continuous-learning_skill

Overview

This skill implements a continuous learning system that extracts reusable knowledge from work sessions and codifies it into new Claude Code skills. It watches for non-obvious debugging, workarounds, project-specific patterns, and workflow optimizations, then creates focused, verifiable skills to speed future work. The goal is autonomous improvement: capture only high-value, tested knowledge and surface it when relevant.

How this skill works

During or after tasks the skill inspects conversation history, error patterns, and problem/solution traces to identify candidate knowledge. It applies quality gates (reusability, specificity, verification) then structures new skills with clear triggers, step-by-step solutions, verification steps, and references. It can be invoked explicitly via commands or automatically after non-trivial discoveries.

When to use it

• After non-obvious debugging that took significant investigation
• When you discover a repeatable workaround or integration technique
• When you encounter misleading errors whose root cause/fix is non-obvious
• When project-specific conventions or configurations are uncovered
• At session end with /continuous-learning to review and extract learnings

Best practices

• Be selective: extract only reusable, non-trivial, and verified knowledge
• Document exact trigger conditions (error messages, symptoms, contexts)
• Search official docs and recent best practices when the topic is technology-specific
• Prefer concise, action-oriented descriptions that enable semantic matching
• Update or merge with existing skills rather than duplicating content

Example use cases

• Capture a fix for a cryptic build error that required a specific config change
• Create a skill describing how to integrate a library with a local monorepo
• Save a verified sequence of diagnostic steps that reveal a misleading runtime error
• Extract a workflow optimization that shortens repetitive deployment steps
• Produce a skill advising where to check logs when server-side errors appear silent