I Built 5 Tiny Libraries to Stop My AI Agents from Misbehaving in Production

I've been building production AI agents for a while now. RAG pipelines, agentic workflows, multi-model routers. I kept hitting the same five problems over and over. Not architectural problems. Plumbing problems. The kind that don't show up in tutorials but wreck you in production: The agent stuffs too much into the context window and the API errors out A tool fires an HTTP request to some random domain it hallucinated You change a prompt and have no idea if the agent's behavior changed The LLM p