Library › Book 3 › The Five Guardrails
The Five Guardrails
9.1🔒Why Guardrails Are Non-NegotiableUnconstrained agents are dangerous. They can:
9.2🔒The Five Guardrails OverviewEach guardrail catches a different failure mode. Together, they form a defense-in-depth system where each layer catches what the previous…
9.3🔒Guardrail 1: Max StepsPrevents the agent from looping infinitely. If the agent takes more than N actions without completing its objective, the test is aborted.
9.4🔒Guardrail 2: Allowed DomainsPrevents the agent from navigating to URLs outside the test environment. Without this, an agent testing a web app could follow links to…
9.5🔒Guardrail 3: TimeoutA wall-clock limit that prevents runaway test execution. Even if the agent is within its step and token budgets, a slow-responding server…
9.6🔒Guardrail 4: Token BudgetA cost ceiling that prevents the agent from making unlimited LLM calls. Each call costs tokens, and costs accumulate fast in multi-step…
9.7🔒Guardrail 5: Action AllowlistRestricts which actions the agent can perform. In production monitoring, you want read-only actions. In CI, you want everything except…
9.8🔒Guardrail Configuration by EnvironmentDifferent environments need different guardrail profiles:
9.9🔒Guardrail Violations: Logging and AlertingEvery guardrail violation should be logged with enough context to diagnose the issue:
9.10🔒Career Translation- Designed and enforced five-guardrail defense-in-depth system (max steps, domain allowlist, timeout, token budget, action allowlist) for…
9.11🔒Q&AInterview Depth CheckPrompt: Your agentic test agent is testing a web application in staging. It follows a "Forgot Password" link that redirects to Google's…