continuous-learning_skill

Overview

This skill captures and preserves actionable patterns discovered during work sessions so they can be reused in future sessions. It extracts error resolutions, user corrections, workarounds, debugging techniques, and project-specific conventions and saves them to a persistent learned-skills directory. The goal is to turn transient discoveries into structured, searchable artifacts that improve productivity over time.

How this skill works

During or at the end of a session the skill scans transcripts, errors, user corrections, and debugging steps for signals of learning. It applies configurable thresholds to decide which moments qualify, structures each finding into a type-specific template, deduplicates against existing patterns, and writes the result to ~/.claude/skills/learned/. Patterns are indexed for fast retrieval and get confidence and usage metadata for progressive refinement.

When to use it

• Automatically when a session ends or on a configured stop hook
• When you say commands like "save learnings" or "extract patterns"
• After resolving a tricky error or exception that required investigation
• When a user corrects an approach revealing a better method
• After discovering a reliable workaround or an effective debugging technique

Best practices

• Keep extracted patterns specific and include context (framework, version, env)
• Capture the root cause, not just the quick fix, to enable correct reuse
• Prefer actionable steps and code examples so patterns are immediately usable
• Configure extraction sensitivity to balance noise vs. missed learnings
• Review and merge similar patterns to avoid duplication and improve confidence

Example use cases

• Store an error_resolution for a recurring TypeError with root cause and fixed code
• Save a user_correction detailing preferred test-naming conventions for a project
• Record a workaround for a framework limitation with caveats and alternatives
• Capture a debugging_technique for diagnosing async race conditions with step-by-step procedure
• Document a project_specific pattern for API naming and examples from the codebase