A systematic framework for AI agents to log errors, track user corrections, and evolve through persistent project memory.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install self-improving-agent-v2
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-agent-v2 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-Improvement Agent Skill?
The Self-Improvement skill is a robust meta-learning framework designed for AI coding agents to ensure they never make the same mistake twice. By capturing unexpected command failures, direct user corrections, and discovered best practices into a structured .learnings/ directory, this tool bridges the gap between ephemeral chat sessions and permanent project knowledge. It is a core component within the Openclaw Skills ecosystem, enabling agents to build a soul and a memory that persist across different development environments.
This skill transforms every interaction into a potential optimization. When an agent encounters a knowledge gap or a missing capability, it documents the event using standardized templates. These entries can then be promoted to workspace-level configuration files such as SOUL.md or AGENTS.md, ensuring that subsequent sessions benefit from previously hard-won insights and refined behavioral guidelines.
Self-Improvement Agent Skill Use Cases
- A terminal command or script execution fails unexpectedly during a coding task.
- A user provides a direct correction like "Actually, this project uses ESM, not CommonJS."
- An agent identifies a recurring workflow pattern that could be simplified or hardened.
- Documenting feature requests or missing tool capabilities for future implementation.
- Promoting temporary session-specific learnings into permanent project memory files like CLAUDE.md or .github/copilot-instructions.md.
How Self-Improvement Agent Skill Works
- The agent detects a trigger such as a tool error, user correction, or a discovered best practice.
- A structured entry is generated and appended to the relevant file in the .learnings/ directory (ERRORS.md, LEARNINGS.md, or FEATURE_REQUESTS.md).
- Each entry is assigned a unique ID (e.g., LRN-YYYYMMDD-XXX) and metadata including priority, area, and status.
- During periodic reviews or session starts, the agent scans these logs to avoid repeating past mistakes.
- High-value, recurring learnings are promoted to workspace-level files or extracted into entirely new, reusable skills.
Self-Improvement Agent Skill Setup
Install the skill via the ClawdHub registry to integrate it with your Openclaw Skills environment:
clawdhub install self-improving-agent
Alternatively, perform a manual installation:
git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent
Finally, initialize the workspace directory to begin logging:
mkdir -p ~/.openclaw/workspace/.learnings
Self-Improvement Agent Skill Data Schema & Taxonomy
The skill organizes data within a .learnings/ directory using three primary Markdown files to categorize insights:
| File | Purpose | Key Metadata |
|---|---|---|
LEARNINGS.md |
Corrections and best practices | Category (e.g., knowledge_gap), Pattern-Key, Recurrence-Count |
ERRORS.md |
Command failures and stack traces | Command Attempted, Error Message, Context, Reproducibility |
FEATURE_REQUESTS.md |
Missing agent capabilities | Complexity Estimate, User Context, Frequency |
All entries use a standardized schema including Logged (timestamp), Priority (low to critical), and Status (pending, resolved, in_progress, or promoted).
Self-Improvement Agent Skill Advanced Features
- Automatic Skill Extraction: Built-in helper scripts to transform verified, non-obvious learnings into standalone Openclaw Skills.
- Hook Integration: Supports UserPromptSubmit and PostToolUse hooks to trigger error detection and evaluation evaluation automatically.
- Inter-Session Communication: Specialized tools like sessions_send and sessions_history to share learnings between different active agents.
- Recurring Pattern Detection: Uses stable Pattern-Keys to track, deduplicate, and escalate issues that occur across multiple tasks.
- Promotion Workflow: Rules-based logic for elevating local logs to global project memory files like SOUL.md, TOOLS.md, or AGENTS.md based on recurrence counts.
SKILL.md
Loading
Related Openclaw Skills
FlowMind Productivity Manager
A unified productivity workspace that integrates goals, tasks, notes, and contacts into a streamlined AI-managed workflow.
Company Research Agent Skill
A multi-source company research tool for generating structured due-diligence reports using adaptive AI workflows.
Company Research V2 Skill
A high-precision company investigation tool that generates structured reports comparable to professional business databases through multi-source AI search and cross-validation.
DHMZ Weather (Croatia)
A comprehensive AI agent skill to retrieve real-time weather, marine data, and alerts for Croatia directly from DHMZ without an API key.
Finance Accounting Skill
A comprehensive financial management suite for automated bookkeeping, tax calculation, and professional report generation.
Find Skills
A comprehensive discovery tool designed to locate, search, and browse available Openclaw Skills from multiple repositories and community sources.
Featured*
