ux-assessmentlisted

# UX assessment — the two-role walkthrough ## What this is and why HexGraph keeps shipping UI interactions that don't behave the way the design intends, caught only when the maintainer manually clicks around. This skill replaces that with a repeatable, agent-driven, role-separated walkthrough: - **Role 1 — the VR analyst** drives HexGraph the way an agent is *meant* to (the MCP server in driver mode, and/or `hexgraph run` / the CLI), following a deliberate ordered sequence that **populates every surface** the researcher will touch. The goal: when Role 2 opens the UI, **nothing is empty** — every panel, tab, modal, and badge has the state it needs. - **Role 2 — the simulated researcher** is a *separate* agent that opens the UI cold (Playwright) on the analyst's project and uses it as a first-time human researcher would, walking `docs/dev/ux-contract.md` **entry by entry**, performing each step, **scoring every dimension** (functional + the qualitative axes), **verifying the backend effect** (the screenshot is evidence, not the check), and **narrating the experience** like a newcomer. The contract (`docs/dev/ux-contract.md`) is the spec; this skill is how it's exercised. The contract documents *intended* behavior — so when the implementation diverges, Role 2 *catches* it as a deviation. That is the point. **This is a living loop.** Re-run it on every major UI change, every fix evaluation, and before a release. The self-audit step (Role 2 flags entries that no