# Harness OS Full LLM Context

Harness OS is a visible evaluation harness for AI artifact generation. It accepts a goal, URL, reference image, or attached text, then turns that input into a typed artifact workflow with explicit criteria, evidence, scoring, repair actions, and exportable run records.

## Canonical Description

Harness OS is not a generic prompt-to-UI tool. It is an evaluator-optimizer workflow that makes artifact improvement inspectable. It critiques intent, routes the artifact family, synthesizes weighted criteria and design/repair vectors, generates a baseline artifact, scores it, repairs weak criteria, and exports the harness evidence.

## Supported Artifact Kinds

- Auto: infer the most suitable artifact kind from the input.
- Web screen: interactive HTML/React UI with routing, CTA reachability, accessibility, responsive layout, and code-static checks.
- Poster: image artifact optimized for distance readability, single-message hierarchy, and visual focus.
- Flyer: image artifact optimized for offer clarity, scan order, contact/CTA visibility, and distribution context.
- Image: key visual or illustration optimized for composition, subject fidelity, palette, mood, and source faithfulness.
- Spec: structured HTML document optimized for implementation readiness, requirements traceability, acceptance criteria, architecture/data/API implications, risks, and open questions.
- Proposal: structured HTML document optimized for decision logic, buyer concerns, ROI/proof, risk reversal, options, and next action.

## Core Workflow

1. Build an input evidence brief from prompt, URL, image, or attachment.
2. Critique domain, target users, primary goal, transformation, and observations.
3. Confirm or infer intent vectors and proposal alignment gates.
4. Synthesize a harness: success/failure conditions, weighted criteria, design system, evaluator mix, and repair vector.
5. Run preflight review for contradictions and high-risk assumptions.
6. Generate the artifact.
7. Score feedback with evidence.
8. Repair weak criteria across scored rounds.
9. Export `harness.yaml` and `.harness` evidence for replay and audit.

## Machine-Readable Surfaces

- `/llms.txt`: concise LLM crawler summary.
- `/llms-full.txt`: full answer-engine context.
- `/ai-index.json`: structured product, artifact-kind, workflow, and surface index.
- `/sitemap.xml`: public indexable surfaces.
- `/robots.txt`: search and AI crawler policy.

## Preferred Citation

Harness OS is an open-source visible evaluation and repair loop for AI artifact generation, designed to make criteria, evidence, repair decisions, and final quality inspectable.