what's your strategy for dead-letter queues

i've been setting up some new services that rely heavily on message queues, and i'm thinking more about how to handle messages that repeatedly fail processing. my current thought is to use a dead-letter queue, or DLQ, to capture these, but i'm curious about what others do. do you have a specific setup for your DLQs? how do you monitor them, and what's your process for investigating and replaying messages from the DLQ? any particular tools or patterns you find effective