subagent-driven-development

name: subagent-driven-development description: Executes implementation plans by dispatching a fresh subagent per independent task with two-stage reviews. Use when executing implementation plans with independent tasks in the current session.

Subagent-Driven Development

Execute plan by dispatching a fresh subagent per task, with two-stage review after each: 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 this skill

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]

vs. executing-plans:

• 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)

How to use it

Setup (once)

1. Read plan file
2. Extract all tasks with full text and context
3. Create task list with all items

Per Task Loop

1. Dispatch implementer subagent (implementer-prompt.md)
   → If questions: Answer them → Re-dispatch
   → Subagent implements, tests, commits, self-reviews

2. Dispatch spec reviewer subagent (spec-reviewer-prompt.md)
   → If issues: Implementer fixes → Re-review
   → Until ✅ spec compliant

3. Dispatch code quality reviewer (code-quality-reviewer-prompt.md)
   → If issues: Implementer fixes → Re-review
   → Until ✅ approved

4. Mark task complete → Next task

After All Tasks

1. Dispatch final code reviewer for entire implementation
2. Use finishing-a-development-branch skill

Prompt Templates

Example Workflow

[Read plan once, extract all 5 tasks, create task list]

── Task 1: Hook installation script ──

[Dispatch implementer with full task text + context]

Implementer: "Should hook install at user or system level?"
You: "User level (~/.config/superpowers/hooks/)"

Implementer: Implemented, 5/5 tests passing, committed.

[Dispatch spec reviewer]
Spec reviewer: ✅ Spec compliant

[Dispatch code quality reviewer]
Code reviewer: ✅ Approved

[Mark Task 1 complete]

── Task 2: Recovery modes ──

[Dispatch implementer]
Implementer: 8/8 tests passing, committed.

[Dispatch spec reviewer]
Spec reviewer: ❌ Missing progress reporting (spec: "every 100 items"), extra --json flag

[Implementer fixes]
Spec reviewer: ✅ Spec compliant

[Dispatch code quality reviewer]
Code reviewer: ❌ Magic number (100) for interval

[Implementer fixes → extracts PROGRESS_INTERVAL constant]
Code reviewer: ✅ Approved

[Mark Task 2 complete] → ...

[After all tasks: final review → finishing-a-development-branch]

Advantages

Red Flags

Never:

• Start on main/master without explicit user consent
• Skip spec compliance review
• Start code quality review before spec is ✅
• Move to next task with open review issues
• Dispatch multiple implementers in parallel (conflicts)
• Make subagent read the plan file (provide full text instead)
• Skip the re-review after fixes (reviewer approved = actually re-reviewed)
• Accept "close enough" on spec compliance

If subagent asks questions: Answer fully before letting them proceed.

If reviewer finds issues: Implementer fixes → reviewer re-reviews → repeat until approved.

If subagent fails: Dispatch a fix subagent with specific instructions. Don't fix manually.

Integration