Verified:GitHub

harness:subagent-driven-development

Dispatch isolated subagents to execute implementation plan tasks with spec and code quality reviews.

3|1|Updated Apr 5, 2026

Search Tags:#implementation planning #subagent #spec compliance #TDD workflow #code quality review #features.json tracing #session-based execution

Ready to run

npx skills add https://github.com/huangbaixun/harness-engineering --skill harness-subagent-driven-development

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: harness:subagent-driven-development
Source: https://github.com/huangbaixun/harness-engineering/tree/main/skills/subagent-driven-development
Command: npx skills add https://github.com/huangbaixun/harness-engineering --skill harness-subagent-driven-development

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps you reliably execute an implementation plan by delegating each task to fresh, isolated subagents and enforcing a two-stage review cycle so work stays spec-compliant and code-quality sound.

Core Features & Use Cases

Fresh subagent per task: Dispatches a dedicated implementer with no inherited session history to reduce context pollution and improve focus.
Two-stage review checkpoints: Runs spec compliance review first, then code quality review, with iterative fix-and-retry until approvals.
Continuous execution within a session: Minimizes human check-ins and proceeds task-by-task unless blocked by unresolved ambiguity or an actual BLOCKED status.
Safety guardrails via harness-delta: Requires reading harness-delta.md first and enforces features.json grounding to prevent orphan subagent work.
Role-based model sizing: Recommends using the least capable model that fits each role (implementer, reviewers, and higher-judgment tasks).

Quick Start

Use it when you have an implementation plan and want to execute its mostly independent tasks using isolated subagents with spec-first then quality-first reviews.

Frequently Asked Questions about harness:subagent-driven-development

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I execute an implementation plan using isolated subagents?▼

To execute an implementation plan with isolated subagents, dispatch a fresh dedicated implementer per task to reduce context pollution, followed by a two-stage spec compliance and code quality review cycle with iterative retries until approvals.

What is subagent-driven development for continuous task execution?▼

Subagent-driven development is a workflow that delegates implementation plan tasks to fresh isolated subagents, enforcing spec compliance and code quality reviews to achieve continuous task execution within a single session without context contamination.

How does the two-stage review cycle work for spec compliance and code quality?▼

The two-stage review cycle runs a spec compliance review first, then a code quality review, iterating with fix-and-retry logic until both stages approve the subagent's implementation work before proceeding.

When should I use subagent-driven development for implementation planning?▼

Use subagent-driven development for single-session development when implementation plan tasks are mostly independent, aiming for continuous task execution without context contamination from inherited session history.

Do I need features.json traceability before dispatching a subagent?▼

Yes, you must enforce features.json grounding for traceability or bug scoping before any subagent dispatch to prevent orphan subagent work and ensure implementation plan alignment.

What are the limitations of continuous session-based execution with subagents?▼

Continuous session-based execution with subagents is limited by unresolved ambiguity or actual BLOCKED statuses, which require human check-ins to prevent proceeding with flawed or incomplete implementation plan tasks.