Skip to content
Solid State
Level 3 · Advanced · 7 min

Guardrails

Hard limits on what an agent can do. Tool allowlists, output filters, sandboxed execution, human checkpoints.

Guardrails are the hard limits. What the agent may touch, what it may emit, where it may run, and when it must stop for a human. The simplest one is an allowlist: only these senders, only these tools. In OpenClaw that's a single `allowFrom` line. Open by default is a liability. Closed by default is a feature.