wayfinder: Skill Manager UX/UI redesign map

[intelligentrascal]

Destination

An approved, implementation-ready UX/UI redesign package for Skill Manager: a unified mental model and information architecture, visual system, interaction and action model, responsive behavior, consistent content language, a user-reviewed prototype, and a sequenced implementation ticket set - while preserving the existing evidence, provenance, verification, and safe-change contracts.

Notes

• Primary user: the human fleet operator maintaining skills across Pi, Claude, OpenCode, Codex, shared locations, and the canonical repository.
• This is a planning map, not an implementation branch. The redesign is not implemented until the route is clear and the handoff ticket set is agreed.
• Preserve the current contracts for evidence levels, provenance and canonical origins, variant states, review/apply safety, cached GitHub facts, and responsive workspace behavior.
• This map is independent of the earlier functional work and must not block those tickets.
• Sessions should consult: grilling, domain-modeling, research, prototype, hallmark, design-taste-frontend, design-critique, design-handoff, and reference-driven design.
• The prototype reference shelf includes Mobbin AXI, 21st.dev, Land-book, Magic UI, and the relevant curated references in Outline's AuxUI Resources and OpenCode Resources documents. Reference pulls must use actual visual inspection and cite the source used.

Decisions so far

• research: Benchmark five skill-management interface types - Five-category primary-source benchmark completed; strongest reusable patterns are reasoned drift states, composable attention filters, explicit preview-before-apply, last-good-state preservation, named attention categories, and visible provenance.
• research: Map direct Skill Manager competitors and near-peers - Ten direct competitors and one adjacent near-peer mapped; the strongest differentiation opportunity is making Skill Manager's repo-canonical provenance and verified-apply safety legible against central-library and disposition-manager products.

Not yet specified

• The default opening surface and navigation order: attention-first, browse/catalog-first, or skill-workspace-first.
• The canonical user-facing mental model and glossary for skills, copies, origins, canonical content, variants, evidence, operations, attention items, and workspaces.
• The revised visual direction beyond the existing dark Genome language: reference set, typography, density, hierarchy, motion, and visual treatment of evidence.
• The action model and content language for sync, origin assignment, upstream checks, adaptation review, verified apply, rollback, and uncertainty.
• Responsive and accessibility rules for desktop, tablet, mobile, keyboard navigation, focus, loading, empty, error, and destructive states.
• The implementation boundary and sequencing needed to preserve functional behavior while replacing the presentation layer.

Out of scope

• New backend capabilities or changes to skill-runtime semantics.
• Rewriting evidence sources, provenance rules, variant verification, or safe-apply behavior.
• Implementing the redesign during this wayfinder pass.

[intelligentrascal]

Charted the redesign as a wayfinder map rather than an implementation ticket.

What changed:

• Rewrote this issue as the canonical map with the agreed destination: an implementation-ready UX/UI redesign package and sequenced implementation ticket set, while preserving the existing evidence, provenance, verification, and safe-change contracts.
• Recorded the human fleet operator as the primary user.
• Added three child decision tickets:
  • research: Benchmark five skill-management interface types
  • prototype: Compare candidate opening surfaces
  • grilling: Define the redesigned mental model and navigation
• The research ticket requires at least five distinct categories: plugin/package managers, prompt/skill/agent registries, dotfile/configuration managers, developer portals/service catalogs, and agent/automation workflow managers.
• The prototype ticket exists specifically to resolve the still-open opening-surface decision with attention-first, browse-first, and skill-workspace-first shells.
• The mental-model ticket is natively blocked by both the research and prototype tickets, so navigation is not decided from intuition alone.

The AFK research ticket has been dispatched to a worker. The prototype and grilling tickets remain human-in-the-loop and intentionally unresolved. No production code was changed in this charting pass.

[intelligentrascal]

Added the reference-driven design pass to the prototype route.

The prototype ticket now owns the visual investigation before shell comparison. It explicitly requires Mobbin AXI with actual screenshot inspection, plus 21st.dev, Land-book, Magic UI, and the relevant references already catalogued in Outline. The map Notes now point future sessions to that reference shelf and require source citation.

The separate five-category research ticket remains focused on comparable product-management mental models and interaction patterns. It is not being overloaded with visual-inspiration work.

[intelligentrascal]

The research child ticket is now resolved.

Its cited benchmark is committed and pushed in acc8538:
docs/research/2026-08-17-benchmark-five-skill-management-interface-types.md

The map's Decisions so far now records the decision at one-line resolution. The next frontier remains the human-in-the-loop prototype, which must use the benchmark plus Mobbin AXI, 21st.dev, Land-book, Magic UI, and the Outline reference shelf before comparing the three opening surfaces.

[intelligentrascal]

The direct-competitor research is now resolved.

The durable artifact is committed and pushed in dc98c0c:
docs/research/2026-08-17-direct-skill-manager-competitors.md

The scan covers 10 true direct competitors and one adjacent near-peer, including the peer supplied by the user. The map now records the direct-competitive decision separately from the earlier cross-category benchmark.

The prototype is now the next frontier and is unblocked. It must use both research notes, then perform the Mobbin AXI and design-reference pass before comparing the three opening surfaces.