continuous-learning

Extracts reusable patterns from Claude Code sessions via a Stop hook.

2|Updated May 30, 2026
One-click install
npx skills add https://github.com/virahitvin8/crafty-gis --skill continuous-learning-virahitvin8
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: continuous-learning
Source: https://github.com/virahitvin8/crafty-gis/tree/main/GIT_STAR/.agent/skills/continuous-learning
Command: npx skills add https://github.com/virahitvin8/crafty-gis --skill continuous-learning-virahitvin8

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires jq.

What problem does it solve? Valuable problem-solving patterns discovered during Claude Code sessions are lost when the session ends, forcing users to repeatedly rediscover the same fixes and workarounds. ## Core Features & Use Cases - Automatic Session Evaluation: Runs as a Stop hook at session end and checks whether the transcript meets a minimum message threshold before analysis. - Configurable Pattern Detection: Detects error resolutions, user corrections, workarounds, debugging techniques, and project-specific conventions while ignoring typos and one-time fixes via config.json. - Learned Skill Storage: Saves extracted patterns as reusable skills under ~/.claude/skills/learned/ for future sessions. - Use Case: After a long debugging session where you resolved a tricky framework quirk, the Stop hook flags the session for pattern extraction so the workaround becomes a reusable learned skill. ## Quick Start Set up the continuous learning Stop hook so reusable patterns from my sessions are automatically extracted into learned skills. Note: this v1 skill is deprecated; the documentation directs new installs to continuous-learning-v2, which uses PreToolUse/PostToolUse hooks and instinct-based learning.

Frequently Asked Questions about continuous-learning

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automatically extract patterns from Claude Code sessions?

Register the evaluate-session.sh script as a Stop hook in ~/.claude/settings.json. At each session end it counts user messages in the transcript and, if the session meets the minimum length, signals that patterns should be extracted into learned skills.

Should I use continuous-learning v1 or v2?

Use continuous-learning-v2 for new installs. The v1 skill is deprecated; v2 is a strict superset that replaces Stop-hook observation with PreToolUse/PostToolUse hooks and uses atomic instincts with confidence scoring instead of full skills.

What patterns does the session evaluator detect?

It detects error_resolution, user_corrections, workarounds, debugging_techniques, and project_specific patterns. It ignores simple_typos, one_time_fixes, and external_api_issues, all configurable in config.json.

Why is the session evaluation skipped for short sessions?

Sessions below the min_session_length threshold (default 10 user messages) are skipped because short sessions rarely contain reusable patterns. You can lower the threshold in config.json if you want shorter sessions evaluated.

What happens if jq is not installed on my system?

The script falls back to built-in defaults: minimum session length of 10 and learned skills stored in ~/.claude/skills/learned/. It prints a warning to stderr but continues running normally.