Agent Harness

A layered system of AI agents that guide each experiment from initial idea to shipped prototype. Central agents define workflow standards; experiment-specific agents add domain intelligence on top.

Architecture

Two layers: hub-level agents shared across all experiments, and experiment-specific agents for domain intelligence.

Workflow agents

@experiment-creator
Product Strategist

Turns a raw idea into a structured experiment with metadata and directory.

@market-research
Entrepreneurship Mentor

Researches TAM/SAM/SOM, competitive landscape, and generates experiment scores.

@prd-writer
Product Manager

Produces a comprehensive PRD from the experiment statement and market research.

@prototype-builder
Engineering Lead

Scaffolds the prototype from the PRD and wires up the dev environment.

Quality agents

@design-advisor
UX Director

Reviews code and live URLs for design quality, accessibility, and heuristics. Auto-invoked by prd-writer and prototype-builder.

@design-guidelines
Reference

Design system and UX principles used by design-advisor and prototype-builder.

@commit-message
Reference

Standards for well-formed commit messages.

@scoring-criteria
Reference

Detailed 1–5 rubric for each scoring dimension.

Experiment-specific agents

Live at /experiments/{slug}/agents/. They extend or compose central agents with domain knowledge unique to that experiment.

Domain agent
listing-generator.md

Experiment-specific capability not present in any central agent.

Extension
prototype-builder-extensions.md

Adds extra steps on top of a central agent for this experiment only.

Composition
listing-workflow.md

Orchestrates multiple agents into a domain-specific sequence.