product-design-harness

# UX3 Product Design Harness - Expanded AI Context

> UX3 is an installable Agentic Design Process for AI-agent teams. It checks
> product direction across User Flow, Evidence Flow, and Business Flow while
> keeping product judgment Designer-in-the-Loop.

Last reviewed: 2026-07-11

Canonical website: https://cis2042.github.io/product-design-harness/

Source repository: https://github.com/cis2042/product-design-harness

## Core Facts

- UX3 is a product decision harness, not a coding harness or UI style library.
- Agentic Design Process means agents participate from product framing through
  research, ideation, build, launch, observation, and feedback.
- Designer-in-the-Loop means designers retain accountable judgment over taste,
  problem framing, trade-offs, ethics, meaning, and uncertainty.
- It is distributed as an installable Skill with prompts, reviewer agents,
  JSON Schemas, examples, and deterministic validators.
- It returns one verdict: continue, verify, or stop_reframe.
- The combined verdict equals the worst lane verdict. Scores are not averaged.
- The weakest flow controls the next action.
- Taste, strategy, ethics, meaning, risk appetite, and final accountability
  remain human-owned.

## Why This Process Exists Now

AI agents can move from an idea to working software before a team has tested
the product direction. Faster execution increases the cost of weak framing,
unsupported claims, and invisible trade-offs. UX3 lets agents expand research,
options, and bounded execution while a designer remains accountable for the
judgment that machines cannot own.

## Three Flows

- User Flow identifies target populations, affected people, situations, tasks,
  workarounds, and desired outcomes.
- Evidence Flow tests claims through sources, signals, counter-signals,
  provenance, freshness, subjective experience, and uncertainty.
- Business Flow identifies the company or project owner, stakeholders, value
  exchange, payment, cost, risk, and durability.

## Decision Kernel

1. Validate lane contracts and evidence references.
2. Reduce to the worst verdict and deterministic weakest flow.
3. Gate uncertainty, risk, reversibility, and human-owned calls.
4. Emit one verdict, next action, execution boundary, and stop conditions.

## Review Modes

- quick_gate: small, reversible, bounded changes.
- standard_gate: new features, workflows, or experiments.
- ux3_council: high uncertainty, low reversibility, meaningful risk, or
  human-owned trade-offs.

## Evidence Boundaries

Evidence tiers range from t0 internal belief to t4 durable retention, workflow
dependency, and sound economics. An evidence tier limits the maximum responsible
action. A strong business case cannot outvote missing user or evidence support.

## Install

`npx skills add cis2042/product-design-harness -g -y`

After installation, ask the agent:

`Use UX3 Product Design Harness. Review this product decision before implementation: <decision>.`

## Canonical Sources

- Canonical answers: answers.md
- Skill entry point: skills/product-design-harness/SKILL.md
- UX3 model: docs/UX3.md
- Full handbook: docs/HARNESS.md
- Contracts: docs/CONTRACTS.md
- Uncertainty: docs/UNCERTAINTY.md
- Human judgment principles: docs/PRINCIPLES.md
- Adoption: docs/ADOPTION.md
- Output schema: schemas/review-result.schema.json
- Validator: scripts/check_review.py

## Limitations

- This knowledge layer does not guarantee ranking, citation, or ingestion by
  any search engine or AI model.
- UX3 does not replace a product designer, accountable owner, or domain expert.
- UX3 is an installable Skill. This repository does not expose a hosted remote
  agent endpoint or claim A2A service compatibility.
- A continue verdict permits only the bounded action recorded in the review.