dispatching-subagentslisted

# Dispatching subagents (sequential, per-task review) Run one task to completion before the next. Review each task's diff before moving on. This is also the safe default whenever writers share a checkout: because tasks run one at a time on the same tree, they cannot race or clobber each other. Only reach for `dispatching-parallel-agents` when the writers are isolated in separate worktrees (or the fan-out is read-only). When in doubt, stay here — sequential never loses work. ## When to use - A task that depends on prior tasks (the plan marks `Independent: no` or `depends on N`). - A task that touches files another task already changed. - A single task small enough to do solo but big enough to want review. - Any task where shipping the wrong thing is worse than shipping nothing. ## When NOT to use - 3+ tasks all marked `Independent: yes` with no shared files → use `dispatching-parallel-agents`. - Tasks small enough to inline (a typo, a one-line rename). - Anything you can finish yourself in a handful of tool calls. A subagent buys a fresh context that has to rediscover what you already know. - Double-checking your own work. The reviewers here read *another* agent's diff with no memory of writing it; a subagent sent to re-read yours is self-critique with extra steps and no extra eyes. - Tasks where you don't yet understand the problem. ## State tracking Record enough per task that a fresh controller after `/clear` knows both what is done and where an unfinished task resu