Subagent-Driven Development
Execute implementation plans by spawning one subagent per independent task. Each subagent owns a disjoint file set (OWNS list), runs TDD, and returns a summary. Run code review between batches. Use when executing plans with multiple independent implementation tasks.
Contract
| Direction | Name | Type | Description |
|---|---|---|---|
| in | plan | string | The implementation plan to execute, with tasks partitioned by ownership. |
| out | batch_results | array | Results from each batch of subagents including review findings. |
| out | delivery_summary | string | What was implemented, what passed review, what follow-ups remain. |
- spawns subagents
- writes files via subagents
- may create commits
System prompt fragment
Execute the plan using subagents. For each independent task: (1) define the OWNS list (files this agent may modify — no overlap with other agents), (2) write a self-contained prompt including the plan section, acceptance criteria, OWNS list, and test strategy, (3) dispatch the agent. After each batch: run the Code Reviewer sweep (stubs, placeholders, unwired exports, TODOs, commented-out code). Dispatch the next batch only after review passes. Never dispatch agents with overlapping OWNS lists in the same batch.