subagent-driven-development

name	subagent-driven-development
description	Use when executing implementation plans with independent tasks in the current session

Subagent-Driven Development

Execute plan by dispatching fresh browser subagents or breaking work into isolated task boundaries per task, with two-stage review after each: spec compliance review first, then code quality review.

Core principle: Fresh context per task + two-stage review (spec then quality) = high quality, fast iteration

When to Use

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"];
}

The Process (Antigravity)

In Antigravity, you manage the orchestration loop directly:

digraph process {
    rankdir=TB;

    "Read plan, extract all tasks with full text, note context, create task.md" [shape=box];
    "Pick next task" [shape=box];
    "Implement task using task_boundary EXECUTION mode" [shape=box];
    "Run verification commands" [shape=box];
    "Spec compliance self-review: does code match spec?" [shape=diamond];
    "Fix spec gaps" [shape=box];
    "Code quality self-review: clean, no debt, good names?" [shape=diamond];
    "Fix quality issues" [shape=box];
    "Mark task complete in task.md" [shape=box];
    "More tasks remain?" [shape=diamond];
    "Use finishing-a-development-branch skill" [shape=box style=filled fillcolor=lightgreen];

    "Read plan, extract all tasks with full text, note context, create task.md" -> "Pick next task";
    "Pick next task" -> "Implement task using task_boundary EXECUTION mode";
    "Implement task using task_boundary EXECUTION mode" -> "Run verification commands";
    "Run verification commands" -> "Spec compliance self-review: does code match spec?";
    "Spec compliance self-review: does code match spec?" -> "Fix spec gaps" [label="no"];
    "Fix spec gaps" -> "Spec compliance self-review: does code match spec?" [label="re-review"];
    "Spec compliance self-review: does code match spec?" -> "Code quality self-review: clean, no debt, good names?" [label="yes"];
    "Code quality self-review: clean, no debt, good names?" -> "Fix quality issues" [label="no"];
    "Fix quality issues" -> "Code quality self-review: clean, no debt, good names?" [label="re-review"];
    "Code quality self-review: clean, no debt, good names?" -> "Mark task complete in task.md" [label="yes"];
    "Mark task complete in task.md" -> "More tasks remain?";
    "More tasks remain?" -> "Pick next task" [label="yes"];
    "More tasks remain?" -> "Use finishing-a-development-branch skill" [label="no"];
}

In Antigravity: How to Orchestrate

Since you are the orchestrator in Antigravity (rather than spawning separate agent processes), follow this pattern per task:

1. Announce: "Implementing Task N: [description]"
2. Set task_boundary to EXECUTION mode for this specific task
3. Implement following the plan step-by-step (use TDD skill)
4. Verify: Run all commands specified in plan, check output
5. Spec Review: Re-read the task spec. Does code do exactly what spec says? Nothing more, nothing less?
6. Quality Review: Is the code clean? Good names? No magic numbers? No duplication?
7. Fix and re-verify if either review fails
8. Update task.md marking the task complete
9. Move to next task

Prompt Templates for Reference

The original prompt templates are in this directory:

• ./implementer-prompt.md — Use as a self-checklist when implementing each task
• ./spec-reviewer-prompt.md — Use as a spec compliance checklist
• ./code-quality-reviewer-prompt.md — Use as a code quality checklist

Red Flags

Never:

• Start implementation on main/master branch without explicit user consent
• Skip spec compliance review before quality review
• Proceed with unfixed issues
• Make parallel changes that could conflict

Always:

• Read the full task description from the plan (don't rely on memory)
• Run verification commands after implementation
• Complete spec review BEFORE quality review (order matters)
• Fix issues from each review before moving to next task

Integration

Required workflow skills:

• using-git-worktrees — REQUIRED: Set up isolated workspace before starting
• writing-plans — Creates the plan this skill executes
• requesting-code-review — Code review checklist for each task
• finishing-a-development-branch — Complete development after all tasks
• test-driven-development — Follow TDD for each implementation step