Summary

This project addresses the problem of AI coding agents losing learned knowledge between sessions. It provides a "meta-skill" that automatically captures reusable "golden paths"—successful workflows, commands, or facts discovered during problem-solving—and persists them for future use. This enhances agent efficiency by preventing repetitive rediscovery of solutions, benefiting users of AI coding assistants like Claude Code and Cursor.

How It Works

The system operates on a simple loop: recognize a moment of earned knowledge (e.g., a complex command sequence, a crucial project fact), automatically capture it as a reusable skill or rule (including failures), and then reuse it in subsequent sessions. This approach leverages the agent's interaction history to build a persistent knowledge base, improving performance over time without explicit user intervention for saving. The capture process distinguishes between multi-step procedures (skills), single facts (memory), and one-off events, ensuring the knowledge base remains relevant.

Quick Start & Requirements

• Primary Install: npx skills add kulaxyz/self-learning-skills (recommended).
• Prerequisites: Node.js/npm for npx. Compatible with 70+ agents including Claude Code, Cursor, Codex, Zed, Aider, and Gemini CLI.
• Links: GitHub repository: https://github.com/kulaxyz/self-learning-skills.

Highlighted Details

• Broad Agent Compatibility: Adapts to Claude Code, Cursor, and agents using AGENTS.md, persisting knowledge in tool-specific formats (skills/<name>/SKILL.md, .cursor/rules/learned/<name>.mdc, AGENTS.md).
• Rigorous Promotion Rule: Skills are only promoted if they include a verified passing check, a named failure pattern, and at least one ruled-out dead-end, ensuring reliability.
• Safety First: Designed to never write secret values directly; instead, it records where secrets are located (e.g., environment variable names).
• Intelligent Triage: Automatically routes lessons to skills/rules, lightweight notes/memory, or skips genuine one-offs, preventing configuration bloat.

Maintenance & Community

The project utilizes a community skills CLI and incorporates feedback, suggesting community involvement. Specific community channels or active contributor details are not detailed in the README.

Licensing & Compatibility

• License: MIT License.
• Compatibility: The MIT license is permissive, allowing for commercial use and integration into closed-source projects.

Limitations & Caveats

Effectiveness is contingent on the agent's ability to correctly identify and capture learning moments. The README does not provide specific performance benchmarks or detailed resource requirements.