How to Set Up Exec Approvals in OpenClaw: Control Which Commands Your Agent Can Run

One of the most common questions in the OpenClaw Discord is: "How do I make my agent ask for permission before running certain commands?" This guide walks you through exec approvals—OpenClaw's system for controlling shell command execution.

Understanding the Exec Approvals Model

First, a key clarification: exec approvals are not configured in openclaw.json. They're handled through a separate approvals system that lives alongside your main config.

The mental model is simple:

• Allowlist = commands that run WITHOUT prompting
• Everything else = requires approval (when configured correctly)

There's no separate "commands that need approval" list. If a command isn't on the allowlist and you have ask: "on-miss" set, it will trigger an approval prompt.

Step 1: Configure Your Approval Settings

Set up exec approvals with these three key settings:

You can configure this via:

• Control UI: Go to Nodes → Exec approvals → pick Gateway or a Node
• CLI: Use the openclaw approvals commands (see below)

Step 2: Add Commands to the Allowlist (No-Prompt Commands)

To let certain commands run without approval, add them to the allowlist:

# For gateway host approvals
openclaw approvals allowlist add --gateway --agent main "/usr/bin/uptime"
openclaw approvals allowlist add --gateway --agent main "/opt/homebrew/bin/rg"

# For node host approvals (macOS app / headless node)
openclaw approvals allowlist add --node <id|name|ip> --agent main "/opt/homebrew/bin/rg"

To inspect your current allowlist:

openclaw approvals get --gateway
openclaw approvals get --node <id|name|ip>

Important Gotcha: Use Resolved Binary Paths

Allowlist entries must match the resolved binary path, not just the command name.

❌ uptime
✅ /usr/bin/uptime

If you're unsure of the full path, use which or type:

which uptime
# /usr/bin/uptime

Step 3: Handling Approval Prompts

When the agent tries to run a non-allowlisted command, you'll get a prompt in the Control UI with two options:

1. Allow once — runs now, doesn't add to allowlist
2. Always allow — adds to allowlist + runs

This is the "approve as you go" workflow that many users prefer.

Alternative: Require Approval for EVERY Command

If you want maximum control and want to approve even allowlisted commands:

This prompts for every single exec call, regardless of allowlist status.

Sandbox Considerations

Here's a subtle but important point: exec approvals only apply when running on gateway or node hosts.

If sandboxing is disabled, host=sandbox may run directly on the gateway host without prompting. To force approvals:

• Use host=gateway or host=node explicitly in your exec calls
• Or enable sandboxing so sandbox requests actually run in isolation

You can set per-session exec defaults in chat:

Quick Reference

Documentation Links

• Exec Approvals Guide
• CLI Approvals Commands
• Exec Tool Reference

This article was synthesized from a helpful Discord discussion. Thanks to the community for the great Q&A!