// 01
Your current setup
Six projects, each with a CLAUDE.md anchor. Global MEMORY.md trimmed to 23 lines. Context stays scoped to whichever folder you launch from.
Six projects, three tools, one global config. Every CLAUDE.md is in place, MEMORY.md is trimmed, and context stays where it belongs. Click any node to dig in.
Six projects, each with a CLAUDE.md anchor. Global MEMORY.md trimmed to 23 lines. Context stays scoped to whichever folder you launch from.
Six phases, in order. Tap a phase to expand it, or press play to walk through the whole sequence. The arrows show where you can influence the outcome.
Each Claude surface has different strengths. Use the right one for the job and you will save time and tokens.
Reusable workflows that fire on trigger words. Only Claude Code supports them.
Six types of files and tools give Claude context and abilities. Here is exactly what each one does, where it lives, and which Claude surfaces support it.
| Config type | What it does | Where it lives | Claude Code | Claude Desktop | claude.ai |
|---|---|---|---|---|---|
| CLAUDE.md | Project identity, rules, stack info. Loaded automatically every session. | your project folder root | YES auto-loads | NO | NO |
| MEMORY.md | Global auto-memory. Loaded into every Claude Code session regardless of folder. | ~/.claude/projects/.../memory/ | YES auto-loads | NO | NO |
| Skills (.md) | Triggered workflow files. Load only when activated by keyword or /command. | ~/.claude/skills/ or <project>/.claude/skills/ | YES on trigger | NO | NO |
| MCP Servers | External tool servers (Pipedrive, browser, Gmail). Give Claude new abilities. | JSON config files + running processes | YES | YES | connectors only |
| Projects | Persistent system instructions + uploaded files per workstream. | claude.ai web interface | NO | NO | YES |
| settings.json | Claude Code behaviour config (permissions, model, themes). | ~/.claude/settings.json | YES | NO | NO |
Glanceable rules for when your sessions start mixing up again.