Research wiki motion skill

Research Wiki Motion

Use this skill for motion work in the research-wiki vault UI, especially:

• Reels/Capture Mode (?capture=1)
• semantic graph animation in apps/vault-ui/frontend/src/components/GraphView.tsx
• HUD/caption motion in apps/vault-ui/frontend/src/styles.css
• short-form video demos where motion must attract attention without making the research tool feel gimmicky

Core Rule

Motion must explain the graph. It must not decorate the screen.

Allowed motion attaches to one of these semantic surfaces:

• node: current concept spotlight, selected paper focus, degree/cluster emphasis
• link: manual link flow, similar-edge ignition, neighborhood bloom
• camera: restrained zoom-to-fit, macro-to-micro focus, no aggressive pans
• HUD: subtle metric/caption pulse, progress state, no full-screen effects

If a motion cannot answer "what research relationship does this reveal?", do not add it.

Hard Bans

Do not add:

• full-screen scan/sweep overlays
• glitch effects
• random decorative particles
• background flashes
• bokeh/orb decoration
• global node blinking
• fast camera whip/pan/zoom
• animations that affect normal graph mode when captureMode is false
• unverified "looks good" claims without screenshots

Direction Choice

When choosing among two or more motion concepts, use anneal-style scoring before implementing deeply.

Draft 2-3 candidates and score each on the same fixture, normally:

Question	Scale
First-2-second hook: is visible change obvious in a silent vertical clip?	0/1/2
Research value: does the viewer understand what the product helps them do?	0/1/2
Semantic attachment: is motion bound to nodes, links, graph scope, or query flow?	0/1/2
Restraint: does the graph remain readable without screen-level decoration?	0/1/2
Capture stability: do HUD/caption/graph avoid overlap at mobile and 9:16 sizes?	0/1/2
Runtime health: console, page errors, failed requests, and bad responses are empty	pass/fail

Pick by the score. Use taste only to break a near tie.

Preferred Patterns

Use these before inventing new effects:

• Concept Spotlight: one or two soft rings from the current concept node.
• Link Bloom: particles along links touching the current node during local expansion.
• Similar Edge Ignite: brighter particles only on kind === "similar" edges.
• Core Collapse: fade/filter low-degree satellites, keep backbone readable.
• Macro-to-Micro Focus: restrained zoom/focus from global cluster map to one concept.
• Workflow Narrative: search/select concept -> local neighborhood -> similar links.

Keep durations slow enough for a phone recording: no critical phase under 2.5s.

Implementation Guardrails

Before editing functions/classes, follow the repo's GitNexus impact rule.

For GraphView.tsx:

• Prefer canvas-local helpers such as drawCaptureSpotlight() and captureLinkParticleCount().
• Keep motion behind captureMode or a capture-step condition.
• Preserve existing drag-end pinning behavior: only user drag should set fx/fy.
• Do not auto-pin nodes just to make a shot stable.
• Avoid adding animation dependencies unless DOM/SVG timeline work clearly needs one.

For styles.css:

• Keep capture layout fullscreen and quiet.
• HUD/caption may pulse subtly.
• Never reintroduce .capture-graph::before screen sweeps or captureSweep.
• Ensure hidden controls also have pointer-events: none in capture mode.

For App.tsx:

• Keep ?capture=1 shareable and reproducible.
• Provide an exit path (Exit and/or Esc) if capture mode is user-entered.
• Do not make capture mode the default app experience.

Verification Checklist

Run the smallest applicable set, then broaden if the change is shared:

cd apps/vault-ui/frontend
npm run test:graph-physics
npm run build
npm run test:smoke

For runtime visual checks, start the app and capture at least:

• normal graph mode: controls visible, capture styles absent
• ?capture=1, 9:16 viewport: global/local/similar/final stages
• mobile-width viewport if adding controls or HUD changes

The browser check must record:

• consoleErrors: []
• pageErrors: []
• failedRequests: []
• badResponses: []
• canvas is nonblank
• no text/HUD/caption overlap
• no full-screen scan/sweep animation

If the change touches backend static serving or repo-wide code, also run:

uv run pytest
npx gitnexus detect-changes
git diff --check

Review Stance

Findings first. Treat visual success as evidence-backed, not assumed.

Report:

• screenshots or paths to captured PNGs
• exact commands run
• whether ordinary graph mode still behaves normally
• remaining visual risk, especially if final judgment needs human review