serve: UX design phase — web-native redesign of the UI (no releases until complete)

[dyoung522]

Owner directive (2026-08-31): the shipped web UI is a too-literal TUI conversion — page-per-command, no intrapage actions (e.g. the status page lists games but you must navigate to another page to switch). Before any release:

1. A UX design phase: audit every page for "actions live where the data is shown", intrapage/cross-element links, and web-native flows; produce a design the owner approves (UI and UX).
2. Implementation on a long-lived feature branch off v2 until the web UI is feature complete and the owner signs off.
3. No releases until further notice — the v2.1.0 prep is parked on dyoung522/v2.1-release (unmerged, untagged; tag-gate green) for whenever a release is eventually cut. v2.0.0's draft stays unpublished; the public still sees v1.30.1.

Part of EPIC #276.

[dyoung522]

Interaction model decided (owner): enhanced server-rendered — WEBUI.md's server-rendered foundation stays, every action renders inline where its data appears, and JS upgrades those to in-place partial updates over the existing endpoints; the no-JS full-page path remains the contract. First artifact: a page-by-page UX audit of the current UI (running now); then the proposed IA + per-page designs for owner approval; then implementation on a long-lived feature branch.

[dyoung522]

Direction sharpened (owner): the redesign is a SPA — most elements on a single page, modals for popups, separate whole pages only where a whole-page design earns it (e.g. search results); the TUI look and feel is discarded entirely. WEBUI.md is relaxed for this project (it targets public-facing apps): single-user localhost, modern browsers assumed, the CLI is the fallback — the no-JS requirement is dropped. Hard constraints that remain: no major security risks, no poor performance, no external dependencies users could struggle to install (everything stays embedded in the single binary). The existing /api/v1 + SSE layer becomes the SPA's backend.

[dyoung522]

Stack decided (owner): vendored Preact + htm — two pinned files in-repo, native ES modules, no npm/Node/bundler anywhere; go build remains the entire build and everything ships embedded in the binary.

[dyoung522]

Design doc committed: docs/plans/2026-08-31-serve-spa-design.md. All forks settled with the owner via mockups: Mission Control home + slide-over→full-page drill-in; omnibar search + search-page escape hatch; inline job progress + activity tray; Launcher style (system default, light/dark override); desktop ≥1080p only; vendored Preact+htm (no Node anywhere); context in the URL path; feature branch webui off v2. One scope decision deliberately left open for the doc review: whether the admin surface (game add, auth, sources, archive import — needs #307) is inside the feature-complete bar or stays CLI-only.

[dyoung522]

Scope decided (owner): full bidirectional parity — the web UI covers everything incl. all admin tools (pulls in #307), and web-first capabilities get CLI equivalents (#324/#254). Spec amended. Feature branch webui created off v2. Unit issues: #327 #328 #329 #330 #331 #332 #333 #334.

[dyoung522]

Ready for the owner's feature-complete UI+UX gate. webui is at c6ba2db (pushed to nerdwerx). All eight units plus a closing wave and a final polish wave are merged; the tracker for the SPA redesign is closed (#327–#334, #307, #324, #254, #335).

Verification. A whole-product epic live review (new-user drive from an empty config, full gates ×3, parity and engineering audits) found four Criticals; a closing wave answered all four (hydrate route fence, the toast rule, full bidirectional parity incl. purge / profile sync / mod re-link / pak conversion / every plan option / the source→game mapping, and a first-run path for custom-source users). A scoped epic re-review then verified every finding live and returned READY; the one remaining Important and the cheap minors landed in a final polish wave. Gates on the merged HEAD: make check green, uncached go test -race ./... green, serve E2E green three times, every ratchet (incl. new ones: one h1 per route, real section headings, accessible names on every input, landmarks, HTML boolean attributes, htm whitespace, entity literals, contrast, SPA/server constant drift), make man no diff.

To review: cd /home/dyoung/Projects/orca/workspaces/linux-mod-manager/webui-home && ./lmm serve (binary v2.0.0-376-gc6ba2db7). A 10-minute demo script — best started from an EMPTY XDG_CONFIG_HOME with a directory source — is in .superpowers/sdd/2026-08-31-webui-impl/epic-rereview.md under "Owner demo script".

Decisions parked for you (also in unit8-carry.md §Post-gate): (1) the library name column absorbs all the slack (vs. table at natural width); (2) motion durations 110/170/240 ms; (3) the light palette's amber darkened for AA (vs. a separate --accent-text token); (4) the conflict-overwrite affordance is inline rather than the design's modal (deliberate; confirm so the doc can be amended); (5) the full mod page and search page are 960 px readable columns on 1920 px (vs. the doc's "unlimited room").

Follow-ups filed, not blockers: #336 (per-hydrate full verify → incremental), #337 (NexusMods dropped tagNames, so search --tag / the tag filter fail against NexusMods; pre-existing).

After sign-off: merge webui → v2 (--no-ff), then the v2.1.0 release prep parked on dyoung522/v2.1-release (flip goreleaser draft:false at prep). No release until you sign off, per the standing directive.

[dyoung522]

Owner sign-off received 2026-09-09 ("the WebUI looks great"; three non-blocking notes filed as #338, #339, #340). webui merged into v2 as b1f4828 (--no-ff). The UX design phase and the feature-complete gate are done. Release versioning (v2.0.0 vs v2.1.0) is being settled separately; no release is cut yet.