Skills subagent-driven-development

📦

subagent-driven-development

Name: subagent-driven-development
Author: ZhanlinCui

Safe

Execute Plans with Subagents and Two-Stage Review

Also available from: Cycleaddict,Cygnusfear,davila7,Doyajin174,obra,DYAI2025,CodingCossack

AI agents often struggle with complex multi-task implementations. This skill solves that by dispatching fresh subagents for each task with mandatory spec compliance and code quality reviews.

Supports: Claude Codex Code(CC)

🥉 73 Bronze

Download the skill ZIP

Upload in Claude

Go to Settings → Capabilities → Skills → Upload skill

Toggle on and start using

Test it

Using "subagent-driven-development". Task: Add input validation to form component

Expected outcome:

Implementer reports: Added validation for email, phone, and required fields. 12/12 tests passing. Self-review: Added error messages matching design system. Spec reviewer: Approved - all validation rules implemented. Code reviewer: Approved - clean implementation with proper error handling.

Using "subagent-driven-development". Task: Create database migration script

Expected outcome:

Implementer reports: Created migration adding users table with indexes. Rollback script included. Self-review: Followed existing migration patterns. Spec reviewer: Approved. Code reviewer: Approved - migration is idempotent and well-documented.

Security Audit

Safe

v1 • 2/24/2026

Static analyzer flagged 38 patterns but all are false positives. DOT diagram syntax (backticks), markdown formatting (crypto patterns), and self-review checklists (reconnaissance patterns) were incorrectly identified. No actual security risks exist - this is a legitimate workflow orchestration skill.

Files scanned

403

Lines analyzed

findings

Total audits

No security issues found

Audited by: claude

Quality Score

Architecture

100

Maintainability

Content

Community

100

Security

Spec Compliance

What You Can Build

Feature implementation from specification

Execute a multi-feature implementation plan where each feature can be developed independently

Refactoring large codebases

Break down refactoring into discrete tasks with quality gates to ensure correctness

Test-driven development at scale

Implement multiple components with TDD, ensuring each passes tests before moving forward

Try These Prompts

Basic: Single Task Implementation

I'm using Subagent-Driven Development. Task 1: Create a function to validate email addresses. Context: This will be used in the user registration flow. Please implement with tests.

Intermediate: Multi-Task Feature

I'm using Subagent-Driven Development to implement user authentication. Task 1: Create login endpoint with password validation. Task 2: Add session management. Task 3: Implement logout with token invalidation. Each task includes full text from my plan document.

Advanced: Full Plan Execution

I'm executing a plan using Subagent-Driven Development. I've created a TodoWrite with 8 tasks extracted from docs/plans/api-v2-migration.md. Starting Task 1: Migrate user endpoints to v2 format. Here is the full task text and context about the migration strategy.

Expert: With Custom Review Criteria

Using Subagent-Driven Development for Task 3: Implement rate limiting. Custom review criteria: Must handle distributed environments, support configurable limits per endpoint, include metrics collection. Spec reviewer should verify these criteria. Code reviewer should check Redis integration patterns.

Best Practices

Always provide full task text and context upfront - do not make subagents read plan files
Answer all clarifying questions before allowing implementation to begin
Enforce the review order: spec compliance must pass before code quality review starts

Avoid

Skipping either review stage or proceeding with unfixed issues
Dispatching multiple implementation subagents in parallel which can cause conflicts
Accepting self-review as a substitute for independent review

Frequently Asked Questions

When should I use this versus parallel session execution?

Use this skill when you want to stay in the same session with continuous progress. Use parallel session execution when tasks can run completely independently without any coordination.

What happens if the spec reviewer finds issues?

The implementer subagent fixes the issues, then the spec reviewer reviews again. This loop continues until the spec is fully compliant before code quality review begins.

Can subagents ask questions during implementation?

Yes. Subagents are encouraged to ask questions at any point if they encounter something unclear. Answering questions before proceeding prevents incorrect implementations.

How many reviewers are involved per task?

Three: the implementer does self-review, then a spec compliance reviewer checks requirements, then a code quality reviewer checks implementation quality.

What if a subagent fails to complete a task?

Dispatch a fix subagent with specific instructions about what needs to be corrected. Do not try to fix manually as this causes context pollution.

Is this skill compatible with all AI assistants?

Yes, it works with Claude, Codex, and Claude Code. The subagent dispatch mechanism uses the Task tool available in these platforms.