superpowers-lite/skills/subagent-driven-development/SKILL.md

---
name: subagent-driven-development
description: Manual-only skill. Activate only after an explicit user request for superpowers-lite:subagent-driven-development.
---

# Subagent-Driven Development

Execute plan in the current branch, dispatching implementation help as needed, then run one overall review after all phases are complete.

**Why subagents:** You delegate tasks to specialized agents with isolated context. By precisely crafting their instructions and context, you ensure they stay focused and succeed at their task. They should never inherit your session's context or history — you construct exactly what they need. This also preserves your own context for coordination work.

**Core principle:** Complete the full plan in one current-branch run, then review the whole implementation once. Use phase reviews only when the user asks or risk justifies the overhead.

**Continuous execution:** Do not pause to check in with your human partner between phases unless blocked. Execute all tasks from the plan in the current branch without stopping. The only reasons to stop are: current branch is `main` or `master`, BLOCKED status you cannot resolve, ambiguity that genuinely prevents progress, or all tasks complete. "Should I continue?" prompts and progress summaries waste their time — they asked you to execute the plan, so execute it.

## When to Use

```dot
digraph when_to_use {
    "Have implementation plan?" [shape=diamond];
    "Tasks mostly independent?" [shape=diamond];
    "Stay in this session?" [shape=diamond];
    "subagent-driven-development" [shape=box];
    "executing-plans" [shape=box];
    "Manual execution or brainstorm first" [shape=box];

    "Have implementation plan?" -> "Tasks mostly independent?" [label="yes"];
    "Have implementation plan?" -> "Manual execution or brainstorm first" [label="no"];
    "Tasks mostly independent?" -> "Stay in this session?" [label="yes"];
    "Tasks mostly independent?" -> "Manual execution or brainstorm first" [label="no - tightly coupled"];
    "Stay in this session?" -> "subagent-driven-development" [label="yes"];
    "Stay in this session?" -> "executing-plans" [label="no - parallel session"];
}
```

**vs. Executing Plans (parallel session):**
- Same session (no context switch)
- Fresh subagent per phase (no context pollution)
- One overall review after all phases are complete by default
- Faster iteration (no human-in-loop between tasks)

## Phase Grouping

Before grouping phases, confirm the current git branch is not `main` or `master` with `git branch --show-current`. If it is, stop and ask the user to switch or create a feature branch. Do not create or switch to a worktree unless the user explicitly asks.

Before dispatching implementers, group tasks into phases:

- Group adjacent tasks when they touch the same files, complete one coherent behavior, or would create review churn if split
- Keep a task as its own phase when it is risky, broad, independent, or likely to need a focused review
- Never group tasks just to hide uncertainty; if grouping makes requirements harder to verify, split the phase
- Record phase membership in TodoWrite so progress is visible at both phase and task level
- Do not review between phases by default; finish all phases first, then review the whole implementation

## The Process

```dot
digraph process {
    rankdir=TB;

    subgraph cluster_per_phase {
        label="Per Phase";
        "Dispatch implementer subagent (./implementer-prompt.md)" [shape=box];
        "Implementer subagent asks questions?" [shape=diamond];
        "Answer questions, provide context" [shape=box];
        "Implementer subagent implements, tests, commits, self-reviews" [shape=box];
        "Mark phase tasks complete in TodoWrite" [shape=box];
    }

    "Read plan, extract all tasks with full text, group into phases, note context, create TodoWrite" [shape=box];
    "More phases remain?" [shape=diamond];
    "Dispatch final code reviewer subagent for entire implementation" [shape=box];
    "Fix Critical/Important findings and re-run verification" [shape=box];
    "Final review approved?" [shape=diamond];
    "Use superpowers-lite:finishing-a-development-branch" [shape=box style=filled fillcolor=lightgreen];

    "Read plan, extract all tasks with full text, group into phases, note context, create TodoWrite" -> "Dispatch implementer subagent (./implementer-prompt.md)";
    "Dispatch implementer subagent (./implementer-prompt.md)" -> "Implementer subagent asks questions?";
    "Implementer subagent asks questions?" -> "Answer questions, provide context" [label="yes"];
    "Answer questions, provide context" -> "Dispatch implementer subagent (./implementer-prompt.md)";
    "Implementer subagent asks questions?" -> "Implementer subagent implements, tests, commits, self-reviews" [label="no"];
    "Implementer subagent implements, tests, commits, self-reviews" -> "Mark phase tasks complete in TodoWrite";
    "Mark phase tasks complete in TodoWrite" -> "More phases remain?";
    "More phases remain?" -> "Dispatch implementer subagent (./implementer-prompt.md)" [label="yes"];
    "More phases remain?" -> "Dispatch final code reviewer subagent for entire implementation" [label="no"];
    "Dispatch final code reviewer subagent for entire implementation" -> "Final review approved?";
    "Final review approved?" -> "Fix Critical/Important findings and re-run verification" [label="no"];
    "Fix Critical/Important findings and re-run verification" -> "Dispatch final code reviewer subagent for entire implementation" [label="re-review if substantial"];
    "Final review approved?" -> "Use superpowers-lite:finishing-a-development-branch" [label="yes"];
}
```

## Model Selection

Use the least powerful model that can handle each role to conserve cost and increase speed.

**Mechanical implementation tasks** (isolated functions, clear specs, 1-2 files): use a fast, cheap model. Most implementation tasks are mechanical when the plan is well-specified.

**Integration and judgment tasks** (multi-file coordination, pattern matching, debugging): use a standard model.

**Architecture, design, and review tasks**: use the most capable available model.

**Task complexity signals:**
- Touches 1-2 files with a complete spec → cheap model
- Touches multiple files with integration concerns → standard model
- Requires design judgment or broad codebase understanding → most capable model

## Handling Implementer Status

Implementer subagents report one of four statuses. Handle each appropriately:

**DONE:** Mark the phase complete and continue to the next phase. Do not review yet unless the user explicitly requested phase reviews.

**DONE_WITH_CONCERNS:** The implementer completed the work but flagged doubts. Read the concerns before proceeding. If the concerns affect correctness or scope, address them before moving to the next phase. If they're observations (e.g., "this file is getting large"), note them and continue; the overall review will evaluate the full implementation.

**NEEDS_CONTEXT:** The implementer needs information that wasn't provided. Provide the missing context and re-dispatch.

**BLOCKED:** The implementer cannot complete the task. Assess the blocker:
1. If it's a context problem, provide more context and re-dispatch with the same model
2. If the task requires more reasoning, re-dispatch with a more capable model
3. If the task is too large, break it into smaller pieces
4. If the plan itself is wrong, escalate to the human

**Never** ignore an escalation or force the same model to retry without changes. If the implementer said it's stuck, something needs to change.

## Prompt Templates

- `./implementer-prompt.md` - Dispatch implementer subagent
- `./spec-reviewer-prompt.md` - Optional phase-level spec review when explicitly requested or unusually risky
- `./code-quality-reviewer-prompt.md` - Overall code quality review after all phases are complete

## Example Workflow

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

[Read plan file once: .superpowers-lite/plans/feature-plan.md]
[Extract all 5 tasks with full text and context]
[Group into 3 phases: Phase 1 = Task 1, Phase 2 = Tasks 2-3, Phase 3 = Tasks 4-5]
[Create TodoWrite with all phases and tasks]

Phase 1: Hook installation script

[Get Task 1 text and context (already extracted)]
[Dispatch implementation subagent with full phase 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

[Mark Phase 1 tasks complete]

Phase 2: Recovery modes and CLI wiring

[Get Task 2 and Task 3 text and context (already extracted)]
[Dispatch implementation subagent with full phase text + context]

Implementer: [No questions, proceeds]
Implementer:
  - Added verify/repair modes
  - 8/8 tests passing
  - Self-review: All good
  - Committed

[Mark Phase 2 tasks complete]

...

[After all tasks]
[Dispatch final code-reviewer]
Final reviewer:
  - Important: Missing progress reporting (spec says "report every 100 items")
  - Minor: Magic number (100)

[Fix findings, rerun relevant verification]
[Re-review if substantial fixes were required]
Final reviewer: All requirements met, ready to merge

Done!
```

## Advantages

**vs. Manual execution:**
- Subagents follow TDD naturally
- Fresh context per phase (less confusion)
- Parallel-safe when phases do not share files (subagents do not interfere)
- Subagent can ask questions (before AND during work)

**vs. Executing Plans:**
- Same session (no handoff)
- Continuous progress (no waiting)
- One overall review after all implementation work

**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
- Overall code review checks spec compliance and code quality after the full implementation exists
- Review loops ensure substantial fixes actually work

**Cost:**
- Fewer review subagent invocations than phase-by-phase review
- Controller does more prep work (extracting all tasks upfront)
- Final review loops may add iterations
- But catches issues early (cheaper than debugging later)

## Red Flags

**Never:**
- Start implementation on `main` or `master`
- Skip the final overall code review
- Proceed with unfixed issues
- Dispatch multiple implementation subagents in parallel when phases share files or state
- 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 final review findings
- Skip review loops for substantial fixes
- Let implementer self-review replace actual review (both are needed)
- Treat phase self-review as the final code review

**If subagent asks questions:**
- Answer clearly and completely
- Provide additional context if needed
- Don't rush them into implementation

**If reviewer finds issues:**
- Fix Critical and Important issues
- Re-run relevant verification
- Re-review if fixes were substantial

**If subagent fails a phase:**
- Dispatch fix subagent with specific instructions
- Don't try to fix manually (context pollution)

## Integration

**Required workflow skills:**
- **superpowers-lite:writing-plans** - Creates the plan this skill executes
- **superpowers-lite:requesting-code-review** - Overall code review after all phases are complete
- **superpowers-lite:finishing-a-development-branch** - Complete development after all tasks

**Optional workflow skills:**
- **superpowers-lite:using-git-worktrees** - Use only when the user explicitly asks for an isolated workspace

**Subagents should use:**
- **superpowers-lite:test-driven-development** - Subagents follow TDD for each task

**Alternative workflow:**
- **superpowers-lite:executing-plans** - Use for parallel session instead of same-session execution