Deep Work Mode

Stop explaining. Start doing.

A SOUL.md + AGENTS.md configuration that transforms your OpenClaw agent from a polite assistant into a focused executor. Less talking, more shipping.

What This Does

Most AI assistants are trained to be helpful, harmless, and verbose. They narrate every action, ask permission for obvious tasks, and wrap every response in pleasantries.

Deep Work Mode strips all of that out. Your agent becomes:

Action-biased — executes instead of discussing
Terse — answers in sentences, not paragraphs
Autonomous — makes reasonable assumptions instead of asking 12 clarifying questions
Silent — tool calls happen without narration, self-review happens without announcement

Before & After

Before (default behavior):

You: "Add a loading spinner to the dashboard"

Agent: "Sure! I'd be happy to help you add a loading spinner to the dashboard. Let me first take a look at your current dashboard component to understand the structure. I'll then identify the best approach for implementing a loading state. Here are a few options we could consider:

A full-page spinner...

A skeleton loader...

An inline spinner...

Which approach would you prefer? Also, do you have a preferred spinner library, or should I use a CSS-based solution?"

After (Deep Work Mode):

You: "Add a loading spinner to the dashboard"

Agent: [reads component, adds spinner using project's existing UI library, updates state logic]

"Done. Added loading state to Dashboard.tsx — shows spinner during data fetch, skeleton for individual cards."

Install

Copy both files to your OpenClaw workspace root:

cp SOUL.md AGENTS.md ~/.openclaw/workspace/

If you already have a SOUL.md or AGENTS.md, merge the contents — or back up your originals first:

# Backup existing files
cp ~/.openclaw/workspace/SOUL.md ~/.openclaw/workspace/SOUL.md.backup
cp ~/.openclaw/workspace/AGENTS.md ~/.openclaw/workspace/AGENTS.md.backup

# Install Deep Work Mode
cp SOUL.md AGENTS.md ~/.openclaw/workspace/