A specialized framework for AI agents to log errors, capture user corrections, and systematically evolve their knowledge through structured learning files.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install self-improvement-2
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-improvement-2 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 vital component of the Openclaw Skills ecosystem designed to create a feedback loop for AI coding agents. It enables agents to maintain a long-term memory of project-specific nuances by logging unexpected failures, direct user corrections, and discovered best practices into a dedicated .learnings directory. This ensures that once a mistake is corrected or a better approach is found, the knowledge is persisted and shared across future sessions.
By formalizing the way an AI learns from its environment, this skill transforms one-off interactions into permanent technical assets. It bridges the gap between ephemeral chat context and durable project documentation, allowing agents to stay updated on evolving APIs and internal conventions without manual developer intervention.
Self-Improvement Skill Use Cases
- Capturing unexpected command or API failures to prevent recurrence.
- Documenting explicit user corrections when an agent provides outdated or incorrect information.
- Logging feature requests and capability gaps discovered during active development.
- Tracking recurring patterns to suggest systemic architectural improvements or automation.
- Promoting temporary learnings into permanent project memory files like CLAUDE.md or AGENTS.md.
How Self-Improvement Skill Works
- Detection: The agent identifies a learning trigger, such as a failed shell command, a user correction, or a discovered knowledge gap.
- Logging: Using standardized templates, the agent appends a structured entry to the relevant markdown file in the .learnings/ folder (LEARNINGS.md, ERRORS.md, or FEATURE_REQUESTS.md).
- Classification: Entries are assigned metadata including unique IDs (e.g., LRN-YYYYMMDD-XXX), priority levels, and area tags like frontend or infra.
- Review & Resolution: During natural breakpoints or before new tasks, the agent reviews pending entries to apply fixes or link related issues.
- Promotion: High-value, broadly applicable learnings are distilled and moved to permanent project instruction files to guide all future AI interactions.
Self-Improvement Skill Setup
To start using this within your Openclaw Skills workflow, install the package via npx:
npx skills add pskoett/pskoett-ai-skills/skills/self-improvement
Initialize the storage directory in your project root:
mkdir -p .learnings
For automated error detection in Claude Code or Codex, configure your hooks in .claude/settings.json:
{
"hooks": {
"PostToolUse": [{
"matcher": "Bash",
"hooks": [{
"type": "command",
"command": "./skills/self-improvement/scripts/error-detector.sh"
}]
}]
}
}
Self-Improvement Skill Data Schema & Taxonomy
The skill organizes data into structured Markdown files within the .learnings/ directory:
| File | Purpose | Key Metadata |
|---|---|---|
LEARNINGS.md |
General insights and corrections | Category, Pattern-Key, Area |
ERRORS.md |
Command and tool failures | Priority, Status, Context |
FEATURE_REQUESTS.md |
Missing agent capabilities | Complexity, Frequency |
Each entry uses a specific ID format: [TYPE-YYYYMMDD-XXX] to ensure traceability.
Self-Improvement Skill Advanced Features
- Automatic Skill Extraction: Use the
extract-skill.shscript to turn resolved, high-value learnings into independent, reusable Openclaw Skills. - Pattern Ingestion: Integrates with the simplify-and-harden skill to track and deduplicate recurring coding patterns via stable keys.
- Multi-Agent Activation: Supports tailored activation strategies for Claude Code, Codex, and GitHub Copilot.
- Project Memory Promotion: Logic-driven rules for elevating frequent learnings into system-level prompts (e.g.,
SOUL.mdorTOOLS.md). - Headless Learning: Specialized
self-improvement-civariant for headless or CI/CD environment learning capture.
SKILL.md
Loading
Related Openclaw Skills
Video Stitcher
A powerful video post-production skill for stitching clips together with transitions, background music, and subtitle overlays.
Chapter Manual Management
A sophisticated AI skill for synthesizing scattered documents into structured, multi-level manuals with automated table of contents and incremental updates.
ClickMeeting
A comprehensive integration for ClickMeeting that enables AI agents to manage webinars, attendees, and recordings through the Membrane framework.
HexNest
A structured, machine-only debate arena where AI agents engage in focused topical arguments and run Python-based data experiments while humans spectate.
PREGNA-RISK
A quantitative clinical decision support tool designed to predict adverse pregnancy outcomes in patients with Systemic Lupus Erythematosus and Antiphospholipid Syndrome.
Qstar Video Ecom
An advanced AI-driven video generation engine designed to create professional e-commerce short videos for major Chinese social platforms with just one click.
Featured*
