[kyleve]

Posted by an AI agent on kve's behalf.

Summary

Add a session-only Mobius Astro prototype for non-linear PR review. One logical workspace per PR owns independently navigable Review, Source, Research and explicitly simulated Preview children, its arrangement/focus, private draft, per-file reviewed versions and separate skip receipts.

• Close outer PR tabs/windows and reopen them through Workspaces with the same child identities and context. Inactive closure keeps the active PR/logical child; active closure selects a remaining neighbor; last closure shows explicit recovery. A resumed Review starts its selected file at the top. Reset restores the sample session.
• Present PR workspaces → Tabs / Windows independently from Views in the named PR → Tile / Float / One view. Float works at every width and Tile stacks when narrow. Drag a child title/filename/icon, resize, or use keyboard movement; stationary filename clicks retain the chooser.
• Group Code, Documents, Localization and Snapshots/assets with quiet groups collapsed and distinct remaining/reviewed/skipped counts. Prepared rich comparisons retain raw source and honest unavailable-content fallbacks. LFS is an attribute. Explicit PR-local skip/undo never advances reviewed baselines; changed/new work returns while unchanged skips persist.
• Author through a real Astro route with imported CSS and bundled client JavaScript, using root-pinned dependencies. The documented native background server binds to the LAN for phone access. CSS hot updates preserve the sample session; page/script changes automatically reload and reset it. The preceding standalone artifact and all historical evidence remain archived.

Run the demo and inspect its sources. The absolute --root is required for the pinned CLI's background child. No second dependency tree, root package/config change or production deployment is added. The reusable planning workflow is a separate PR #503.

Two synthetic PRs/two revisions and prepared comparisons keep the study bounded. No real GitHub PR closure, deletion, repository/account data, native OS windows, Git/LFS/parsers, code editing/execution, persistence or sync is implemented. Cross-device open/closed policy remains under Q-09.

Why

Review work needs durable logical ownership independent of its current presentation. The experiment makes workspace return, multiple contextual views, per-file comparison history and explicit skipped work concrete before native state design. Astro supplies the requested source-update loop and LAN access without changing those interactions.

Trigger	Result
Close and reopen a PR workspace	Keep child IDs/routes, frames/layout, logical focus, receipts and draft; review resumes at its file top.
All workspaces closed, resized or pushed	Stay explicitly empty until reopen; a sample push changes the retained revision without opening a workspace.
View full context or skip then receive new changes	Preserve the actual reviewed basis; skip is separate and content-version scoped.
Preview lacks content or omits source details	Expose available source/pointer/metadata honestly; unavailable binary bytes cannot be marked reviewed.
Edit CSS versus page/client source	CSS updates preserve the sample; a full reload deliberately resets it.

Screenshots

Live Astro/Chromium captures at eba910265bc74b8ef858c1d71289ce31745b0132, light appearance, 1× scale, synthetic PR42 v1. Phone starts in One view; tablet/desktop in Tile. All three initial PNGs are byte-identical to matched standalone captures.

Phone · 390 × 844	Tablet · 834 × 1112

Matched before/after and resumed-review captures retain exact hashes. Close/reopen, title-bar, layout and type/skip evidence remain historical in the same walkthrough.

Review focus

The current delivery seam is index.astro plus its imported sources and prototype-local config. The client script is byte-identical to the archived inline script; CSS differs only by removing its closing-block separator blank line. The dev server uses native lifecycle commands, ignored local output/cache and prototype-scoped file access. An outside-root @fs request is denied. No runtime test-state API was added when the script became module-scoped.

Fresh independent review found no actionable migration issues. Ten Chromium/WebKit dev/build cases pass, with a final equivalence recheck of the CSS-only whitespace correction. The author and coordinator reran their served checks on final sources. WebKit mouse closure of an inactive workspace after typing in Research may move DOM focus to body while retaining logical child focus; this also occurs in the archived artifact. Keyboard recovery/context pass, so this is an unchanged browser-focus limit.

Earlier reviews corrected floating order, focus/recovery, search Enter handling, skip accessible naming, pointer/object-size labeling and narrow metadata readability; their linked execution records preserve findings and rechecks. User evaluation remains pending; 01a stays executing and STU-330 In Review. Mockup layouts/renderers/schema are hypotheses, not native architecture acceptance.

Testing

• Exact-source migration verification pins all five sources/configs, commands and nine captures. Astro 7.4 / Node 24.22 check: four files, zero errors/warnings/hints; build: one static route with bundled assets. Explicit route/config ESLint and client syntax pass. Root default lint does not select Prototype.
• Live edit/revert probes prove CSS HMR and page/script full reload without manual refresh, with expected session semantics and no duplicate Add handler. Machine-LAN HTTP and native background status/logs pass; temporary build-preview server stopped. The requested dev server remains running with ownership/stop instructions in the handoff.
• Eight observable scenarios × three viewports pass, plus six selected WCAG scans. Cover review/skip/draft retention, independent source/search/preview children, close/reopen, push resurfacing and correct remaining/full basis, rich/source content, file-top return, exact filename drag/click, keyboard movement/resize. No page errors, document overflow or external data requests.
• Ten independent Chromium 156 / WebKit 27.2 dev/build cases and four final coordinator localhost/LAN cases pass. Playwright 1.64.1 / axe 4.14.0; actual phone Wi-Fi, native devices, VoiceOver and full accessibility are untested.
• Nine captures visually inspected; three initial before/after pairs byte-identical. Archive retention, 332 local links/anchors, Markdown fences and staged whitespace pass. Prior source/evidence remain intact; runtime probes are ad hoc evidence, not CI guards.
• Lifecycle, title-bar, layout, and type/skip matrices retain historical scope and limits. Two user-facing revisions do not claim a third live push; later-head baseline selection has separate seeded policy evidence. Native tests/snapshots/SwiftFormat cannot exercise this browser tooling change.

Stack and tracking

Second PR in stack #501, above planning PR #497; parent remains ecb3c017d. This slice owns the browser experiment, 01a execution record, retained evidence and delegated execution checkpoint. STU-330 remains In Review under STU-329. Publication does not complete evaluation or authorize native implementation.

[kyleve]

Posted by an AI agent on kve's behalf.

Mobius workspace study — Astro development route, desktop 1440 × 1000. Live Chromium capture, light,1×, synthetic PR 42 v1, artifact eba910265bc74b8ef858c1d71289ce31745b0132. The initial rendered screen is byte-identical to the pre-Astro standalone capture; the real route now participates in live source updates and LAN serving.

Phone 390 × 844, tablet 834 × 1112, and matched before/after plus incremental-review captures cover the other served sizes/states. CSS updates retain the sample session; page or script edits automatically reload/reset it. Run/stop instructions use root-pinned tools and the native background lifecycle with an absolute project root.

The screen's owning source moved from the original inline index.html to src/pages/index.astro, imported CSS and client script. This existing discussion is preserved; pre-astro.html retains the previous file's exact bytes. Verification records actual reload, three-size observable regression, fresh Chromium/WebKit dev/build review and final-source coordinator checks. No actionable migration findings remain. Actual phone Wi-Fi and native devices are untested; user evaluation stays pending. Earlier lifecycle/layout/type evidence remains historical.