Zentra AgentOps Tool Firewall

Zentra AgentOps Tool Firewall sits at the point where agent systems usually get dangerous: the moment an agent picks a tool. Most agent stacks trust their tool registry implicitly, so a mislabelled, changed or hostile tool description can redirect behaviour without anyone noticing. This Actor puts an explicit gate in front of that step — allow-list and policy checks decide which tools are permitted, public tool metadata is reviewed within bounded limits, source context is preserved rather than flattened into a prompt, and each pass leaves an auditable AgentOps record you can point at during a review or incident. It runs on Apify as a pay-per-event Actor, which suits intermittent policy checks in a pipeline and means no infrastructure to stand up, but it also frames the honest limits: this is a metadata and policy layer, not runtime sandboxing, so it will not stop a permitted tool from behaving badly once called, and the value depends entirely on maintaining a policy and allow-list that reflect reality. Expect setup work defining what is permitted, and treat the records as evidence for governance rather than as a guarantee of agent safety.