A high-performance memory management framework that implements automated context compression and multi-tiered storage for long-form AI interactions.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install auto-summarization-loop
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 auto-summarization-loop 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 Auto-Summarization Loop?
The Auto-Summarization Loop is a sophisticated solution designed for high-fidelity AI personas to manage conversation context automatically. By utilizing Openclaw Skills, this system establishes a multi-level memory architecture consisting of core memory, working memory, and long-term memory. This approach ensures that the AI maintains its core persona while efficiently processing massive amounts of dialogue data without hitting token limits or incurring unnecessary costs.
This skill is particularly effective for developers building persistent AI agents that need to remember user preferences, historical events, and established consensus over weeks or months of interaction. It utilizes a sliding window strategy and dual-watermark triggers to balance performance and memory integrity, making it a cornerstone for professional-grade Openclaw Skills implementations.
Auto-Summarization Loop Use Cases
- Creating high-fidelity AI roleplay characters with persistent long-term memory.
- Optimizing token consumption to reduce API operational costs in enterprise chatbots.
- Managing complex multi-session dialogues where historical context is critical for task completion.
- Building structured user profiles that evolve naturally through conversation analysis.
How Auto-Summarization Loop Works
- The system initializes a MemoryConfig defining context limits, watermark ratios, and the summarization logic.
- Every new message triggers a watermark check to monitor the current token usage of the working memory.
- If the soft limit (70%) is reached, the skill initiates an asynchronous background compression of older messages into a long-term summary.
- If the hard limit (90%) is breached, a synchronous interception occurs to force immediate summarization and prevent context overflow.
- The prompt builder assembles a consolidated input for the LLM, combining the system prompt, historical summaries, user facts, and the active sliding window of recent messages.
Auto-Summarization Loop Setup
To integrate this module into your project, follow these initialization steps in your Python environment:
from scripts.memory_manager import (
AutoSummarizationLoop,
MemoryConfig,
MemoryBlock,
TriggerType
)
# Configure limits for your specific model
config = MemoryConfig(
max_tokens=200000, # Maximum context limit
soft_limit_ratio=0.7, # Async compression at 70%
hard_limit_ratio=0.9, # Sync compression at 90%
working_messages=20, # Keep last 20 messages intact
summary_max_tokens=2000, # Max length for generated summary
summarize_fn=your_summarize_func
)
# Initialize the loop
manager = AutoSummarizationLoop(config)
Auto-Summarization Loop Data Schema & Taxonomy
The skill organizes session data into a structured hierarchy to ensure rapid retrieval and persistence:
| Data Component | Type | Description |
|---|---|---|
| core_memory | String | The immutable System Prompt and persona definition. |
| working_memory | JSON Array | The sliding window containing the most recent conversation turns. |
| long_term_summary | String | The compressed historical context generated by the LLM. |
| user_facts | Object | Structured data points about the user (preferences, name, occupation). |
| last_update | Timestamp | Records the last time the memory block was modified. |
Auto-Summarization Loop Advanced Features
- Dual-Watermark Strategy: Seamlessly transitions between background async processing and emergency sync compression.
- Structured Persona Extraction: Automatically identifies and stores user facts, event timelines, and agreed-upon consensus during summarization.
- Extensible Summarization: Hook into various LLM providers to handle the summarization loop independently of the main chat model.
- Sliding Window Optimization: Maintains a precise number of recent messages to ensure immediate local context remains coherent while using Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
SaaS Churn Analysis Skill
A specialized AI skill for calculating complex SaaS retention metrics, building cohort tables, and identifying at-risk customers to improve net revenue retention.
AI Intelligent Image Generation
A high-performance image generation engine supporting text-to-image, image-to-image, and batch processing for professional creative workflows.
Bilibili Data Query Tool
A powerful data retrieval skill for Bilibili that allows AI agents to search videos, fetch creator statistics, and monitor real-time trends without requiring an API key.
Remotion Bozo
A comprehensive Remotion video project template and toolkit for building professional, React-powered videos with ease.
GTM Meeting Prep
A professional-grade intelligence tool that generates tailored pre-meeting briefs, account research, and attendee insights for sales and growth teams.
EdrawMind Mind Map Generation
A powerful integration that transforms structured Markdown or natural language into professional, editable EdrawMind maps with custom layouts and styles.
Featured*
