A systematic maintenance skill that organizes, prunes, and validates AI agent memory files to prevent information decay.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install memory-sorting
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-sorting 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-Sorting?
Memory-Sorting acts as a comprehensive maintenance engine for AI long-term memory, specifically designed to optimize MEMORY.md and the topics/ directory. In the ecosystem of Openclaw Skills, this tool ensures that an agent's knowledge base remains coherent by treating information management like a structured wardrobe organization. It systematically scans for technical debt within memory files, providing a clear path to high-fidelity data.
The skill provides a vital layer of governance for AI-driven workflows by identifying memory rot—outdated, conflicting, or redundant information that can lead to agent hallucinations or inefficient processing. By implementing a strict scan-and-approve workflow, it keeps the human-in-the-loop while automating the tedious task of knowledge pruning.
Memory-Sorting Use Cases
- Regular weekly maintenance of complex AI memory repositories to ensure high retrieval accuracy.
- Resolving contradictions between old project requirements and current daily logs.
- Detecting and deleting orphaned files that are no longer indexed or relevant in the primary memory map.
- Reducing token overhead by consolidating fragmented topics and enforcing strict line and size limits.
How Memory-Sorting Works
- Perform a systematic scan of the topics/ directory and MEMORY.md to build a comprehensive file inventory and metadata map.
- Execute a six-category problem detection phase focusing on duplicates, expiration, contradictions, orphans, fragmentation, and storage limits.
- Verify the reality of memory entries by cross-referencing documented paths against the actual file system and current configurations.
- Generate a detailed health report and cleanup proposal for user review, requiring an explicit approval gate before any file modification.
- Execute approved modifications sequentially across all affected files and generate a final status report with execution details.
Memory-Sorting Setup
To utilize Memory-Sorting, ensure your agent has the necessary file system permissions enabled within your configuration. This skill is built to integrate seamlessly with the standard Openclaw Skills architecture.
# Trigger the sorting process manually via command
/memory-sorting
The skill requires FileRead and FileWrite access to modify documentation, and exec permission to perform grep-based global searches for conflict detection.
Memory-Sorting Data Schema & Taxonomy
Memory-Sorting organizes and validates data based on the following structural taxonomy:
| Component | Type | Data Elements Analyzed |
|---|---|---|
| topics/*.md | Markdown | frontmatter (name, type, mtime), full content body |
| MEMORY.md | Markdown | Index pointers, line counts (max 200), byte size (max 25KB) |
| memory/*.md | Daily Logs | Timestamps and latest conclusions used as a timeline reference |
| Execution Reports | JSON/MD | Health scores, PO (Must-do) tasks, and validation results |
Memory-Sorting Advanced Features
- Intelligent Health Scoring: Evaluates memory quality on a 0-100 scale across index completeness, timeliness, and thematic aggregation.
- Reality Verification Engine: Automatically identifies and proposes the removal of memories referencing files or configurations that no longer exist in the codebase.
- Selective Approval Logic: Support for granular control where users can approve all changes, reject specific numbered IDs, or add exceptions.
- Context-Aware Exclusions: Smartly ignores false positives such as shared source URLs, contact info in different contexts, or recent TODO lists within a 7-day window.
SKILL.md
Loading
Related Openclaw Skills
WeChat MP Reader
A production-ready skill for AI agents to directly extract titles and full body text from WeChat Official Account links using Playwright.
Education Skill
A local, zero-dependency learning assistant that generates 7-day study roadmaps and interactive quizzes for developers.
Mindmap
A terminal-based utility to convert indented text lists into structured ASCII tree diagrams for visualization and documentation.
Study Plan Generator
An AI-driven scheduling tool that generates scientifically optimized study plans for exams, certifications, and daily learning.
OpenClaw Trust Validation
A robust validation framework that prevents OpenClaw agents from blindly trusting outdated or incorrect memory during decision-making.
Knab
A specialized on-chain research agent for the AIMS protocol that provides deep visibility into vault yields, reserves, and transaction data across multiple blockchains.
Featured*
