审计

VirusTotal

可疑

ClawScan

可疑

Static analysis

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在 warning/failed 结果,不能写成本站确认安全。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills