Show HN: We're using LLMs to classify risk before execution in prod Hey HN! I'm Andrios, founder of Hoop.dev, an OSS layer-7 gateway for infra access. We just released a new integration: put LLMs between devs' or agents' actions and databases or Kubernetes. The model gives a more nuanced analysis of the action, not only the syntax. Like: is this deleting data? Updating large number of entries? How risky is it? Then it decides if it will allow the execution, send it to human approval, or block it completely. Because Hoop sits at the network layer, this process happens in-transit, as data is passing through the gateway. Very low setup required. As product teams put agents in production, we're seeing security and SRE teams also shipping agents to enforce controls, and this is a nice way of deploying them. What do you think about this approach? Any feedback is super welcome here. Project is here: https://github.com/hoophq/hoop |