Can I manually trigger a learning session?

Yes, use the /learn command at any time to review your session history and prioritize high-value knowledge for extraction.

Does it use external data for skill creation?

The system can perform web searches during the extraction process to verify current best practices and include official references in the skill documentation.

Where are the new skills stored?

Skills are saved as .mdc files in .claude/rules/ or as SKILL.md files in .claude/skills/ for persistent access across future sessions.

What are the quality criteria for a new skill?

A skill must be reusable, non-trivial, specific, and verified to have worked in the current context before it is codified.

How does the Learn skill identify new knowledge?

It triggers via the /learn command or specific phrases like 'save this as a skill', analyzing recent work for non-obvious solutions, patterns, or tool-specific insights.

Autonomous Skill Learning

Name: Autonomous Skill Learning
Author: udecode

byudecode

•

1,127

•

Learning & Documentation

Extracts, codifies, and refines reusable knowledge from development sessions into persistent Claude Code skills.

The Learn skill enables Claude to autonomously improve over time by identifying non-obvious solutions, project-specific patterns, and tool integrations discovered during work. By mapping skill extraction to Test-Driven Development (TDD) principles, it ensures that every debugging win, workaround, or architectural decision is documented as a reusable resource. This creates a self-optimizing development environment where past discoveries are never lost and future tasks benefit from previously acquired context.

Key Features

011,127 GitHub stars

02Session retrospective mode via the /learn command

03Automatic MDC rule generation and frontmatter formatting

04TDD-based skill creation (Red/Green/Refactor)

05Web research integration to verify best practices

06Automated knowledge extraction from chat history

Use Cases

01Codifying non-obvious debugging fixes for misleading error messages

02Capturing project-specific architectural patterns and conventions

03Documenting tool-specific optimizations and undocumented workarounds

Key Features

011,127 GitHub stars

02Session retrospective mode via the /learn command

03Automatic MDC rule generation and frontmatter formatting

04TDD-based skill creation (Red/Green/Refactor)

05Web research integration to verify best practices

06Automated knowledge extraction from chat history

Use Cases

01Codifying non-obvious debugging fixes for misleading error messages

02Capturing project-specific architectural patterns and conventions

03Documenting tool-specific optimizations and undocumented workarounds