Top 5 Claude Cowork Tips I Wish I Knew from Day One

TL;DR

Five foundational Claude Cowork setup tips covering file management, token efficiency, and project migration to avoid costly mistakes from the start.

Key Points

• 1.Use Obsidian as a free markdown reader for Cowork files. Install Obsidian, point it to your Cowork folder, and all MD files render with proper headings and formatting — making editing far easier without touching Claude's interface directly.
• 2.Keep your root claw.md under 300 lines to cut token waste. Trimming from 600 to 250 lines reduced one user's token usage by ~25%; move task-specific rules to reference files with one-line pointers instead of loading them every session.
• 3.Separate prescriptive rules from factual memory using two clear tests. If an entry uses 'always/never' language, it belongs in claw.md; if it describes a changeable fact or status, it belongs in memory.mmd — mixing them tanks output quality.
• 4.Apply a memory diet: 150-line ceiling, structured sections, and an archive.mmd. Root memory.mmd should have active projects, scheduled tasks, and core facts; when it hits 150 lines, Cowork auto-archives old entries to archive.mmd, which isn't loaded each session so it has no token cost.
• 5.Migrate Claude Projects into Cowork to eliminate editing and memory limitations. Project instructions become a workstation claw.md, project memory becomes memory.mmd, and knowledge files go into a resources folder — enabling direct edits and structured, compounding context.
• 6.Distinguish workstations from skills by asking 'place I work vs. thing I do.' Workstations handle ongoing areas with accumulated context (e.g. newsletter HQ); skills are repeatable, autopilot checklists with predictable outputs (e.g. subject-line scorer, workspace audit).