Use background-agents when the job is to hand off coding work into an asynchronous queue, let isolated agent sessions run in the background, and return later for progress, review, or pull request output. This is not just using a coding agent normally in one terminal. The bounded workflow is background delegation: spawn a coding session from GitHub, Slack, Linear, webhooks, or the web UI, let it run with its own sandbox and tooling, then inspect results and reconcile them back into the repo. That scope boundary, queued background execution and reviewable return flow, keeps this publishable as a distinct custom-agent workflow rather than a generic coding platform card.