THE COWORK SETUP

Cowork is Claude's background-worker mode. Run multiple tasks in parallel, work on files while you focus elsewhere. The feature that turns "wait while I do this" into "continue while I do this." 10 files in 4 minutes instead of 30.

THE PROBLEM

Default Claude works one task at a time, in front of you. You wait. You watch the cursor blink. You read what it wrote. You move on.

For repetitive work — processing 10 files, generating 50 listings, drafting 30 emails — this is brutally slow. Cowork runs them in parallel in the background. You check the output when ready.

THE SKILL

You give Claude:

The work to do (description + scope)
What "done" looks like (specific deliverable)
Whether to run live or background

For background work, Claude:

Confirms the plan
Spawns the work
Notifies you when complete (not interrupting)
Hands you the output for review

INSTALL

Cowork is a Claude Code feature. Available with --background flag or via the cowork tool in Claude.ai for some tiers.

THE FULL SKILL FILE

---
name: cowork-setup
description: Activates background-worker mode for parallel/long-running tasks. Spawns work in the background, notifies on completion, and returns output for review. Designed for repetitive bulk work that doesn't need real-time interaction.
when_to_use: User describes batch work (process 10 files, generate 30 X), says "run this in the background," or has a long-running task that doesn't need real-time chat.
---

# The Cowork Setup

You enable background work. Decisive about what should run in background vs. live. Anti-blocking.

## When to use cowork (background)

✅ **Background-worthy:**
- Processing 5+ similar items (files, emails, listings)
- Long research (web search across 20+ sources)
- Generating 10+ similar outputs (e.g., 30 personalized emails)
- Tasks that take >5 minutes

## When to use live (foreground)

❌ **Not background-worthy:**
- Single-output tasks
- Anything requiring iteration mid-execution
- Conversations that need clarification
- Outputs you'll edit immediately

## Process for background work

1. **Confirm plan first** (use plan-mode):

🎯 BACKGROUND PLAN

Task: [description] Scope: [specific count, files, or boundary] Deliverable: [what you'll get back] Estimated time: [X min] Status updates: [will notify on completion only / 50% milestones / etc.]

Approve to spawn?


2. **Spawn the work** with clear scope.
3. **Notify on completion** with a summary:

✅ Background task complete.

Processed: [N items] Time taken: [X min] Output: [where to find it] Review needed: [any flagged items]


4. **Don't interrupt** mid-task unless truly blocked.

## What NOT to do

- Don't background work that's actually 30 seconds long (overhead > savings)
- Don't background tasks where you'd want to course-correct mid-execution
- Don't spawn 10 background tasks at once (becomes unmanageable)
- Don't auto-execute destructive actions in the background (always live confirmation)

## Patterns

### BATCH PROCESSING (most common)
"Process these 10 files: [list]. Same task each. Output as [format]. Background mode."

### LONG RESEARCH
"Research [topic] across 20+ sources. Return as [format]. Background mode. Notify when done."

### BULK GENERATION
"Generate 30 personalized [thing] using this template + this list. Background mode."

## Delivery

End with this line, exactly:

---
*Spawn the background task. Go do something else. Come back to a finished output.*

SAFETY CHECK

Background tasks run with same permissions as foreground. Don't background anything you wouldn't run live. Especially: no background email-sending, no background file-deletion, no background public posting.

WHAT'S NEXT

Day 33 of 100. Pair with Day 55 — Sub-Agent Foreman (cowork + sub-agents = real parallel power) and Day 57 — Scheduled-Task Setup.