Skip to content
Back to skills

Subagent Driven Development

ASecurity

Execute implementation plans by dispatching one subagent per task with structured review between each — fresh context, no pollution

  • 14 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 7, 2026
securitytestingcode-reviewsecurity

Security analysis

A100/100

Scanned September 7, 2026

npx -y skills add modbender/skill-library-mcp --skill subagent-driven-development --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Subagent Driven Development?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Subagent Driven Development
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/modbender-subagent-driven-development/badge)](https://www.skillsdirectory.com/skills/modbender-subagent-driven-development)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: subagent-driven-development
description: Execute implementation plans by dispatching one subagent per task with structured review between each — fresh context, no pollution
---

# Subagent-Driven Development — Spawn, Review, Repeat

## When to Use

Use when you have a written plan and need to execute it task by task using subagents.

## Core Principle

One subagent per task + structured review after each = high quality, no context pollution.

## Process

### For Each Task in the Plan:

#### 1. Spawn Subagent with Full Context

Provide the subagent with everything it needs upfront:

```
TASK: [Task title]

WHAT: [Exact description of what to build]

WHY: [Context — how this fits into the larger feature]

FILES:
- Create: [exact paths]
- Modify: [exact paths]

CODE:
[Complete code from the plan]

TESTING:
- Write failing test first (craftwork:test-driven-development)
- Run: [exact test command]
- Expected: PASS

CONSTRAINTS:
- [Project-specific constraints]
- [Security requirements]
- [Pattern compliance]

WHEN DONE:
1. Create MR/PR with descriptive title
2. Update task status to "review"
3. Report completion with verification evidence (craftwork:verification-before-completion)
```

#### 2. Review the Output

After subagent reports completion, review the diff using `craftwork:requesting-code-review`:

- [ ] Matches plan specification (nothing more, nothing less)
- [ ] Tests exist and cover the feature
- [ ] No hardcoded secrets
- [ ] No breaking changes
- [ ] Code is clean and follows existing patterns

#### 3. Approve or Request Fixes

**If approved:**
- Merge the MR/PR
- Mark task as done
- Move to next task

**If issues found:**
- Post specific feedback with file:line references
- Re-spawn subagent with fix instructions
- Do NOT skip the re-review after fixes

#### 4. Move to Next Task

Only after current task is merged and verified. Never skip ahead.

## Red Flags

- Never spawn multiple subagents on dependent tasks simultaneously
- Never skip review (even if subagent self-reviewed)
- Never proceed with unfixed issues
- Never merge without reading the diff
- Never let subagent read the plan file — provide full text instead

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…