This is a real guide, generated for an example user. Yours is built from a short interview about how you actually work — your saved prompts, your projects, your do-not-break workflows. Free during beta.
How to move from Claude Code to Claude Cowork.
A step-by-step guide to move Claude Code workflows into Claude Cowork — copy-ready, checkable, and checked against each tool’s own docs.
- Prepared for
- An example 3-person marketing team
- Sources
- 15 verified sources
7 steps, 4 paste-ready assets, 11 to confirm in Claude Cowork.
- CLAUDE.md repo context
- Terminal agentic loop
- Custom slash commands
- Skills and MCP integrations
Setup steps
Work through these in order. Check each off as you go; your progress is saved on this device.
Download and install Claude Cowork from claude.com/product/cowork on your macOS or Windows machine. Sign in with the same Anthropic account you use for Claude Code — Cowork honors the same subscription and shares usage limits, so there's no new billing to set up. Cowork is the native desktop chat surface; the same Claude Agent SDK loop you know from the terminal runs inside the chat window. Expect a GUI chat window rather than a `claude` REPL.
Interactive chat sessionInteractive chat replIn the Cowork composer, type `/` and run the `/setup-cowork` skill. It walks you through installing a role-matched plugin (engineering is the closest fit to your Claude Code dev workflow). Cowork plugins are curated bundles of skills, sub-agents, MCP servers (connectors), slash commands, and hooks — the same plugin concept as Claude Code's `/plugin`, just installed through a GUI. Installing the engineering plugin pre-configures the right skills and connectors so you don't wire each one up manually.
A few things to confirm
We flag these up front so nothing surprises you — a short list to double-check in Claude Cowork as you go.
- 11 items are flagged to confirm in Claude Cowork — check the menu paths and field names as you go.
- GAPS / things to verify (all guidance here is provisional and grounded in provisional FPM data — verify each in the UI): - Subagents for parallel work: the FPM results only describe Claude Code's sub-agents and background/agent-view (and its agent SDK); there is NO Cowork capability or migration pattern in the provided data for delegating parallel subtasks.
Paste-ready assets
Workflows from Claude Code, rewritten to paste straight into Claude Cowork.
CLAUDE.md kept at repo root with conventions and architecture, auto-loaded every turn.
# Project: <your-repo-name> ## Attached folders - <path/to/repo> (added via "Add folder") ## Instructions (paste of CLAUDE.md — this is the auto-load substitute) ### Architecture - <high-level module map / service boundaries> - <key directories and what lives where> ### Conventions - Language/style: <e.g. TypeScript strict, 2-space indent, no default exports> - Commit/PR conventions: <e.g. conventional commits, PR template> - Test command: <e.g. npm test> — run before declaring a task done - Build command: <e.g. npm run build> ### Constraints (pin these — do not drift) - Never edit files in <generated/ or vendor/> - Always update <docs/CHANGELOG.md> when changing public APIs - Ask before touching <migrations/ or infra/> ### Working style - Tone: direct. Skip preamble; show the diff/plan, then act. - For multi-file changes, summarize the plan before editing.
Same-day walkthrough
This sample was built for an example user. Yours gets built around how you work.
A 15–25 minute interview captures your workflows, then the guide is generated for you — same structure, your prompts, your projects, your gotchas. Nothing in your stack is changed or cancelled by any of it.