A persistent learning framework that captures errors, corrections, and best practices to enable continuous improvement for AI coding agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install hi
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 hi 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 Skill?
The Self-Improvement skill is a sophisticated logging and feedback mechanism designed to help AI agents learn from their mistakes and user interactions. By systematically capturing command failures, user corrections, and newly discovered best practices into a structured .learnings directory, this skill ensures that every session contributes to a more intelligent and capable agent. It bridges the gap between ephemeral chat history and permanent project knowledge, making it a cornerstone for those using Openclaw Skills in complex development environments.
This skill provides the infrastructure for an agent to perform self-reflection and knowledge promotion. It allows for the distillation of verbose session data into actionable rules, which can be promoted to global workspace files or project-specific documentation. This creates a virtuous cycle where the agent becomes more specialized and efficient the more it is used.
Self-Improvement Skill Use Cases
- Automatically logging unexpected command or API failures to prevent recurrence.
- Capturing specific user corrections to align the agent with project-specific coding styles.
- Documenting and prioritizing feature requests mentioned during the development process.
- Identifying recurring technical debt through pattern detection in error logs.
- Promoting verified best practices into long-term project memory files.
How Self-Improvement Skill Works
- The agent identifies a learning trigger, such as a failed shell command, a user correction, or a discovered knowledge gap.
- A structured markdown entry is created in the .learnings directory, categorized by type (Learning, Error, or Feature Request).
- Each entry is assigned a unique ID and metadata including priority, area, and status for easy tracking and filtering.
- During periodic reviews or natural session breaks, the agent identifies recurring patterns or critical fixes that need to be addressed.
- Validated and broadly applicable learnings are promoted to permanent files like CLAUDE.md or specific Openclaw Skills directories to update the agent's core behavior.
Self-Improvement Skill Setup
Install the skill using the ClawdHub registry for seamless integration with your environment:
clawdhub install self-improving-agent
Alternatively, you can perform a manual installation by cloning the repository into your local directory:
git clone https://github.com/peterskoett/self-improving-agent.git ~/.openclaw/skills/self-improving-agent
To initialize the required log files, create the directory structure within your workspace:
mkdir -p ~/.openclaw/workspace/.learnings
For users who want automated prompts, enable the session hooks via the CLI:
openclaw hooks enable self-improvement
Self-Improvement Skill Data Schema & Taxonomy
The skill maintains a rigorous taxonomy within the .learnings directory to ensure data can be processed by other Openclaw Skills. Data is distributed across three main files:
| File | Entry Type | Key Metadata |
|---|---|---|
LEARNINGS.md |
Corrections & Gaps | Category (correction, knowledge_gap, best_practice), Area, Status |
ERRORS.md |
Command Failures | Error message, Context, Reproducibility, Priority |
FEATURE_REQUESTS.md |
User Requests | Complexity estimate, Frequency, Implementation suggestions |
All entries follow the TYPE-YYYYMMDD-XXX naming convention to facilitate cross-referencing and automated promotion workflows.
Self-Improvement Skill Advanced Features
- Skill Extraction: Automatically transform validated, high-value learnings into standalone Openclaw Skills using provided helper scripts.
- Inter-Session Messaging: Use the sessions_send and sessions_history tools to share critical learnings across different active agent sessions.
- Automated Detection Hooks: Configure UserPromptSubmit and PostToolUse hooks to trigger error evaluation and learning reminders after every interaction.
- Multi-Agent Support: Standardized templates and configuration logic for Claude Code, Codex, and GitHub Copilot integration.
- Workspace Promotion: Logic for elevating behavioral patterns to SOUL.md and tool-specific gotchas to TOOLS.md for workspace-wide persistence.
SKILL.md
Loading
Related Openclaw Skills
Deep Dialogue System
A sophisticated multi-agent system designed for deep self-exploration through conversational coaching, personality profiling, and structured session synthesis.
Firm VS Bridge Pack
A powerful bridge utility for synchronizing code context and editor session states between VS Code and the OpenClaw Gateway.
Firm Suppliers Pack
A comprehensive procurement toolkit for AI agents to handle supplier sourcing, TCO analysis, and risk monitoring.
MCP Specification Compliance Audit Pack
A comprehensive suite of seven specialized audit tools designed to validate AI agent compliance with the MCP 2025-11-25 protocol specification.
WhatsApp Cloud API Messaging Skill
A technical reference and workflow guide for implementing programmatic WhatsApp messaging through the Meta Cloud API.
Deep Private Search
A secure, VPN-mandated search skill using local SearXNG for multi-iteration private research without tracking or rate limits.
Featured*
