After 8 months of daily Claude Code use, I have distilled my workflow into a 5-layer system.
Layer 1: CLAUDE.md — Project Memory
This is the foundation. CLAUDE.md is a file at your project root that Claude reads automatically every session.
Bad CLAUDE.md: Generic advice like "write clean code, use Git."
Good CLAUDE.md: Specific tech stack, exact commands, architecture diagram, unique conventions.
Rule: only write what is unique to your project.
Layer 2: Plan Mode — Direction Before Speed
Trigger: any change touching 3+ files. Claude explores code, designs a plan, waits for your approval before writing anything.
Layer 3: Small Tasks — One Unit, One Commit
Each task changes ONE logical unit. After each task, the project is still runnable. You know exactly which commit introduced a bug.
Layer 4: Git — Your Undo Button
Commit after every small task. The commit history IS your project journal.
Layer 5: Worktree — Parallel Without Collision
git worktree add ../bugfix .claude/worktrees/hotfix
cd ../bugfix
# Fix bug in isolation
cd -
git worktree prune
Bonus: Context Compounding
| Level | Vehicle |
|---|---|
| Project | CLAUDE.md |
| Session | Memory system |
| Knowledge Base | Linked notes |
After 8 months, Claude does not just understand my project — it understands how I think.
Quick Start
Day 1: Write a proper CLAUDE.md
Day 2: Use Plan Mode before cross-file changes
Day 3: Break tasks into small steps
Day 5: Try Worktree
Day 7: Write your first Memory
Day 14: Review commits for patterns
☕ Support This Content
If my articles saved you time, scan the QR code below to buy me a coffee.
Follow @mgj for weekly AI engineering content.
Top comments (0)