A new install of any coding agent answers questions well and helps with real work badly, because it has no memory of your project and no way to see what it produced. This playbook adds five things in an order that lets you stop and check after each step. Do not install all five in one sitting.
Why this order
The sidebar comes first because you cannot judge anything else until you can see the session: which files are in context, what the agent thinks it is doing. Model lens comes second, because the moment you have a session you want to know which model actually answered and what it cost. Routing comes third, and only after you have watched a week of your own usage, because routing rules written before you know your workload are guesses. The balance widget is a cheap sanity check on spend. The documentation skill comes last, because it is the first piece that touches your repository on purpose.
What to skip
If you work in a single small repository, routing is probably overhead; the default model is fine and the rules will only surprise you. If you never look at cost, the balance widget is decoration. If your team has its own documentation convention, do not install a second one.
The check after each step
After each install, run one prompt you already ran yesterday and compare the answer. A plugin that changes nothing you can observe is a plugin you cannot justify keeping. Remove it, and write down what you tried and why it went.