> ## Documentation Index > Fetch the complete documentation index at: https://mintlify.com/obra/superpowers/llms.txt > Use this file to discover all available pages before exploring further. # Subagent-Driven Development > Dispatch fresh subagents per task with two-stage review for fast, high-quality iteration ## Overview Subagent-Driven Development executes implementation plans by dispatching a fresh subagent for each task, followed by two-stage review: spec compliance first, then code quality. **Core principle:** Fresh subagent per task + two-stage review (spec then quality) = high quality, fast iteration ## When to Use ```mermaid theme={null} graph TD A[Have implementation plan?] -->|Yes| B[Tasks mostly independent?] A -->|No| C[Manual execution or brainstorm first] B -->|Yes| D[Stay in this session?] B -->|No - tightly coupled| C D -->|Yes| E[subagent-driven-development] D -->|No - parallel session| F[executing-plans] ``` Use when: * You have an implementation plan from [Writing Plans](/skills/writing-plans) * Tasks are mostly independent * You want to stay in the current session (not switch to parallel session) **vs. Executing Plans (parallel session):** * Same session (no context switch) * Fresh subagent per task (no context pollution) * Two-stage review after each task * Faster iteration (no human-in-loop between tasks) ## The Process 1. Read plan file once: `docs/plans/YYYY-MM-DD-feature.md` 2. Extract all tasks with full text and context 3. Create TodoWrite with all tasks Read the plan file once upfront. Extract all task details so you can provide them to subagents without making them read the file. Use `./implementer-prompt.md` template to dispatch subagent with: * Full task text (already extracted) * Context about where task fits * Scene-setting (what's already built, what this adds) **Implementer subagent workflow:** 1. May ask questions before starting 2. Implements following TDD 3. Tests the implementation 4. Commits the changes 5. Self-reviews the code **If subagent asks questions:** * Answer clearly and completely * Provide additional context if needed * Let them proceed once answered Use `./spec-reviewer-prompt.md` template to dispatch reviewer: * Provide task spec * Provide what was implemented **Spec reviewer confirms:** * ✅ All spec requirements met * ✅ Nothing extra added (no YAGNI violations) * ❌ Lists any missing requirements * ❌ Lists any extra features **Do NOT proceed to code quality review until spec compliance is ✅** Wrong order wastes effort reviewing code that doesn't match spec. **If reviewer finds issues:** 1. Implementer (same subagent) fixes them 2. Spec reviewer reviews again 3. Repeat until ✅ Use `./code-quality-reviewer-prompt.md` template: * Provide git SHAs for the changes * Ask for code quality assessment **Code reviewer assesses:** * Strengths of implementation * Issues (Important / Minor / Nitpick) * Approval or required changes **If reviewer finds issues:** 1. Implementer fixes them 2. Code reviewer reviews again 3. Repeat until approved Once both reviews approved: * Mark task completed in TodoWrite * Move to next task 1. Dispatch final code reviewer for entire implementation 2. Use `superpowers:finishing-a-development-branch` skill ## Visual Workflow ```mermaid theme={null} graph TD A[Read plan, extract tasks, create TodoWrite] --> B[Dispatch implementer] B --> C{Asks questions?} C -->|Yes| D[Answer questions] D --> B C -->|No| E[Implements, tests, commits, self-reviews] E --> F[Dispatch spec reviewer] F --> G{Spec compliant?} G -->|No| H[Implementer fixes] H --> F G -->|Yes| I[Dispatch code quality reviewer] I --> J{Approved?} J -->|No| K[Implementer fixes] K --> I J -->|Yes| L[Mark complete] L --> M{More tasks?} M -->|Yes| B M -->|No| N[Final review + finishing skill] ``` ## Prompt Templates The skill includes three prompt templates: `./implementer-prompt.md` Dispatch to implement task with TDD `./spec-reviewer-prompt.md` Verify implementation matches spec `./code-quality-reviewer-prompt.md` Assess code quality and suggest improvements ## Example Workflow ``` You: I'm using Subagent-Driven Development to execute this plan. [Read plan file once: docs/plans/2026-03-09-metrics.md] [Extract all 5 tasks with full text and context] [Create TodoWrite with all tasks] --- Task 1: Hook installation script [Get Task 1 text and context (already extracted)] [Dispatch implementation subagent with full task text + context] Implementer: "Before I begin - should the hook be installed at user or system level?" You: "User level (~/.config/superpowers/hooks/)" Implementer: "Got it. Implementing now..." [Later] Implementer: - Implemented install-hook command - Added tests, 5/5 passing - Self-review: Found I missed --force flag, added it - Committed: abc123f [Dispatch spec compliance reviewer] Spec reviewer: ✅ Spec compliant - All requirements met - Nothing extra added [Get git SHAs, dispatch code quality reviewer] Code reviewer: Strengths: Good test coverage, clean implementation Issues: None ✅ Approved [Mark Task 1 complete] --- Task 2: Recovery modes [Get Task 2 text and context (already extracted)] [Dispatch implementation subagent with full task text + context] Implementer: [No questions, proceeds] Implementer: - Added verify/repair modes - 8/8 tests passing - Self-review: All good - Committed: def456a [Dispatch spec compliance reviewer] Spec reviewer: ❌ Issues: - Missing: Progress reporting (spec says "report every 100 items") - Extra: Added --json flag (not requested) [Implementer fixes issues] Implementer: - Removed --json flag - Added progress reporting - Committed: ghi789b [Spec reviewer reviews again] Spec reviewer: ✅ Spec compliant now [Dispatch code quality reviewer] Code reviewer: Strengths: Solid implementation Issues (Important): Magic number (100) should be constant [Implementer fixes] Implementer: - Extracted PROGRESS_INTERVAL constant - Committed: jkl012c [Code reviewer reviews again] Code reviewer: ✅ Approved [Mark Task 2 complete] --- [Tasks 3-5 similar...] --- All tasks complete! [Dispatch final code reviewer] Final reviewer: All requirements met, ready to merge Using finishing-a-development-branch skill... ``` ## Advantages ### vs. Manual Execution * Subagents follow TDD naturally * Fresh context per task (no confusion) * Parallel-safe (subagents don't interfere) * Subagent can ask questions (before AND during work) ### vs. Executing Plans * Same session (no handoff) * Continuous progress (no waiting) * Review checkpoints automatic ### Efficiency Gains * No file reading overhead (controller provides full text) * Controller curates exactly what context is needed * Subagent gets complete information upfront * Questions surfaced before work begins (not after) ### Quality Gates * Self-review catches issues before handoff * **Two-stage review: spec compliance, then code quality** * Review loops ensure fixes actually work * Spec compliance prevents over/under-building * Code quality ensures implementation is well-built ### Cost **Higher invocation cost:** * More subagent invocations (implementer + 2 reviewers per task) * Controller does more prep work (extracting all tasks upfront) * Review loops add iterations **But catches issues early (cheaper than debugging later)** ## Red Flags **Never:** * Start implementation on main/master branch without explicit user consent * 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) * Skip scene-setting context (subagent needs to understand where task fits) * Ignore subagent questions (answer before letting them proceed) * Accept "close enough" on spec compliance (spec reviewer found issues = not done) * Skip review loops (reviewer found issues = implementer fixes = review again) * Let implementer self-review replace actual review (both are needed) * **Start code quality review before spec compliance is ✅** (wrong order) * Move to next task while either review has open issues **Critical: Review Order** Spec compliance MUST be ✅ before code quality review starts. Reviewing code quality for code that doesn't match spec wastes effort. ## Handling Issues ### If Subagent Asks Questions * Answer clearly and completely * Provide additional context if needed * Don't rush them into implementation * Questions are good - they prevent wrong work ### If Reviewer Finds Issues 1. Implementer (same subagent) fixes them 2. Reviewer reviews again 3. Repeat until approved 4. Don't skip the re-review ### If Subagent Fails Task * Dispatch fix subagent with specific instructions * Don't try to fix manually (context pollution) * Provide clear context about what failed and why ## Checklist Per task: * [ ] Extracted task text and context upfront * [ ] Dispatched implementer with complete information * [ ] Answered any questions before proceeding * [ ] Implementer completed (implemented, tested, committed, self-reviewed) * [ ] Dispatched spec compliance reviewer * [ ] Spec reviewer approved (or issues fixed and re-reviewed) * [ ] Dispatched code quality reviewer (only after spec ✅) * [ ] Code reviewer approved (or issues fixed and re-reviewed) * [ ] Marked task complete in TodoWrite After all tasks: * [ ] Dispatched final code reviewer * [ ] Used finishing-a-development-branch skill ## Related Skills * **[Writing Plans](/skills/writing-plans)**: Creates the plan this skill executes * **[Executing Plans](/skills/executing-plans)**: Alternative for parallel session execution * **[Test-Driven Development](/skills/test-driven-development)**: Subagents use TDD for implementation Two-stage review (spec then quality) catches different types of issues. Spec compliance ensures you built the right thing. Code quality ensures you built it right.