Security Check
Exec Approval Policy: Gate Shell Access
OpenClaw agents can execute shell commands via the system.run tool. Without an approval policy, agents run commands without human review.
Why It Matters
Shell access turns prompt injection into remote code execution. An attacker who crafts the right message can make your agent run arbitrary commands — installing malware, exfiltrating data, or pivoting to other systems.
How to Fix
Configure an exec allowlist of permitted commands, or enable approval-based execution that requires human confirmation for each command.
openclaw.json
{
"tools": {
"exec": {
"allowlist": ["git status", "ls", "cat"],
"approvals": true
}
}
}