A dual-layer Openclaw skill that empowers AI agents to learn from user feedback while maintaining proactive task momentum.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install self-improving-proactive-agent
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-proactive-agent 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 is a unified behavioral framework designed to bridge the gap between static memory and autonomous action. By merging two critical layers of agent logic, this skill allows an agent to evolve based on explicit user corrections while simultaneously pushing work forward without constant supervision. It is a core component for users of Openclaw Skills who require agents that don't just remember better, but operate with higher levels of follow-through and context awareness.
This skill provides a robust architecture for capturing durable preferences and managing active task states. It ensures that an agent remains productive by identifying the next useful move, recovering context independently before asking the user for help, and strictly adhering to safety boundaries during proactive execution.
Self-Improving Proactive Agent Use Cases
- When an agent needs to capture and apply durable user preferences or coding style corrections across multiple sessions.
- For managing complex, multi-step tasks where the agent must maintain state and recover context after a drift.
- In scenarios requiring autonomous heartbeat behavior to follow up on blockers or pending decisions.
- When a unified behavior model is needed to replace separate, overlapping learning and proactivity skills.
How Self-Improving Proactive Agent Works
- The agent monitors interactions for explicit evidence such as direct corrections, stated preferences, or repeated successful workflows.
- Captured information is routed to specific storage paths: durable lessons enter the self-improving memory, while active task data is stored in proactivity session files.
- Before requesting input, the agent performs context recovery by reading the working buffer and session state to identify the missing delta.
- The agent pushes the next useful move—such as preparing drafts or checking for blockers—while staying within defined safety boundaries.
- A promotion and decay logic periodically reviews logs, promoting frequent corrections to durable rules and archiving unused data to keep the context window clean.
Self-Improving Proactive Agent Setup
To deploy this skill within your agent's environment, you must initialize the required directory structure and configuration files. Execute the following commands in your workspace:
# Create directory structure
mkdir -p ~/self-improving/projects ~/self-improving/domains ~/self-improving/archive
mkdir -p ~/proactivity/memory
# Initialize core state files
touch ~/self-improving/memory.md ~/self-improving/corrections.md
touch ~/proactivity/session-state.md ~/proactivity/heartbeat.md
Ensure that your agent has read and write permissions for these paths as part of your Openclaw Skills integration.
Self-Improving Proactive Agent Data Schema & Taxonomy
The skill utilizes a structured file system to separate long-term learning from short-term execution state:
| Directory | Primary Files | Data Type | Purpose |
|---|---|---|---|
~/self-improving/ |
memory.md, corrections.md |
Markdown | Durable preferences and recent learned lessons. |
~/proactivity/ |
session-state.md |
Markdown | Tracks objective, last decision, blockers, and next move. |
~/proactivity/memory/ |
working-buffer.md |
Markdown | Volatile context for fragile or tool-heavy operations. |
~/self-improving/ |
index.md |
Markdown | Storage map and topic index for domain-scoped learning. |
Self-Improving Proactive Agent Advanced Features
- Intelligent Promotion Logic: Automatically promotes a lesson to HOT memory if repeated 3 times within 7 days.
- Proactive Heartbeat: Detects missing next moves and surfaces prepared recommendations without user prompting.
- Context Recovery Flow: Advanced logic to read session buffers and state files before asking the user to restate work.
- Automated Decay: Automatically demotes or archives unused learnings after 30 to 90 days to maintain performance.
- Safety Guardrails: Hard boundaries preventing autonomous financial transactions, data deletion, or public messaging without explicit approval.
SKILL.md
Loading
Related Openclaw Skills
Vocab Deep Dive
A comprehensive vocabulary analysis tool designed to provide exhaustive linguistic insights and exam-specific preparation strategies for English learners.
yfinance Skill
A comprehensive financial data client for querying US and Hong Kong stock markets via Yahoo Finance integration.
Aistro Astrology Consultation
Aistro is a professional astrology AI skill providing personalized birth charts, compatibility reports, and daily horoscopes through precise astronomical calculations.
Report Generator
A comprehensive automation skill for transforming raw financial and business data into professional, chart-heavy PDF reports and dashboards.
Binance Spot WebSocket Skill
A specialized skill for subscribing to and capturing real-time Binance Spot public market data streams through the uxc WebSocket transport.
OpenClaw Cost Optimization Consultant
An intelligent advisor for Openclaw Skills designed to analyze AI model pricing and provide strategic recommendations for reducing API expenditures.
Featured*
