A self-improving memory system that enables AI agents to learn from every session and compound knowledge over time.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install compound-engineering
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 compound-engineering 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 Compound Engineering?
Compound Engineering provides a framework for AI agents to improve their own performance through systematic self-reflection. By integrating these Openclaw Skills into your workflow, you enable an agent to analyze past sessions, identify successful patterns, and avoid repeating mistakes. This creates a continuous improvement cycle where tomorrow's agent is always more capable than today's, effectively building a custom knowledge base tailored to your specific coding environment.
The skill focuses on extracting high-value data points like user preferences, architectural decisions, and technical gotchas from raw chat logs. This information is then synthesized into structured Markdown files, ensuring that the agent's long-term memory remains organized and actionable without requiring manual intervention from the developer.
Compound Engineering Use Cases
- Automating nightly reviews of AI chat sessions to extract long-term technical insights.
- Building a persistent memory of user preferences and project-specific coding standards.
- Maintaining a granular audit trail of decisions and architectural choices in daily log files.
- Updating agent instructions dynamically based on what worked or failed in previous tasks.
- Reducing repetitive instruction overhead by formalizing learned patterns into AGENTS.md.
How Compound Engineering Works
- The agent performs daily work across various sessions, chats, and coding tasks, generating logs and context.
- A scheduled nightly review (manually or via cron) scans all activity from the last 24 hours.
- The system extracts specific data points including patterns, mistakes, user preferences, and pending items.
- Memory files such as MEMORY.md (long-term) and daily markdown logs are automatically updated with the new insights.
- Changes are committed to Git, providing the agent with fresh, updated instructions for the next day's work.
Compound Engineering Setup
Install the skill and trigger manual or automated reviews using the following commands:
# Review last 24 hours and update memory
npx compound-engineering review
# Create hourly memory snapshot
npx compound-engineering snapshot
# Set up automated nightly review (cron)
npx compound-engineering setup-cron
For users integrating with Clawdbot, you can add a nightly review job to your configuration:
{
"id": "compound-nightly",
"schedule": "30 22 * * *",
"text": "Review sessions, extract learnings, and update MEMORY.md and AGENTS.md."
}
Compound Engineering Data Schema & Taxonomy
The skill organizes knowledge using a hierarchical Markdown structure:
| File | Level | Purpose |
|---|---|---|
MEMORY.md |
Long-term | Stores persistent patterns, user preferences, and core project context. |
memory/YYYY-MM-DD.md |
Daily | Records granular session logs, specific decisions, and immediate learnings. |
AGENTS.md |
Instruction | Updates agent behavioral rules and workflow configurations based on feedback. |
HEARTBEAT.md |
Automation | Optional triggers for hourly snapshots to ensure no context is lost during long sessions. |
Compound Engineering Advanced Features
- Automated Hourly Snapshots: Create granular state tracking to capture progress within a single workday.
- Clawdbot Integration: Native support for scheduled heartbeat tasks and automated Git commits.
- Pattern Extraction Engine: Specific logic to distinguish between noise and significant architectural learnings.
- Self-Correcting Instructions: Automatically updates agent persona and rules based on user likes and dislikes.
- Version-Controlled Wisdom: All memory updates are committed to Git, allowing for easy rollback or audit of the agent's learning path.
SKILL.md
Loading
Related Openclaw Skills
Compound Engineering
An automated learning framework that enables AI agents to review their work, extract patterns, and compound knowledge daily.
Component Gen
Turn natural language descriptions into fully functional React components with TypeScript and Tailwind support.
Comment Generator
An AI-driven CLI tool that injects meaningful inline comments into source code to explain the intent and reasoning behind complex logic.
Codemod Generator
An AI-driven CLI tool that generates jscodeshift scripts to automate complex, large-scale code refactoring via natural language commands.
Core Vitals Fixer
Automatically scan and fix Largest Contentful Paint, Cumulative Layout Shift, and First Input Delay issues using AI guidance.
Core Web Vitals Fixer
An AI-powered CLI tool designed to scan codebases and provide instant, actionable fixes for Core Web Vitals performance bottlenecks.
Featured*
