delegation-handofflisted

# Background Delegation Operations How to dispatch fire-and-forget `delegate_task` subagents so the work actually lands. Covers the parent's three jobs: **verify the premise before dispatch**, **write a brief the child can finish inside its budget**, and **inspect artifacts and finish the last mile when it can't**. This is the orchestration discipline, not the implementation discipline. For the plan-execution loop (fresh subagent per task, two-stage review), that lives in the bundled `subagent-driven-development` skill. ## When to use - The user says "do this with a background agent" / "don't clutter this thread." - Any `delegate_task` dispatch that will run more than a couple of minutes. - Any task phrased as "fix X" / "finish X" / "X is broken" where X's state is a _claim_ rather than something you have measured. --- ## 1. Verify the premise BEFORE you dispatch A brief is usually written from a one-line human summary or from a **previous session's completion claim**. Both can be wrong. A subagent dispatched on a false premise will find a way to "fix" something that isn't broken. **Rule: spend 1–2 cheap read-only calls establishing ground truth first, then put the measured facts into the brief flagged as "verified starting state — trust this."** Two real examples from one session: | Stated premise | Measured reality