An autonomous intelligence layer that enables agents to learn from mistakes, reflect on task outcomes, and manage a tiered memory system.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install self-improving-local
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 self-improving-local 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 Self-Improving + Proactive Agent?
The Self-Improving + Proactive Agent skill provides a sophisticated framework for AI agents to evolve their capabilities through self-reflection, self-criticism, and organized memory management. Instead of remaining static, an agent equipped with this skill actively evaluates its own work, identifies patterns in user corrections, and optimizes its internal knowledge base. This creates a compounding intelligence effect where the agent becomes more precise and aligned with user preferences over time.
Built on a tiered storage architecture, the skill ensures that critical, frequently used information remains highly accessible in HOT memory, while project-specific or domain-specific data is kept in WARM storage. By automating the lifecycle of knowledge—from initial correction to long-term archiving—this skill allows Openclaw Skills to maintain high performance across diverse workspaces without manual configuration maintenance.
Self-Improving + Proactive Agent Use Cases
- When a user provides explicit corrections like "No, that is not right" or "I prefer X over Y."
- After completing a multi-step workflow where the agent needs to verify if the outcome met the initial intent.
- When identifying repeatable patterns or preferences that should be applied to future tasks automatically.
- To maintain separate learning contexts for different projects, such as specific coding styles for a Flutter app versus a Python backend.
- To prevent the recurrence of bugs or mistakes by logging and promoting lessons learned from failures.
How Self-Improving + Proactive Agent Works
- Signal Detection: The agent monitors interactions for explicit corrections, preference signals, or repeated instructions.
- Self-Reflection: Upon completing significant work, the agent pauses to evaluate the outcome versus the intent and identifies potential improvements.
- Logging: New lessons are initially recorded in
corrections.mdor scoped project files within the~/self-improving/directory. - Tiered Promotion: Patterns that are successfully applied three times within seven days are promoted to HOT memory (
memory.md) for constant availability. - Maintenance & Compaction: A recurring heartbeat process reviews the memory state, demoting unused patterns to WARM storage or archiving them to COLD storage to maintain context efficiency.
- Contextual Retrieval: During execution, the agent loads relevant memory namespaces based on the current project or domain to ensure accurate and personalized responses.
Self-Improving + Proactive Agent Setup
To initialize the self-improving memory architecture, ensure the base directory exists and run the setup workflow. Openclaw Skills users should follow these steps:
# Create the required directory structure
mkdir -p ~/self-improving/projects/ ~/self-improving/domains/ ~/self-improving/archive/
# Initialize the memory templates
touch ~/self-improving/memory.md
touch ~/self-improving/corrections.md
After creating the directory, integrate the steering rules into your workspace by referencing the setup.md guide provided within the skill. This will configure your AGENTS and HEARTBEAT.md files to enable proactive self-learning.
Self-Improving + Proactive Agent Data Schema & Taxonomy
The skill organizes information in a hierarchical, tiered structure to optimize token usage and relevance:
| File / Path | Tier | Description |
|---|---|---|
memory.md |
HOT | Global rules and confirmed preferences (limit 100 lines); always loaded. |
projects/ |
WARM | Per-project learnings and context-specific rules; loaded on match. |
domains/ |
WARM | Domain-specific expertise (e.g., code, writing, communications). |
archive/ |
COLD | Historical or decayed patterns; loaded only on explicit query. |
corrections.md |
LOG | The last 50 explicit user corrections for evaluation. |
heartbeat-state.md |
STATE | Metadata tracking last maintenance runs and promotion/demotion actions. |
Self-Improving + Proactive Agent Advanced Features
- Automatic Promotion/Demotion: Implements a usage-based lifecycle where active patterns stay HOT and inactive ones are archived after 90 days.
- Namespace Isolation: Keeps project-specific logic contained, preventing cross-project configuration pollution.
- Transparent Citations: The agent can explicitly cite the memory source for its actions (e.g., "Using X from projects/foo.md").
- Security Boundaries: Hard-coded restrictions prevent the storage of credentials, health data, or sensitive third-party information.
- Conflict Resolution: Logic to handle contradictory rules by prioritizing project-specific instructions over global defaults.
SKILL.md
Loading
Related Openclaw Skills
AceToolz QR Code Generator
The AceToolz QR Code Generator enables AI agents to instantly create and share hosted QR code images for any URL or text block.
AceToolz Password Generator
A secure, real-time password generation skill powered by the AceToolz API for AI agents.
Claude Usage Analyzer
A powerful tool to track, analyze, and optimize your Claude Code token consumption and project costs.
EO Quickstart
A streamlined onboarding skill designed to configure the Everything Openclaw plugin and launch multi-expert AI collaboration projects in minutes.
Bilibili Automation Skill
An AI-powered automation tool for Openclaw Skills that enables searching, subtitle extraction, and content summarization for Bilibili videos.
DingTalk CLI Auto Skill
An automation powerhouse for DingTalk that leverages the official dws CLI to manage enterprise communications, schedules, and tasks through an AI agent.
Featured*
