walkthrough-storyboard

Turns "make a demo of X" into a recorder-ready scenario: a schema: 1 YAML of timed beats — moves, clicks, zooms, cards — that the sibling walkthrough-record skill executes. It fixes the failure mode of tours that ramble: zoom everywhere, wall-of-text callouts, no narrative spine, brittle selectors.

When NOT to use

Executing/rendering/exporting the video, or recorder errors → walkthrough-record.
In-app onboarding features (intro.js-style tooltips), slide decks, or YouTube ad copywriting — this skill only produces recorder scenarios.

Workflow

Define the viewer's job in one sentence ("create the first dashboard", "invite a teammate"). Write the ending first: what success looks like on screen. One narrative level per video — overview, task tutorial, or deep workflow; mixing them is what makes demos ramble.
Walk the real app. Identify each beat's target element and collect stable selectors (data-testid > ids > structural CSS). If none exist, ask for them or propose adding them — brittle selectors are the #1 cause of dead scenarios. Note anything sensitive on screen → mask list; logged-in flows → note that the user should export a storageState auth file (kept out of git).
Storyboard the beats per references/storyboard-patterns.md: hook first (hero card ≤ 2.5 s, show the payoff early), then 3–6 chapters of zoom in → act → card → zoom out, then a result/CTA beat. Discipline that separates polished from frantic: at most one zoom per logical step, zoom scale 1.3–2.0 (2.2+ only for tiny controls), cards adjacent to — never covering — the action, title 3–8 words, body 8–20 words, card up 1.2–4 s.
Write the YAML against references/scenario-schema.md (pinned to schema: 1). Prefer waits on selectors over sleeps; let move durations auto-compute from distance; add an assert after state-changing actions.
Pick the look per references/theming.md: default coral-glass, indigo, paper (for dark apps), or an inline theme: block. Card contrast ≥ 4.5:1. Then decide the frame: output.frame puts the capture on a wallpaper canvas with rounded corners and a shadow (background: auto derives it from the app; chrome: true adds a macOS window header). Full-bleed stays the default — frame marketing and landing-page videos, leave in-app help and anything that will be cropped unframed.
Validate and hand off. If the recorder is installed, lint the YAML with its browser-free mode: node <walkthrough-record>/scripts/record.mjs scenario.yaml --validate-only (exits 0 on a clean schema, never launches Chromium). Then hand the scenario to walkthrough-record to record; iterate on pacing from what the frames show, not from imagination.

Output spec

A single scenario YAML (plus optional theme file) that:

passes the recorder's pre-flight validator (schema: 1, known keys only)
runs 30–90 s for marketing/tour content (3–7 min only for training content)
has a hook in the first 3 s, ≤ 1 zoom per step, card copy within limits, masks for anything sensitive, and stable selectors throughout

Gotchas

Zoom/ring/card/mask targets are CSS selectors only (engine-side querySelector); Playwright-style text=/role selectors work only for move/click/type. Don't mix them up.
Don't script mid-zoom scrolling or interactions that open viewport-positioned portals during a zoom — sequence them zoom → settle → act.
wait: {until: networkidle} hangs on websocket-heavy apps; wait on a selector.
Respect the defaults instead of hand-timing every step — preClick/postClick pauses and distance-based move durations already encode good pacing; override only for deliberate slow reveals.
Keep GIF ambitions out of the storyboard: if the deliverable includes a README gif, mark ONE 5–10 s beat as the gif segment for the record skill to trim to.

Files

references/scenario-schema.md — the full schema-1 authoring reference (steps, top-level keys, defaults)
references/storyboard-patterns.md — beat templates, zoom/cursor/card craft rules with sources
references/theming.md — theme keys, presets, contrast rules