subagent-driven-development
Scannednpx machina-cli add skill a5c-ai/babysitter/subagent-driven-development --openclawSubagent-Driven Development
Overview
Execute plan by dispatching fresh subagent per task with two-stage review: spec compliance first, then code quality.
Core principle: Fresh subagent per task + two-stage review = high quality, fast iteration.
When to Use
- Have implementation plan with mostly independent tasks
- Want to stay in current session
- Want automatic review checkpoints
Two-Stage Review
- Spec Compliance - Did they build what was requested? (nothing more, nothing less)
- Code Quality - Is it well-built? (clean, tested, maintainable)
Spec MUST pass before quality review begins.
Red Flags
- Never skip either review stage
- Never proceed with unfixed issues
- Never dispatch multiple implementation subagents in parallel
- Never let implementer self-review replace actual review
Agents Used
agents/implementer/- Fresh subagent per taskagents/spec-reviewer/- Verifies spec complianceagents/code-quality-reviewer/- Verifies code qualityagents/code-reviewer/- Final review of entire implementation
Tool Use
Invoke via babysitter process: methodologies/superpowers/subagent-driven-development
Source
git clone https://github.com/a5c-ai/babysitter/blob/main/plugins/babysitter/skills/babysit/process/methodologies/superpowers/skills/subagent-driven-development/SKILL.mdView on GitHub Overview
Subagent-Driven Development executes an implementation plan by dispatching a fresh subagent per task within the current session. Each task undergoes a two-stage review—spec compliance first, then code quality. This approach yields higher quality, faster iteration, and clear checkpoints.
How This Skill Works
For each independent task, a new subagent is spawned from the implementer flow. Tasks first pass through the spec-reviewer to verify exact requirements; only if spec passes does the code-quality-reviewer evaluate cleanliness, tests, and maintainability. A final code-reviewer provides an end-to-end assessment of the completed implementation.
When to Use It
- Have an implementation plan with mostly independent tasks.
- Want to stay in the current session while tasks execute.
- Need automatic review checkpoints before progressing.
- Prefer fresh subagents per task to ensure isolation.
- Require a final, integrated review before finishing.
Quick Start
- Step 1: Invoke the skill via babysitter process: methodologies/superpowers/subagent-driven-development
- Step 2: Break the plan into independent tasks and dispatch a fresh subagent for each
- Step 3: Run the two-stage review for each task: spec-compliance first, then code-quality; finalize with the end-to-end review
Best Practices
- Never skip either the spec-compliance or code-quality review stages.
- Ensure all issues are fixed before advancing to the next stage.
- Do not dispatch multiple implementation subagents in parallel.
- Avoid letting the implementer self-review; rely on dedicated reviewers.
- Keep task scope tightly aligned to the original spec to avoid drift.
Example Use Cases
- Implementing a dashboard with multiple independent widgets.
- Adding distinct API endpoints where each endpoint is a separate task.
- Refactoring isolated components within a module.
- Creating a multi-step onboarding flow with independent steps.
- Sequencing data pipeline tasks where each stage is independent and reviewed separately.
