Subagent-Driven Development executes a written plan by dispatching a fresh implementer subagent for every task and a reviewer after it that checks spec compliance and code quality. A fix loop handles findings, a ledger file records progress so the run survives context compaction, and a broad whole-branch review closes the run before the finishing skill takes over.
What You Get
- Implementer and task-reviewer prompt templates, plus a re-review prompt
- A per-plan workspace and ledger under .superpowers/sdd
- Reference guides for pre-flight conflict scans, fix loops, resuming and the final review
- Guidance on choosing a model per role
- A hand-off to the finishing-a-development-branch skill
Common Use Cases
- Running a plan of mostly independent tasks with review gates
- Resuming a long run after context compaction
- Keeping a coordinator’s context small on a large plan
What Makes It Unique
Each task gets a fresh implementer and a fresh reviewer, so no context builds up between tasks. A ledger lets a controller that lost its place resume without re-dispatching finished tasks, which the skill calls the most expensive failure seen in real sessions.
Under The Hood
The SKILL.md points to prompt templates and reference files. Helper shell scripts create a git-ignored workspace for each plan, print task briefs, and package a diff range for a reviewer. The agent runs shell and git commands in your working tree, writes the ledger and briefs to disk, and dispatches subagents for implementation and review. Some helper scripts have no file extension, so the label could not read them.