4.7 KiB
4.7 KiB
| name | description |
|---|---|
| subagent-driven-development | Use when executing implementation plans with independent tasks in the current session |
Subagent-Driven Development
Execute plan by dispatching fresh subagent per task, with two-stage review after each: spec compliance review first, then code quality review.
Why subagents: You delegate tasks to specialized agents with isolated context. They should never inherit your session's context or history — you construct exactly what they need. This preserves your own context for coordination work.
Core principle: Fresh subagent per task + two-stage review (spec then quality) = high quality, fast iteration
When to Use
Use when you have an implementation plan (from writing-plans skill) with multiple independent tasks.
The Process
- Read plan file —
docs/superpowers/plans/[name].md - Extract all tasks with full text and context
- Create TodoWrite with all tasks
- For each task: a. Dispatch implementer subagent with full task text + context b. If implementer asks questions → answer, provide context, re-dispatch c. Implementer implements, tests, commits, self-reviews d. Dispatch spec reviewer subagent → confirms code matches spec e. If spec issues → implementer fixes, re-review f. Dispatch code quality reviewer subagent → reviews quality g. If quality issues → implementer fixes, re-review h. Mark task complete in TodoWrite
- After all tasks: Dispatch final code reviewer for entire implementation
- Invoke finishing-a-development-branch to complete
Implementer Subagent Prompt Template
When dispatching an implementer subagent, provide:
You are implementing Task [N] from the implementation plan.
## Task
[Full task text from plan]
## Context
- Project: Construction Delivery Control (React + Supabase)
- Stack: React 18, React Router 7, Tailwind CSS, Framer Motion, Vitest
- Test convention: TDD — write failing test first, minimal code to pass
- File conventions: Components by domain in src/components/, services in src/services/
## Files to modify
[List exact file paths]
## Requirements
- Follow TDD: write failing test first, then minimal code
- Commit after each passing test
- Self-review your work before reporting complete
- Do NOT modify files outside this task's scope
Report status when done: DONE, DONE_WITH_CONCERNS, NEEDS_CONTEXT, or BLOCKED.
Spec Reviewer Subagent Prompt Template
Review whether the implementation matches the spec for Task [N].
## Task Spec
[Full task text from plan]
## What was implemented
[Summary from implementer]
## Files changed
[git diff or file list]
Check:
1. Does the code implement all requirements from the task?
2. Does the code add anything NOT requested?
3. Are there any missing edge cases or error handling from the spec?
Report: ✅ Spec compliant, or ❌ with specific issues.
Code Quality Reviewer Subagent Prompt Template
Review the code quality for Task [N].
## What was implemented
[Summary]
## Files changed
[git diff between BASE_SHA and HEAD_SHA]
Review for:
1. Code clarity and readability
2. Proper error handling
3. Consistent naming and style (follow existing project conventions)
4. Test quality and coverage
5. No magic numbers or hardcoded values
6. Proper separation of concerns
Report strengths, issues (with severity), and approval status.
Handling Implementer Status
- DONE: Proceed to spec compliance review
- DONE_WITH_CONCERNS: Read concerns before proceeding. Address if about correctness/scope.
- NEEDS_CONTEXT: Provide missing context and re-dispatch
- BLOCKED: Assess blocker — context issue? Provide context. Reasoning issue? Use more capable model. Plan issue? Escalate to human
Model Selection
- Mechanical tasks (1-2 files, clear spec) → use fast/cheap model
- Integration tasks (multi-file, coordination) → use standard model
- Architecture/review → use most capable available model
Red Flags
Never:
- Start implementation on main/master branch
- Skip reviews (spec compliance OR code quality)
- Proceed with unfixed issues
- Dispatch multiple implementation subagents in parallel (conflicts)
- Make subagent read plan file (provide full text instead)
- Ignore subagent questions
- Accept "close enough" on spec compliance
- Start code quality review before spec compliance is ✅
- Move to next task while either review has open issues
Integration
Required workflow skills:
- superpowers:using-git-worktrees — REQUIRED: Set up isolated workspace before starting
- superpowers:writing-plans — Creates the plan this skill executes
- superpowers:finishing-a-development-branch — Complete development after all tasks