ACAgentic Craft

FoundationHazelJS

Guardrail

Preventive or detective control enforcing safety, policy, or quality constraints on agent behavior.

Authors
editorial-team
Published
Last reviewed

Direct answer

Preventive or detective control enforcing safety, policy, or quality constraints on agent behavior.

Plain language

Rails that keep the agent on the path—filters, allow-lists, validators, and blockers.

Technical explanation

Includes schema validation, ToolRegistry allow/deny, PolicyEngine, PII redaction, content filters, RunBudget enforcers, and approval requirements—preferably in runtime code via @hazeljs/guardrails + agent policy.

Example

Block outbound tool URLs not on an allow-list using guardrails before ToolExecutor egress.

Related

← All glossary terms