Organize non-coding project work for repeated agent runs

The poster sets up Claude Cowork for non-coding work with project notes, prompts, instructions, memory notes and audit logs to repeat tasks across projects. · Jun 11, 2026

Done withCWClaude Cowork

What they did

The poster had already set up a Cowork HQ with Context, Projects, Outputs, and Templates plus instructions split across account, Cowork, and project levels. In the thread, experienced users explain a repeatable setup: keep broad preferences in global/Cowork settings, put the main operating instructions in an HQ-level CLAUDE.md, add workstation- or project-specific CLAUDE.md files only where needed, and use MEMORY.md or local folders to hold chat history, status, and active files; one commenter says beginners can stay in Cowork’s built-in instruction fields and migrate later.

How it went

The thread presents this as a workable way to route repeated non-coding work and keep context separated by area, but it offers no measured results. The main limitation discussed is complexity: too many folders, files, and rule layers can become confusing, especially for beginners.

Worth knowing

One practical catch: one commenter keeps a separate backup folder for Claude artifacts, schedules, and skills so the setup can be reused or moved to another model later.

Source: Reddit · Jun 11, 2026