extract-skills-from-claude-code-sessions

name	extract-skills-from-claude-code-sessions
description	Automatically extract reusable skills from Claude Code session transcripts using LLM analysis and wire them into a Stop hook
version	1.0.0
source	auto-extracted
extracted	"2026-04-10T00:00:00.000Z"
metadata	{"tags":["claude-code","skills","automation","self-improvement","hooks"]}

Extract Skills from Claude Code Sessions

Wire a Stop hook that fires when Claude Code ends a session. The hook runs skill-extractor.py, which reads the transcript, calls an LLM (via OpenRouter) to identify skill-worthy patterns, and writes a SKILL.md file if extraction succeeds. Use openai library for API calls, strip JSON fence artifacts with .strip(), and request concise content (<1500 tokens) to avoid truncation. Test end-to-end with synthetic skill-worthy conversations before deploying.