TLDR Tech / AI
Don't rely on instructions, use Agent Hooks to enforce guardrails
Agent Hooks allow developers to interject themselves into an agent's workflow while they are doing work rather than after the work is done. They can stop agents from ignoring rules. This means agents can be used for deterministic checks that work 100% of the time. This article discusses how to create two agent hooks: one that ensures agents never use an 'input' tag directly, and another that ensures agents never claim work is done while tests are still failing.
Read the full story