Network Allowlist for OpenClaw Skills: Safe Defaults and Practical Setup
If a skill has network access, it can send data anywhere.
The safe pattern is simple:
Default deny, then allow only the domains you can justify.
Step 1: Default to no network
Treat “network” as an exception, not a default.
Start here: /guides/openclaw-security
Step 2: Require a per-skill allowlist
For every networked skill, write down:
- which domains it must call
- why those calls are necessary
- what data might leave your machine
Step 3: Separate “fetch docs” from “send data”
If a skill only needs to fetch public docs:
- allow only documentation domains
- block uploads where possible
- avoid sending repo contents to third-party services
Step 4: Treat fileRead + network as high risk
This combo can exfiltrate secrets.
Mitigations:
- keep secrets out of the workspace
- scan before running: /guides/credential-protection
- use a sandbox: /guides/sandbox-setup
Step 5: Re-verify after updates
Permissions drift happens. Re-run: