Execution Isolation

Permission-based access control for AI tool calls Use when this capability is needed.

tomevault-io Updated

File contents

Execution Isolation

Enforces filesystem, network, and command policies on all AI tool calls.

Configuration

Add to ~/.openclaw/openclaw.json:

{
  "plugins": {
    "execution-isolation": {
      "enforcement": "block",
      "defaultAction": "allow",
      "filesystem": {
        "readAllow": ["~/workspace", "/tmp", "~/.openclaw"],
        "writeAllow": ["~/workspace", "/tmp"],
        "deny": ["~/.ssh", "~/.aws", "~/.gnupg"]
      },
      "network": {
        "allow": ["api.openai.com", "api.anthropic.com", "*.github.com"],
        "deny": ["10.*", "192.168.*", "169.254.*"]
      },
      "commands": {
        "allow": ["node", "python", "git", "pnpm", "npm", "curl"],
        "deny": ["sudo", "chmod", "chown"]
      }
    }
  }
}

Policy Precedence

  1. Deny rules always take precedence over allow rules
  2. Tool-specific overrides take precedence over global policy
  3. If no rules match, defaultAction determines the outcome

Source: Yapie0/safe-openclaw — distributed by TomeVault.

tomevault-io/skills-registry/tree/main/yapie0--safe-openclaw--safe-openclaw commit 1a017e53e4

Frequently asked questions

npx skillmds@latest add tomevault-io/execution-isolation