LLM response caching

Agent-cache's tool cache returns the first ticket's ID when the tool writes instead of reads

The library puts LLM completions, tool results and session state behind one Valkey or Redis connection. Its tool keys are a function name plus an argument hash, and the post says that breaks for tools that mutate state.