Make Claude Code Remember What Matters

TL;DR You just solved something non-trivial. The context is fresh. Instead of waiting for auto-memory to decide what to save, tell Claude: "extract what we just did as a skill." In 30 seconds you have a reusable workflow in .claude/skills/.

Auto-memory saves what Claude considers relevant. But it doesn't capture complete workflows — the steps, the decisions, the gotchas you discovered along the way. That's lost when you close the session.

The trick is simple: right after solving something hard, ask Claude to turn it into a reusable skill. You don't need a special command. Claude already knows how to create skills. You just need the habit of asking at the right moment.

Result:

You: extract what we just did to fix the memory leak
     into a reusable skill in .claude/skills/

Claude: creates ~/.claude/skills/fix-memory-leak/SKILL.md with:
  - The diagnostic steps we followed
  - The patterns we searched for (event listeners without cleanup)
  - The standard fix
  - The verification command

You: /fix-memory-leak
# Next time, the entire workflow is one command away.

How to do it

1. Spot the moment

You just solved something that:

• Took more than 10 minutes
• Involved non-obvious steps
• Could happen again in the future

2. Ask on the spot

Extract what we just did as a skill in .claude/skills/<name>/SKILL.md.
Include the steps, the patterns we searched for, and how to verify it works.

Claude has all the conversation context — the files it read, the errors it found, the final solution. This is the best moment to capture it.

3. Refine the result

Claude generates a SKILL.md with frontmatter and steps. Review it: trim what's unnecessary, add what's missing. The skill is yours — a markdown file you can edit.

4. Use it next time

/fix-memory-leak

Or let Claude invoke it automatically when it detects a similar problem (if the frontmatter description is good enough).

When to capture

Reference

Official docs: Skills — Create your first skill