Memory in Practice

Level: 🌱 Beginner Related: Memory

← Back to Auto Memory · ← Back to Setting Your Environment

Sarah has been using Claude Code for a couple of weeks on a Node.js project. She has a CLAUDE.md, but keeps finding herself correcting Claude on the same things — the package manager, naming conventions, test patterns. She doesn’t realize Claude has been learning from those corrections all along.

Step 1: Claude Already Knows

Monday morning. Sarah opens a new session and asks Claude to add a utility function with tests.

Claude creates the function and runs the tests:

pnpm vitest run src/utils/format-date.test.ts

Sarah pauses. She didn’t tell Claude to use pnpm — her CLAUDE.md just says “run tests before committing.” But last week, she’d corrected Claude: “Use pnpm, not npm.” Claude remembered.

She also notices Claude used camelCase for the function name — another correction she made days ago. These small things used to require re-explaining every session. Now they just work.

Step 2: Making a Correction That Sticks

Later that day, Claude generates an API handler and names the file formatDate.handler.ts.

Sarah corrects it: “We use kebab-case for filenames — it should be format-date.handler.ts.”

Claude fixes the filename and saves the convention to memory. From now on, every session on this project will use kebab-case filenames without being told.

Step 3: Explicitly Asking Claude to Remember

While debugging a deployment issue, Sarah discovers something that isn’t in any docs: the staging environment requires a specific header (X-Staging-Auth) for API calls to work.

She tells Claude directly:

Remember that staging API calls need the X-Staging-Auth header.
The value comes from the STAGING_TOKEN env var.

Claude saves this to memory. Next time she (or Claude) works on anything involving staging, the information is there — no digging through Slack threads or deployment docs.

Step 4: Checking What Claude Knows

A few days later, Sarah is curious what Claude has accumulated. She runs:

/memory

Her MEMORY.md looks like this:

# Project Memory

## Build & Test
- Uses pnpm, not npm
- Test runner: vitest (not jest)
- Run pnpm vitest run for tests

## Conventions
- camelCase for variables and functions
- kebab-case for filenames
- API routes under /api/v2/

## Environment
- Staging API needs X-Staging-Auth header (from STAGING_TOKEN env var)
- Local dev requires Redis: docker compose up redis

## Preferences
- Prefers concise commit messages
- Wants test files co-located with source (not in separate __tests__ dir)

Some of these she told Claude explicitly. Others Claude picked up from corrections during normal work. All of them save her time every session.

Step 5: Promoting Memory to CLAUDE.md

Sarah notices that the kebab-case filename convention and the pnpm requirement aren’t just her preferences — they’re team standards. But they’re only in her personal auto memory, so other developers using Claude Code won’t benefit.

She opens CLAUDE.md and adds:

## Conventions
- Use kebab-case for filenames (e.g., format-date.handler.ts)
- Use pnpm for all package operations (not npm or yarn)
- Test files live alongside source files, not in __tests__/

Then she runs /memory and removes the now-redundant entries from her auto memory. The team-wide conventions live in CLAUDE.md (shared via git), while her personal notes (like the staging header quirk) stay in auto memory.

Key Takeaways

Memory works automatically. Claude saves corrections and patterns without you asking. You don’t need to set anything up.
Memory is personal. Your auto memory is yours alone — it doesn’t affect teammates or other projects.
You can be explicit. Tell Claude to remember (or forget) anything. It’s not limited to things Claude discovers on its own.
Review periodically. Run /memory every few weeks to prune outdated entries and spot things worth promoting.
Promote team knowledge. When a memory entry applies to everyone, move it to CLAUDE.md so it’s shared and versioned.
Memory complements CLAUDE.md. CLAUDE.md is the team playbook. Auto memory is your personal notebook. Together they give Claude full context every session.