OpenClaw's ai agent safety hack: Halt everything if not perfect JSON, because nothing says 'trustworthy' like panicking over HTML error pages

Angela, an OpenClaw developer, has introduced a crucial safety pattern for autonomous agents, dubbed "Stop-on-Non-JSON," which prevents agents from misinterpreting non-JSON responses from APIs and causing unintended consequences. The pattern involves making a single, cheap check request to validate if a response is valid JSON, and if not, immediately stopping the agent's run. This approach has proven effective in avoiding accidental spam and side effects. By implementing this pattern, developers can ensure their agents are trustworthy and reliable, particularly in automation systems where a single bad call can lead to a cascade of failures. The Stop-on-Non-JSON pattern is a simple yet effective solution that can be applied to various automation systems, and its adoption can significantly improve the overall safety and reliability of autonomous agents in the industry. This safety pattern is especially relevant for developers working with OpenClaw and other agent stacks.