v0.2.0 — Intent-first UX context

UX Skills 0.2.0 introduces an intent-first architecture for AI-assisted UX work.

setup-ux now supports both new and existing products. Greenfield projects use an adaptive conversation to capture product intent; existing projects inspect the repository, evidence, and design system first before resolving only material gaps.

The project context now begins with:

.ux/INTENT.md
.ux/CONTEXT.md
.ux/DESIGN-SYSTEM.md
.ux/DECISIONS.md

This release also introduces progressive context loading, progressive Skill references, visual architecture flows, migration from v0.1 context, example projects, and expanded setup evaluation coverage.

Intent tells the agent what good means. Context tells it what is true. Skills tell it how to reason.

The designer experience stays the same: install UX Skills, run setup-ux once, then work normally.

v0.2.1 — Intent-led outcome loops

UX Skills 0.2.1 adds lightweight long-horizon behavior without turning the project into an agent-orchestration framework.

The main change is a new shared Outcome behavior across every installed skill.

For substantial multi-step work, UX Skills now keeps product intent active throughout execution, prioritizes the highest-impact unresolved gap before polishing already-adequate work, and checks the actual experience against intent before declaring the work complete.

A finished checklist is no longer treated as proof that the problem is solved.

Intent is stable, not frozen

INTENT.md remains the durable north star, but it can evolve when explicit human direction or authoritative evidence genuinely changes or clarifies the intended outcome.

Routine implementation should not rewrite intent.

Agents should make the smallest useful refinement rather than regenerating the file, and implementation drift cannot retroactively redefine what the product was supposed to do.

Optional working state

Long-running work can now use an optional:

.ux/STATE.md

This is temporary working memory for things like:

current objective;
current phase;
meaningful completed work;
highest-impact unresolved gap;
blockers;
next action;
last intent check.

STATE.md is not created by setup-ux, is not a fifth core context file, and should disappear or stop being maintained when the work no longer needs it.

Model-agnostic long-horizon execution

UX Skills does not prescribe manager/implementer agents, dashboards, agent counts, /goal commands, or any specific orchestration approach.

Agents may use sub-agents or parallel execution when their environment supports it, while the designer continues to work normally.

Evaluation and validation

This release also adds evaluation coverage for:

small intent clarification versus full intent replacement;
implementation attempting to rewrite intent after the fact;
unnecessary creation of working-state artifacts;
long-running work where every planned task is complete but the intended experience is still not solved.

The shared skill validator now enforces the new Outcome behavior across the suite.

The designer experience remains the same:

Install UX Skills
→ run setup-ux once
→ work normally

UX Skills handles the additional execution discipline underneath.

v0.1.3 — User grounding without UX ceremony

Adds user grounding while keeping simple design work free of unnecessary research, personas, or discovery steps.

• added user-grounding to answer who the work is for, what is actually known about them, whether personas are useful, and what research is worth doing;
• added User to the shared behavior contract across every skill;
• added an explicit anti-ceremony guardrail so simple work does not trigger unnecessary research, personas, discovery, or long questionnaires;
• strengthened setup-ux to capture useful user/task evidence without manufacturing personas;
• added routing and authoring guidance for user grounding and non-activation on ordinary well-understood tasks.