Memory Dream is a structured memory classification and periodic consolidation system designed to keep Openclaw Skills agents efficient and context-aware.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install memory-dream-consolidation
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install memory-dream-consolidation using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Memory Dream?
Memory Dream provides a robust framework for managing an AI agent's long-term memory through structured classification and a periodic consolidation process known as a Dream Cycle. Inspired by advanced agent architectures, it ensures that your Openclaw Skills agents remain performant by preventing memory files from becoming bloated, contradictory, or disorganized. By categorizing information into user, feedback, project, and reference types, the system enables precise recall and significantly improves the agent's ability to follow complex user preferences over time.
Memory Dream Use Cases
- Initializing a new AI agent with a scalable long-term memory infrastructure.
- Performing periodic maintenance during heartbeat checks to prevent context overflow in long-running projects.
- Automatically pruning and consolidating memory files when they exceed the 200-line or 25KB limit.
- Resolving contradictions between past logs and current reality to maintain a single source of truth for Openclaw Skills.
How Memory Dream Works
- Trigger Gates: The system evaluates if three conditions are met: at least 3 days since the last Dream, 3 days of new daily notes, and no other Dream process currently in progress.
- Phase 1 - Orient: The agent reads the existing MEMORY.md and scans recent daily notes to understand the current memory landscape.
- Phase 2 - Gather: High-priority entries, including important decisions, success stories, and lessons learned, are extracted from ephemeral daily logs.
- Phase 3 - Consolidate: Information is merged into appropriate sections, relative dates are converted to absolute timestamps, and feedback is structured with Why and How logic.
- Phase 4 - Prune: The agent enforces strict size limits by removing completed projects and stale data, ensuring that Openclaw Skills remain focused and lightweight.
Memory Dream Setup
To integrate Memory Dream into your agent workflow, follow these steps:
- Add the memory classification table and Dream procedure to your AGENTS.md file.
- Initialize your MEMORY.md with a size limit header:
# > 📏 Limit: ≤ 200 lines / 25KB | Last Dream: YYYY-MM-DD
- Create or update
memory/heartbeat-state.jsonto include thelastDreamAtfield. - Add a Dream trigger check task to your HEARTBEAT.md file to automate the lifecycle within Openclaw Skills.
Memory Dream Data Schema & Taxonomy
Memory Dream organizes data across several files to maintain a clean metadata taxonomy:
| Component | File Path | Description |
|---|---|---|
| Long-term Memory | MEMORY.md | Categorized entries (User, Feedback, Project, Reference) with absolute dates. |
| State Tracking | heartbeat-state.json | Stores metadata like lastDreamAt and lastDreamResult. |
| Ephemeral Logs | memory/YYYY-MM-DD.md | Raw daily notes used as the source for the Gather phase. |
All feedback entries must follow a schema consisting of the Rule/Fact, the Why (reasoning), and the How to apply (contextual triggers).
Memory Dream Advanced Features
- Success-based recording ensures agents learn from effective patterns, not just past errors.
- Automated size enforcement prevents performance degradation by strictly maintaining a 200-line limit.
- Verification protocols require agents to check current file states before trusting recalled memory to prevent drift.
- Cross-file deduplication merges near-duplicate entries into a single cohesive record within Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
voice-minimax
A professional integration that converts text into natural speech using MiniMax TTS and delivers it as voice messages or files to Lark via lark-cli.
Complex Task Orchestrator
A robust framework for decomposing, executing, and recovering complex, multi-step AI agent tasks without losing context or data.
Skill Sync
A powerful utility to audit, deduplicate, and synchronize AI agent skills across multiple environments into a single canonical source of truth.
Doubao Copywriter Skill
An automated AI writing assistant that leverages the Doubao platform to generate high-quality copy for social media, blogs, and video scripts.
Claude Code Integration for OpenClaw
A professional-grade programming integration that bridges the Claude Code CLI with OpenClaw for autonomous software engineering tasks.
Skill Index
An automated inventory management tool that scans, categorizes, and indexes your entire collection of Openclaw Skills.
Featured*
