Develop Code with Subagent Reviews

Subagent-Driven Development executes an implementation plan by dispatching a fresh subagent per task, followed by two-stage review after each - spec compliance review first, then code quality review. The core principle is that a fresh subagent plus two-stage review yields high quality and fast iteration, all within the same session (no context switch, unlike a parallel-session alternative).

Applicability follows a decision tree:

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"];
}

No plan or tightly-coupled tasks mean manual execution or more planning first; a parallel session instead uses the sibling executing-plans skill.

Per task, the controller dispatches an implementer subagent (./implementer-prompt.md) with the full task text and context (never making the subagent read the plan file itself); the implementer asks clarifying questions if needed, then implements, tests, self-reviews, and commits. A spec reviewer subagent (./spec-reviewer-prompt.md) checks the result against the spec - any gap sends it back to the same implementer to fix, then re-review, repeating until compliant. Only then does a code quality reviewer subagent (./code-quality-reviewer-prompt.md) check the implementation, with the same fix-then-re-review loop until approved, before the task is marked complete in TodoWrite. After every task is done, a final code reviewer subagent reviews the entire implementation before handing off to superpowers:finishing-a-development-branch.

Required sibling workflow skills: superpowers:writing-plans creates the plan this skill executes, superpowers:requesting-code-review supplies the review template for reviewer subagents, and dispatched implementer subagents themselves follow superpowers:test-driven-development. Explicit red flags: never skip either review stage, never run multiple implementer subagents in parallel (causes conflicts), never accept "close enough" on spec compliance, never start the code-quality review before spec compliance is confirmed, and never let self-review substitute for the actual reviewer pass - both are required.

Developers executing a multi-task implementation plan in a single session who want automatic quality gates - spec compliance before code quality, both with fix-and-re-review loops - rather than manually tracking task completion and review state themselves.