| name | subagent-driven-development |
| description | Use when executing implementation plans with independent tasks in the current session. |
Subagent-Driven Development
Execute implementation plans sequentially in a single session by delegating independent tasks to subagents, enforced by a two-stage review: spec compliance followed by code quality.
Pre-computation Announcement: "I'm using the subagent-driven-development skill to implement the plan."
Continuous Execution Rule: Run continuously to completion without pausing for human check-ins between tasks. Stop only if blocked, encountering ambiguous requirements, or when all tasks are complete.
When to Use
- Detailed, checkbox-tracked implementation plan exists.
- Tasks are largely independent.
- Rapid iteration without manual file-reading overhead or context pollution is desired.
Execution Workflow
- Initialize: Read plan, extract task descriptions and target files, initialize checkbox tracking.
- Execute Task:
- Dispatch Implementer: Start fresh subagent with full task text and necessary context/files.
- Answer Questions: Respond immediately to implementer questions and re-dispatch.
- Stage 1: Spec Compliance Review: Dispatch spec reviewer subagent to verify code satisfies spec without extra scope. Implementer fixes flagged issues.
- Stage 2: Code Quality Review: Dispatch code quality reviewer to verify tests, readability, and design patterns. Implementer fixes flagged issues.
- Checkpoint: Mark complete, commit changes, proceed immediately to next task.
- Wrap-up: Perform final codebase review.
Implementer Statuses
- DONE: Proceed directly to Spec Compliance review.
- DONE_WITH_CONCERNS: Evaluate concerns; resolve scope/correctness issues before review, otherwise proceed to review.
- NEEDS_CONTEXT: Provide missing details and re-dispatch.
- BLOCKED: Do not force retries. Provide context, upgrade model capacity, or escalate plan defects to human.
Prompt Templates
./implementer-prompt.md - Task implementation../spec-reviewer-prompt.md - Specification alignment verification../code-quality-reviewer-prompt.md - Code quality and test verification.
Benefits & Critical Rules
Key Benefits
- Zero subagent file-reading overhead (controller supplies all text and context).
- Strong TDD and self-review discipline.
- Two-stage review gates prevent regression and scope creep.
Prohibitions
- Never skip spec compliance or code quality reviews.
- Never proceed to the next task while current task reviews have open issues.
- Never make subagents read raw plan files (pass task text directly).
- Never substitute implementer self-review for independent review stages.
- Never begin code quality review before spec compliance is 100% approved.
