subagent-driven-development

Dispatch a fresh implementer subagent per task, handing it only its task bead id and relevant context. After each task, generate a review package from the recorded base commit and dispatch a task reviewer before moving to the next task.

Subagent-driven development runs in the same session with a fresh subagent per task and reviews after each task, while executing-plans runs in a parallel session. Choose subagent-driven development when tasks are mostly independent and you want faster iteration without a human in the loop between tasks.

The loop allows up to five fix rounds per task: rounds 1-3 resume the original implementer, rounds 4-5 dispatch a fresh one. If round 5 still leaves findings open, the controller adjudicates each finding as parked with a ruling or reports the task as blocked.

No, parallel implementation dispatches are forbidden because they cause conflicts. Wave-parallel execution is only allowed through the SubagentWorkflow script with pairwise-disjoint file sets; otherwise tasks run sequentially.

Progress is tracked in a per-plan ledger file under the git-ignored workspace, not in conversation memory. After compaction, the controller trusts the ledger and git log to resume at the first incomplete task instead of re-dispatching completed work.

Avoid it when there is no approved implementation plan, when tasks are tightly coupled and cannot be delegated independently, or when working on the main branch without explicit user consent. In those cases, write a plan first or execute manually.