5 Context Layers That Make Claude Cowork Actually Work

Most people treat Claude like a chat box retyping the same context every session. Cowork loads 5 layers before you type a single word. Here's what they are and how to build them. What we cover: → Layer 1: Claude.md - your global identity file, loaded automatically every session → Layer 2: Projects - scoped workspaces per client or topic, context switches with you → Layer 3: Files - your local folder is part of the workspace, no copy-pasting required → Layer 4: Skills - describe a workflow once, invoke it forever (live demo: YouTube thumbnail skill) → Layer 5: Connectors - live connections to your calendar, CRM, and docs, data flows in directly → The Harness Mindset: 3 questions to ask after every session so the next one starts smarter This is video 4 of the Claude Cowork series. Timestamps: 00:00 — Introduction: 5 layers of persistent context 00:16 — Layer 1: Claude.md — your identity file, always loaded 01:33 — Global instructions in Cowork settings 02:46 — Layer 2: Projects — scoped workspaces per client or topic 03:38 — How projects store instructions, files, and conversations 04:11 — Demo: YouTube thumbnail skill invoked inside a project 05:56 — Layer 3: Files — local folder always accessible 06:45 — Layer 4: Skills — describe once, invoke forever 07:20 — How the thumbnail skill was created from a conversation 08:50 — When and how to create a skill 09:43 — Layer 5: Connectors — live connections to your tools 10:36 — Demo: checking calendar events via Google Calendar connector 11:42 — Connecting Airtable and querying your data 12:24 — What Cowork already knows before you type anything 12:40 — The Harness Mindset: every session is a deliverable 13:22 — 3 questions to ask after every session 14:21 — What improvement looks like: week 1 vs one month in