Agent Skills: Subagent-Driven Development

Subagent-Driven Development

Execute plan by dispatching fresh subagent per task, with three-stage review after each:

1. Spec compliance review
2. Code quality review (ios-reviewer or backend-reviewer)
3. UX review

Core principle: Fresh subagent per task + three-stage review = high quality with great UX

When to Use

• Have an implementation plan with defined tasks
• Tasks are mostly independent (can be done in any order)
• Want to stay in the current session (no parallel sessions)

The 3-Stage Review Process

┌─────────────────────────────────────────────────────────────┐
│                    PER-TASK WORKFLOW                        │
├─────────────────────────────────────────────────────────────┤
│  1. IMPLEMENT      Implementer subagent builds the feature  │
│         ↓                                                   │
│  2. SPEC REVIEW    Does code match requirements?            │
│         ↓          (./spec-reviewer-prompt.md)              │
│  3. CODE REVIEW    Is code quality good?                    │
│         ↓          (ios-reviewer or backend-reviewer)       │
│  4. UX REVIEW      Is user experience good?                 │
│         ↓          (ux-reviewer)                            │
│  5. COMPLETE       Task marked done                         │
└─────────────────────────────────────────────────────────────┘

Stage Details

Stage 1: Implementation

• Dispatch implementer subagent with full task text
• Implementer asks questions if unclear
• Implementer implements, tests, self-reviews, commits

Stage 2: Spec Compliance Review

• Verify implementer built what was requested (nothing more, nothing less)
• If issues found → implementer fixes → re-review
• Only proceed when ✅ spec compliant

Stage 3: Code Quality Review

• Use ios-reviewer for Swift/SwiftUI code
• Use backend-reviewer for Supabase/TypeScript code
• Check patterns, performance, security
• If issues found → implementer fixes → re-review
• Only proceed when ✅ code approved

Stage 4: UX Review

• Use ux-reviewer for all user-facing code
• Check flow clarity, feedback states, accessibility
• If issues found → implementer fixes → re-review
• Only proceed when ✅ UX approved

Prompt Templates

• ./implementer-prompt.md - Dispatch implementer subagent
• ./spec-reviewer-prompt.md - Dispatch spec compliance reviewer
• ./code-quality-reviewer-prompt.md - Dispatch code quality reviewer (uses ios-reviewer or backend-reviewer agent)
• ./ux-reviewer-prompt.md - Dispatch UX reviewer

Choosing the Right Code Reviewer

| Code Type | Agent | |-----------|-------| | Swift, SwiftUI, SwiftData, iOS | ios-reviewer | | TypeScript, Supabase, Edge Functions, RLS | backend-reviewer | | Mixed (full-stack feature) | Both in sequence |

Example Workflow

You: I'm using Subagent-Driven Development to execute this plan.

[Read plan file: docs/plans/feature-plan.md]
[Extract all tasks, create TodoWrite]

Task 1: Add pour deletion with confirmation

[Dispatch implementer with full task text]

Implementer:
  - Added delete button to PourDetailView
  - Confirmation dialog implemented
  - SwiftData deletion with sync
  - 3 tests passing
  - Committed

[Dispatch spec reviewer]
Spec reviewer: ✅ Spec compliant - all requirements met

[Dispatch ios-reviewer for code quality]
iOS reviewer:
  Strengths: Good SwiftData usage, proper @MainActor
  Issues (Important): Missing haptic feedback on delete

[Implementer fixes]
Implementer: Added haptic feedback

[Re-dispatch ios-reviewer]
iOS reviewer: ✅ Approved

[Dispatch ux-reviewer]
UX reviewer:
  Strengths: Clear confirmation dialog, good error message
  Issues (Important): No undo option after deletion
  Suggestions: Add "Deleted" toast with undo button

[Implementer fixes]
Implementer: Added undo toast with 5-second window

[Re-dispatch ux-reviewer]
UX reviewer: ✅ UX approved - excellent recovery flow

[Mark Task 1 complete]

Task 2: ...

Advantages

Quality gates:

• Self-review catches obvious issues
• Spec review prevents over/under-building
• Code review ensures technical quality
• UX review ensures user success (often missed by technical reviews)

Fresh context:

• Each subagent starts clean (no accumulated confusion)
• Implementer can ask questions before starting
• Reviewers evaluate objectively

Fast iteration:

• No human-in-loop between tasks
• Issues caught early (cheaper than debugging later)
• Review loops ensure fixes actually work

Red Flags

Never:

• Skip any of the three review stages
• Proceed with unfixed issues from any reviewer
• Start code quality review before spec compliance passes
• Start UX review before code quality passes
• Dispatch multiple implementation subagents in parallel
• Make subagent read plan file (provide full text instead)
• Accept "close enough" on any review stage

If reviewer finds issues:

• Same implementer subagent fixes them
• Same reviewer re-reviews
• Repeat until approved
• Don't skip re-reviews

Integration

Required skills:

• indie-stack:writing-plans - Creates the plan this skill executes
• indie-stack:requesting-code-review - Code review guidance

Reviewers use:

• indie-stack:ios-reviewer - Swift/SwiftUI code quality
• indie-stack:backend-reviewer - Supabase/TypeScript quality
• indie-stack:ux-reviewer - User experience quality

Implementers should use:

• indie-stack:testing-flexible - Testing guidance (encouraged, not blocking)