feat-harness

Set up a Planner-Generator-Evaluator agent harness for feature development.

Updated Mar 21, 2026
One-click install
npx skills add https://github.com/Derek-X-Wang/skills --skill feat-harness
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: feat-harness
Source: https://github.com/Derek-X-Wang/skills/tree/main/feat-harness
Command: npx skills add https://github.com/Derek-X-Wang/skills --skill feat-harness

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Provides a product-level orchestration pattern that coordinates four specialized agents (Planner, Generator, Evaluator, and Auditor) to plan, build, evaluate, and audit features with clear traceability and reusable templates.

Core Features & Use Cases

  • Four autonomous agents with file-based communication so team members can review outputs and audit history.
  • Per-project harness setup that bootstraps agent configs and evaluation criteria into <project>/.claude/harness/, including templates from references/agents/ and references/criteria/.
  • End-to-end feature development workflow: spec planning, feature implementation, evaluation, and optional external audit, with iterative feedback loops.

Quick Start

Bootstrapping a harness creates the project-specific harness directory and agent configurations so you can begin feature work immediately.

Frequently Asked Questions about feat-harness

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

FAQPage Schema
How do I orchestrate multi-agent workflows for feature development?

Multi-agent orchestration for feature development is handled by a Planner-Generator-Evaluator harness that coordinates specialized agents via file-based communication to preserve an audit trail and enable iterative improvement.

What is a planner-generator-evaluator agent harness used for?

A planner-generator-evaluator agent harness is used to orchestrate feature development by assigning structured roles for spec planning, implementation, evaluation, and optional auditing with clear traceability and reusable templates.

How do I set up a multi-agent harness in my project directory?

To set up a multi-agent harness, bootstrap the project-specific configuration which writes agent configs and evaluation criteria into the project's .claude/harness directory using templates loaded from references.

Can I use file-based communication to preserve an audit trail between autonomous agents?

File-based communication between autonomous agents is supported to preserve an audit trail, allowing team members to review outputs and audit history throughout the iterative feature development workflow.

Does the agent harness support iterative feedback loops for feature implementation?

Iterative feedback loops for feature implementation are supported by the Evaluator and optional Auditor agents, which apply loaded evaluation criteria and rubrics to surface specs and feedback through a standardized directory structure.

What are the limitations of using a four-agent harness for feature planning?

Limitations include the requirement to bootstrap project-specific configs into a dedicated harness directory and reliance on file-based communication, which may introduce overhead for smaller projects not needing full audit traceability.