Subagent-Driven Development (SDD)
Overview
SDD uses specialized agents to implement and review code. Instead of doing everything yourself, you:
- Plan the work (break into tasks)
- Dispatch an implementer agent per task
- Review with specialized reviewers
- Fix and re-review until quality passes
Workflow
1. Create a Plan
Write a plan with clear, independent tasks. Use the superteam-tasks format:
```superteam-tasks
- title: Create user model
description: Define User type with validation
files: [src/models/user.ts, src/models/user.test.ts]
- title: Add authentication
description: JWT-based auth with login/logout
files: [src/auth.ts, src/auth.test.ts]
```
Each task should be:
- Independent — can be implemented without other tasks
- Testable — has clear success criteria
- Small — completable in one agent session
2. Load and Run
/sdd load plan.md — Load the plan
/sdd run — Run SDD for current task
/sdd status — Check progress
/sdd next — Skip to next task
3. Review Cycle
For each task, the SDD orchestrator:
- Dispatches implementer (with TDD enforcement)
- Runs spec review (does code match spec?)
- Runs quality review (is code well-written?)
- Optionally runs security and performance reviews in parallel
- On failure: re-dispatches implementer with specific findings
- On max retries: escalates to you
4. Available Agents
| Agent |
Role |
Tools |
| scout |
Fast recon, find relevant code |
read, grep, find, ls, bash |
| implementer |
TDD implementation |
all tools |
| spec-reviewer |
Verify spec compliance |
read, grep, find, ls |
| quality-reviewer |
Code + test quality |
read, grep, find, ls |
| security-reviewer |
Vulnerability scanning |
read, grep, find, ls |
| performance-reviewer |
Performance analysis |
read, grep, find, ls |
| architect |
Design + structure review |
read, grep, find, ls |
5. Direct Team Use
You can also use the team tool directly without SDD:
"Dispatch scout to find all database access patterns"
"Run security-reviewer and performance-reviewer in parallel on src/auth/"
"Chain: scout finds the code, then architect reviews the structure"
Best Practices
- Small tasks — each task should touch 1-3 files
- Clear specs — reviewers need to know what to check against
- Don't skip reviews — they catch real bugs
- Fix, don't argue — when a reviewer finds an issue, fix it
- Monitor costs — use
/team to check session cost
Converted and distributed by TomeVault — claim your Tome and manage your conversions.
1---2name: superteam-subagent-driven-development3description: Break work into tasks, dispatch specialized agents, review iteratively Use when this capability is needed.4---56# Subagent-Driven Development (SDD)78## Overview910SDD uses specialized agents to implement and review code. Instead of doing everything yourself, you:111. Plan the work (break into tasks)122. Dispatch an implementer agent per task133. Review with specialized reviewers144. Fix and re-review until quality passes1516## Workflow1718### 1. Create a Plan1920Write a plan with clear, independent tasks. Use the `superteam-tasks` format:2122````markdown23```superteam-tasks24- title: Create user model25 description: Define User type with validation26 files: [src/models/user.ts, src/models/user.test.ts]27- title: Add authentication28 description: JWT-based auth with login/logout29 files: [src/auth.ts, src/auth.test.ts]30```31````3233Each task should be:34- **Independent** — can be implemented without other tasks35- **Testable** — has clear success criteria36- **Small** — completable in one agent session3738### 2. Load and Run3940```41/sdd load plan.md — Load the plan42/sdd run — Run SDD for current task43/sdd status — Check progress44/sdd next — Skip to next task45```4647### 3. Review Cycle4849For each task, the SDD orchestrator:501. Dispatches **implementer** (with TDD enforcement)512. Runs **spec review** (does code match spec?)523. Runs **quality review** (is code well-written?)534. Optionally runs **security** and **performance** reviews in parallel545. On failure: re-dispatches implementer with specific findings556. On max retries: escalates to you5657### 4. Available Agents5859| Agent | Role | Tools |60|-------|------|-------|61| scout | Fast recon, find relevant code | read, grep, find, ls, bash |62| implementer | TDD implementation | all tools |63| spec-reviewer | Verify spec compliance | read, grep, find, ls |64| quality-reviewer | Code + test quality | read, grep, find, ls |65| security-reviewer | Vulnerability scanning | read, grep, find, ls |66| performance-reviewer | Performance analysis | read, grep, find, ls |67| architect | Design + structure review | read, grep, find, ls |6869### 5. Direct Team Use7071You can also use the `team` tool directly without SDD:7273```74"Dispatch scout to find all database access patterns"75"Run security-reviewer and performance-reviewer in parallel on src/auth/"76"Chain: scout finds the code, then architect reviews the structure"77```7879## Best Practices8081- **Small tasks** — each task should touch 1-3 files82- **Clear specs** — reviewers need to know what to check against83- **Don't skip reviews** — they catch real bugs84- **Fix, don't argue** — when a reviewer finds an issue, fix it85- **Monitor costs** — use `/team` to check session cost8687---88> Converted and distributed by [TomeVault](https://tomevault.io/claim/coctostan) — claim your Tome and manage your conversions.89<!-- tomevault:4.0:skill_md:2026-04-13 -->