GanyuanRan/Aegis v2.12.0

v2.12.0: UI/UX governance on demand

on GitHub

UI/UX governance on demand

• Add ui-ux-governance, a compositional rule skill for user-facing interfaces,
  interaction flows, and API changes that affect visible states or recovery.
• Prefer lower total user effort under equivalent goals and accepted
  requirements: understanding, decisions, operations, waiting, and recovery.
  Preserve accessibility, informed control, and necessary safeguards.
• Load only applicable experience and verification sections. Reuse the
  project's design system and accepted references; keep tiny wording changes
  scoped and exclude backend changes with no user-facing effect.
• Carry applicable criteria through design, planning, execution, debugging,
  review, and verification. Existing workflows retain their decision and
  completion roles; the former inline UI checklist points to the shared owner.

Evidence and upgrading

• Add positive, lifecycle, API-recovery, tiny-change, backend-negative,
  explicit-mode, missing-evidence, and safeguard-pressure scenario contracts,
  plus natural and explicit Codex skill-loading smoke prompts.
• Keep missing browser/device and usability evidence explicit. Bounded skill
  observations and loading checks do not establish measured product usability
  or a new benchmark score. Other hosts retain their documented evidence scope.
• The offline benchmark policy fixture has a limited target mapping and fails
  with Windows Python's AMD64 identifier. Windows validation can run the
  unchanged checks in an isolated Linux container; this verification path does
  not change host installation requirements.
• Synchronize all eight package/plugin version fields at 2.12.0. Update via
  the existing host installation route and restart or reload as that host
  requires. Both manual global routing-prefix mirrors remain compatible;
  no manual re-copy is needed for this release.

Validation scope

• The original tests/e2e/run-all.sh --full --host-profile fast passed in
  isolated Linux: Layer 1 43/0, completion contract, and Layer 2/3 6/0 each.
• Natural and explicit UI/UX Codex smokes passed with successful command results
  and emitted skill bodies independently checked, beyond trigger-parser PASS.
• An accepted-design application check in actual Edge headless passed 30 checks
  at 1280, 390, and 360px using local HTTP rejection/success fixtures. It covers
  input preservation, pending duplicate suppression, keyboard retry, and layout.
• Separate fresh tiny, backend-only, and explicit-mode cases preserved scoped
  edits. The tiny case still read the whole short verification reference;
  section-selective loading remains advisory model discipline.

These are representative method-behavior observations, not measured usability,
physical-device or screen-reader verification, or a new benchmark result.