Subagent Driven Development

Subagent-Driven Development

Execute plan by dispatching fresh subagent per task, with two-stage review after each: spec compliance review first, then code quality review.

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

When to Use

• Have an implementation plan (spool change with tasks.md)
• Tasks are mostly independent
• Want to stay in this session (vs. parallel session with spool-apply-change-proposal)

vs. spool-apply-change-proposal:

• 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)

The Process

1. Setup: Read plan, extract all tasks, set up tracking
2. Per Task:
   • Dispatch implementer subagent
   • Answer any questions
   • Implementer implements, tests, commits, self-reviews
   • Dispatch spec reviewer subagent
   • If issues: implementer fixes, re-review
   • Dispatch code quality reviewer subagent
   • If issues: implementer fixes, re-review
   • Mark task complete: spool tasks complete <change-id> <task-id>
3. Completion: Dispatch final code reviewer, then use spool-finishing-a-development-branch

Setup

# Get the change context
spool agent instruction apply --change <change-id>

# Read tasks.md
cat .spool/changes/<change-id>/tasks.md

# Extract all tasks with full text and context upfront

Per Task Workflow

1. Mark Task Started

spool tasks start <change-id> <task-id>

2. Dispatch Implementer Subagent

Provide:

• Full task text (not just reference)
• Context: what came before, what comes after
• Relevant file paths
• Expected outcome

Subagent implements using TDD:

1. Write failing test
2. Run to confirm failure
3. Implement
4. Run to confirm pass
5. Commit

3. Spec Compliance Review

Dispatch spec reviewer subagent with:

• The task specification
• Git diff of changes

Reviewer checks:

• All spec requirements met?
• No extra functionality added?
• Correct files modified?

If issues: implementer subagent fixes, re-review until ✅

4. Code Quality Review

Dispatch code quality reviewer subagent with:

• Git SHAs for review
• Code review template

Reviewer checks:

• Code quality
• Test coverage
• Style/conventions

If issues: implementer subagent fixes, re-review until ✅

5. Mark Task Complete

spool tasks complete <change-id> <task-id>

6. Next Task or Finish

If more tasks: repeat from step 1 If done: dispatch final reviewer, then use spool-finishing-a-development-branch

Prompt Templates

• ./implementer-prompt.md - Dispatch implementer subagent
• ./spec-reviewer-prompt.md - Dispatch spec compliance reviewer subagent
• ./code-quality-reviewer-prompt.md - Dispatch code quality reviewer subagent

Red Flags

Never:

• Start implementation on main/master without explicit user consent
• Skip reviews (spec compliance OR code quality)
• Proceed with unfixed issues
• Dispatch multiple implementation subagents in parallel (conflicts)
• Make subagent read plan file (provide full text instead)
• Skip scene-setting context
• Accept "close enough" on spec compliance
• Start code quality review before spec compliance is ✅
• Move to next task while either review has open issues

If subagent asks questions:

• Answer clearly and completely
• Provide additional context if needed

If reviewer finds issues:

• Implementer fixes them
• Reviewer reviews again
• Repeat until approved

Integration

Required workflow skills:

• spool-using-git-worktrees - Set up isolated workspace before starting
• spool-write-change-proposal - Creates the plan this skill executes
• spool-requesting-code-review - Code review template for reviewer subagents
• spool-finishing-a-development-branch - Complete development after all tasks

Subagents should use:

• spool-test-driven-development - Subagents follow TDD for each task

Alternative workflow:

• spool-apply-change-proposal - Use for human-in-loop execution with batch checkpoints