3-Step Token Management: CLAUDE.md + Context Folder + Fresh Chat Per Task
See more in Claude Code / Skills →Exact instruction
- Create CLAUDE.md with your reusable rules and context (Project instructions in Claude.ai, or global CLAUDE.md in Claude Code).
- Create a context/ folder with reference docs — reference them by name only when needed rather than loading everything upfront.
- Start a fresh chat for each distinct task — don't continue yesterday's long chat for today's quick task.
Adopt?
Yes: The context folder pattern (step 2) is underused — keeping heavy reference docs separate and pulling them on-demand is more efficient than loading everything into every session.
Find the resource
Comment 'SNAIL' on the original post (@wright_mode) for the full token-saving guide.
show original captionhide original caption
Comment SNAIL for my full guide on saving Claude tokens Your token problem isn't fixed by a bigger plan, it's fixed by smaller context. Think of every chat like a backpack you keep overstuffing. Pack your CLAUDE.md, build a context folder, and start a fresh chat per job so you're not dragging the whole day into a 5-minute task.