What kind of artifacts can it create?

It can generate Rules (declarative heuristics), Skills (executable command sequences), Hooks (automatic triggers), or Agent updates (enhanced personality/capabilities).

What is meta-pattern detection?

It is a process where the skill looks at clusters of learnings; if many rules relate to one topic, it suggests creating a dedicated Skill to handle that domain more efficiently.

How does Compound Learnings identify what to save?

It scans session logs for specific headers like 'Actionable Takeaways' and 'What Worked', then uses a frequency table to suggest artifacts only when a pattern appears across multiple sessions.

Where does the skill store its data?

It operates within the .maestro/ directory of your project, looking at cache/learnings and writing to the rules, skills, and hooks subdirectories.

Can I review changes before they are applied?

Yes, the skill is designed to present proposals including the rationale and draft content, requiring user approval via the AskUserQuestion tool before creation.

Compound Learnings

Name: Compound Learnings
Author: scooter-lacroix

byscooter-lacroix

0•

Productivity & Workflow

Transforms ephemeral session insights into permanent rules, skills, and automation hooks to continuously improve your development environment.

Compound Learnings is a meta-capability for Claude Code that builds a persistent organizational memory for your development projects. By analyzing recent session logs, it identifies recurring patterns, success heuristics, and common pitfalls to suggest permanent improvements. Instead of losing insights when a session ends, this skill categorizes them into declarative rules, executable skills, or automated hooks. It utilizes a signal-based threshold system to ensure only high-value, recurring patterns are promoted, effectively allowing your AI development environment to grow smarter and more specialized with every task you complete.

Key Features

010 GitHub stars

02Meta-pattern detection to consolidate multiple rules into unified skills

03Signal-based thresholding to identify high-confidence improvements

04Categorization of insights into Rules, Skills, Hooks, or Agent updates

05Automated extraction of patterns from recent session learnings

06Structured proposal system for manual or batch approval of new artifacts

Use Cases

01Improving project setup based on recurring debugging patterns

02Codifying project-specific standards after a major feature implementation

03Converting a manual multi-step workflow into a permanent executable skill

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add scooter-lacroix/maestro compound-learnings

For use in Claude.ai and ChatGPT

Download Skill