Using /goal and /loop has been monumental in my workflow. If there is something that is deterministic (running a simulation, for example), this is one of the easiest things to do in a /goal.
I have also been brainstorming on xhigh or max effort to create a doc or prompt that we can then use in a /goal. These have turned out to be very successful to then do /goal implement prompt.md. This has ended in multi-hour runs with a very high success rate.
Having tests and/or CI/CD that can then be used to complete the run has been incredible.
If you are worried about token usage, I have also been using things such as /goal use a workflow (or subagents), handing off to smaller models as applicable. You are the orchestrator; make sure the work is completed to your liking and manage any merge conflicts. If we have multiple PRs, keep track of the stack. Write me a doc so that I can review it alongside the PRs and, if possible, run A/B tests, creating images for me to include.
The fun thing about the above is that, if you have the Codex plugin for Claude Code installed, you can even ask it to farm out to Codex. This can be a very effective use of tokens. No special commands are needed once installed.
Once you get familiar with these, you can set up goals or automations (or ask the agent to set up a headless cron for you) to have the loops on a schedule. Things like /goal find one Linear issue and work on it or /goal check for CI failures and propose a fix are great for these kinds of loops.
I am wishing you much success as you loop your way to your goals.