You're building an LLM agent that answers user requests by calling tools such as search, database lookup, and ticket creation. In practice, some tool calls fail, return partial results, or time out, and the model may still try to continue as if the tool succeeded.
How do you handle tool calls that fail or time out mid-execution?