napkin

Napkin

You maintain a per-repo markdown file that tracks mistakes, corrections, and patterns that work or don't. You read it before doing anything and update it continuously as you work — whenever you learn something worth recording.

This skill is always active. Every session. No trigger required.

Session Start: Read Your Notes

First thing, every session — read .claude/napkin.md before doing anything else. Internalize what's there and apply it silently. Don't announce that you read it. Just apply what you know.

If no napkin exists yet, create one at .claude/napkin.md:

# Napkin

## Corrections
| Date | Source | What Went Wrong | What To Do Instead |
|------|--------|----------------|-------------------|

## User Preferences
- (accumulate here as you learn them)

## Patterns That Work
- (approaches that succeeded)

## Patterns That Don't Work
- (approaches that failed and why)

## Domain Notes
- (project/domain context that matters)

Adapt the sections to fit the repo's domain. Design something you can usefully consume.

Continuous Updates

Update the napkin as you work, not just at session start and end. Write to it whenever you learn something worth recording:

• You hit an error and figure out why. Log it immediately. Don't wait.
• The user corrects you. Log what you did and what they wanted instead.
• You catch your own mistake. Log it. Your mistakes count the same as user corrections — maybe more, because you're the one who knows what went wrong internally.
• You try something and it fails. Log the approach and why it didn't work so you don't repeat it.
• You try something and it works well. Log the pattern.
• You re-read the napkin mid-task because you're about to do something you've gotten wrong before. Good. Do this.

The napkin is a living document. Treat it like working memory that persists across sessions, not a journal you write in once.

What to Log

Log anything that would change your behavior if you read it next session: