Expand Artstr Stacks spec into a HyperCard behaviour model

Replace the thin single-action-per-layer model with real interactive
logic:

- Cards gain onEnter/onExit chains; layers gain an onClick chain. Each is
  an ordered action chain, not a single link.
- v1 action set: goto-card, go-back, open-url, show-layer, hide-layer,
  toggle-layer — reveals, secret panels, multi-step buttons, branching.
- New action-builder surface: a no-JSON panel with reorderable rows,
  id-based target pickers, and authoring warnings.
- Runtime tracks a per-card layer-visibility map (reset on re-entry) and
  executes chains; action overlay regions are focusable buttons.
- Safety: onEnter/onExit cannot navigate (no enter->goto->enter loops);
  chains capped and non-recursive; unknown action types ignored.
- Re-phased A (format) -> B (behaviour authoring) -> C (runtime) ->
  D (route) -> E (stack logic: variables/conditionals, future tier).

Interactivity stays a declarative, interpreted attribute of existing
layers — no new layer types, no HTML/JS, no eval.

Co-Authored-By: Claude Opus 4.7 <[email protected]>
This commit is contained in:
thePR0M3TH3ANandClaude Opus 4.7 committed 2026-05-22 07:05:12 -04:00
1 parent 19ee00f14a
commit 39f0ed0eb0
1 file changed
+318 -213
+318 -213
View File
@@ -4,63 +4,77 @@
Turn Artstr into a Nostr-native **HyperCard**: a user builds a **stack**
of fullscreen **cards**, each one an Artstr canvas (layers, drawings,
text, images, shapes, QR), and any layer can be made **interactive** —
click it to jump to another card or open a link. The whole stack
publishes as one self-contained Nostr event and plays in a fullscreen
**interactive viewer**.
text, images, shapes, QR), and any object can be made to **do
something** — a click runs an ordered **action chain** that navigates,
reveals, hides, or links. The whole stack publishes as one self-contained
Nostr event and plays in a fullscreen **interactive runtime**.
Think interactive zines, micro-sites, portfolios, album booklets, comics,
choose-your-path pages — all decentralized, forkable, no hosting.
Think interactive zines, micro-sites, portfolios, comics, choose-your-
path pages, simple games — decentralized, forkable, no hosting.
### The core realisation
**A Stack is a Slide Deck with interactive layers.** The Deck Builder
already gives us ~70% of this — a Stack is not a greenfield feature, it's
a focused extension of `templateMode: 'deck'`. What's genuinely new is
*layer actions* and an *interactive (non-linear) viewer*.
**A Stack is a Slide Deck whose cards have interactive logic.** The Deck
Builder already gives us the card container, sorter, edit round-trip,
embed-inline storage, publish/fork path, and a fullscreen renderer. Stacks
are a focused extension of that — not a greenfield feature. What's
genuinely new is the **behaviour system**: triggers, action chains, and a
runtime that executes them.
> A stack is a collection of visual cards where any object can do
> something — and "something" is an ordered list of actions, not a single
> link.
### Product goals
1. **Stack format** — a Template-tab layout: an ordered set of fullscreen
cards, each an Artstr canvas, at a stack-wide aspect ratio (16:9, 9:16,
1:1, 4:3, or custom — not locked to 16:9 like slide decks).
2. **Interactive layers** — any layer can carry an `action`: jump to a
card, or open a URL. Transparent **hotspot** layers make
click-anywhere regions trivial.
3. **Interactive viewer** — a fullscreen runtime that renders a card and
routes clicks on actionable layers. Non-linear: the stack decides where
a click goes, not just next/prev.
4. **Nostr-native** — publish/fork/edit-in-place as a `kind:30078`
2. **Interactivity** — a card has `onEnter` / `onExit` action chains; any
layer has an `onClick` action chain. Actions navigate between cards,
open links, and show / hide / toggle layers on the current card.
3. **Action builder** — a no-code panel: *"When this is clicked: 1. Hide
layer X 2. Show layer Y 3. Go to card Z."* It should feel like Apple
Shortcuts, not like writing code.
4. **Interactive runtime** — a fullscreen player that renders a card,
routes clicks through `onClick` chains, runs `onEnter`/`onExit`, and
tracks per-card layer visibility.
5. **Nostr-native** — publish / fork / edit-in-place as a `kind:30078`
addressable event, shareable by `naddr`, encryptable via the
private-publishing envelope.
### Non-goals (v1)
- **Arbitrary HTML / CSS / JavaScript.** Stacks are a *safe declarative
schema* Artstr renders — never executable web content. No iframes, no
script, no custom CSS. This is the single most important constraint.
- **Forms, inputs, databases, server logic** — a stack is a presentation
artifact, not an app.
- **Per-card transitions / animations** — future polish.
- **Blossom media upload** — v1 uses image URLs, like the rest of Artstr.
- **`zap`, `open-nostr`, `toggle-layer` actions** — namespace reserved,
not built in v1 (`zap` alone drags in the whole Lightning flow).
- **Scrolling / multi-viewport cards** — a card is one fixed-size canvas.
- **Arbitrary HTML / CSS / JavaScript.** A stack is a *safe declarative
schema* Artstr interprets — never executable web content. No iframes,
no script, no custom CSS, no `eval`. An action chain is data. This is
the single most important constraint.
- **A scripting language.** v1 has triggers + action chains, *not*
variables, counters, or conditionals. Those are a defined later tier
(Phase E) — the action model is built to accept them without a rewrite.
- **New layer *types*** for interactivity — `onClick` is an *attribute*
of the existing layers. (Audio / video layers are a real future
addition, but later.)
- **Asset library, modals, Nostr embeds, Blossom upload, static-HTML
export, stack templates** — all later.
- **Extra triggers** (`hover`, `timer`, `keypress`, `double-click`) —
later; v1 triggers are `click`, `card-enter`, `card-exit`.
- **Forms, inputs, server logic** — a stack is a presentation artifact,
not an app.
### Decisions (proposed — confirm before building)
1. **New `templateMode: 'stack'`**, reusing the Deck Builder's
card-container engine — *not* a flag on `deck`. A Stack and a Slide
Deck are distinct products (free aspect + interactivity vs. locked
16:9 + linear presenting), even though they share most code.
2. **v1 action set: `goto-card` and `open-url` only.** Everything else is
reserved namespace.
3. **`open-url` shows a confirm** with the destination before navigating
— a published stack is untrusted content.
4. **One event = one stack** (all cards embedded inline), exactly like a
deck. Multi-event stacks are explicitly not v1.
5. Additive schema — `action` on layers, `id` on cards, a `stack` payload
object. No `SCHEMA_VERSION` bump required.
1. **New `templateMode: 'stack'`**, reusing the Deck Builder's card-
container engine via an `isDeckLike(mode)` helper — not a flag on
`deck`. A Stack and a Slide Deck are distinct products.
2. **Interactivity is an attribute, not a layer type** — cards carry
`onEnter`/`onExit`, layers carry `onClick`; each is an action chain.
3. **v1 action set:** `goto-card`, `go-back`, `open-url`, `show-layer`,
`hide-layer`, `toggle-layer`. Everything else is reserved namespace.
4. **`open-url` shows a confirm** with the destination before navigating.
5. **One event = one stack** (all cards embedded inline), like a deck.
6. Additive schema — `onClick` on layers, `onEnter`/`onExit` + `id` on
cards, a `stack` payload object. No `SCHEMA_VERSION` bump required.
---
@@ -78,33 +92,38 @@ wholesale. An `isDeckLike(mode)` helper covers both `'deck'` and
load / fork path; `naddr` sharing; the community feed card with a
count badge.
- **The whole canvas editor** — layers, text, shapes, the pen/pencil
tools, clipping, undo/redo all apply to a card unchanged.
tools, clipping, undo/redo all apply to a card unchanged. Every layer
already has a stable `id` (`makeLayerId()`) and a `visible` flag — the
exact hooks `show`/`hide`/`toggle` actions need.
- **Theme composition** — `composeSlideForDeck` is reused
(`ignoreStackTheme` mirrors `ignoreDeckTheme`).
- **Encryption** — a stack payload is deck-shaped, so the
`PRIVATE_PUBLISHING_FEATURE.md` envelope wraps it with zero extra work.
- **The fullscreen renderer** — Presenter Mode already renders a card
full-screen, letterboxed, via `renderCustomArtPreviewDOM`. The
interactive viewer is that runtime plus an action-routing overlay.
interactive runtime is that renderer plus an action layer.
### Genuinely new work
1. A per-stack **aspect ratio** (decks are locked 16:9).
2. The **`action`** field on layers + the editor UI to set it.
3. Stable **card `id`s** (so `goto-card` survives reordering).
4. The **interactive viewer** — action hit-testing over the rendered card.
2. The **behaviour model** — `onClick` on layers, `onEnter`/`onExit` on
cards, each an ordered **action chain**.
3. The **action-builder UI** in the editor.
4. The **interactive runtime** — executes chains, tracks per-card layer
visibility, hit-tests clickable layers.
5. A public **viewer route** (`/p/<naddr>`).
### Current-state facts the plan is built on
- `templateMode: 'deck'` — embedded ordered cards, the sorter view,
edit-a-card-in-the-canvas-editor with write-back, publish/fork as
`casewrap-deck`, feed card with a count badge. (See
`SLIDE_DECK_FEATURE.md`.)
edit-a-card-with-write-back, publish/fork as `casewrap-deck`, feed card
with a count badge. (See `SLIDE_DECK_FEATURE.md`.)
- Presenter Mode (`#presenterMode`) — fullscreen, renders a card with
`renderCustomArtPreviewDOM`, keyboard nav, an Audience (slide-only)
view.
- Layer renderer positions every layer as an absolutely-placed div from
inch coordinates; `renderCustomArtPreviewDOM` does the same at preview
scale, with `pointer-events:none` on layers.
`renderCustomArtPreviewDOM`, keyboard nav, capture-phase keyboard
ownership, an Audience (slide-only) view.
- Every layer has a stable `id` and a `visible` boolean; the layer
renderer positions each layer as an absolutely-placed div from inch
coordinates.
- `casewrap-*` typed `kind:30078` events; `/share/<naddr>` cold-resolves;
`vercel.json` rewrites `/share/:id` and `/u/:npub` to the SPA.
@@ -123,14 +142,16 @@ wholesale. An `isDeckLike(mode)` helper covers both `'deck'` and
stack: {
aspect: '16:9', // 16:9 | 9:16 | 1:1 | 4:3 | custom
width: 1920, height: 1080, // canvas size every card shares
theme: { fontFamily: '', background: '' }, // same as deck.theme
theme: { fontFamily: '', background: '' },
startCard: '<cardId>', // entry card (default: first)
cards: [
{
id: 'c_a1b2c3', // stable; goto-card targets this
name: 'Cover', // optional label in the sorter
name: 'Cover',
background: '#111111',
layers: [ … ], // standard layers, may carry `action`
layers: [ … ], // standard layers; may carry onClick
onEnter: [ <action>, … ], // runs when the card is shown
onExit: [ <action>, … ], // runs when leaving the card
ignoreStackTheme: false
},
…
@@ -139,53 +160,90 @@ wholesale. An `isDeckLike(mode)` helper covers both `'deck'` and
}
```
- A `card` is the deck's slide entry generalised: it gains a stable `id`,
drops `notes`, and its `layers` may carry `action`s. Canvas dimensions
live at the **stack** level (every card shares the aspect) rather than
per-card — all cards in a stack are the same size.
- **Feeding the shared renderer:** the deck's renderer / theme composition
read a per-entry `{ slide: {width,height,background}, layers }` shape.
For a stack, the engine synthesises that on the fly — `width`/`height`
from `stack`, `background` from the card — so `renderCustomArtPreviewDOM`
and `composeSlideForDeck` are reused untouched. (`ignoreStackTheme`
mirrors `ignoreDeckTheme`.)
- **`startCard`** holds a card `id`; if that card is deleted it falls back
to the first card.
- A `card` is the deck's slide entry generalised: a stable `id`, no
`notes`, plus `onEnter` / `onExit` action chains.
- **Feeding the shared renderer:** the deck renderer reads a per-entry
`{ slide:{width,height,background}, layers }` shape. For a stack the
engine synthesises that — `width`/`height` from `stack`, `background`
from the card — so `renderCustomArtPreviewDOM` / `composeSlideForDeck`
are reused untouched.
- **`startCard`** holds a card `id`; if that card is deleted it falls
back to the first card.
### The `action` on a layer
### Layer interactivity
Any layer may carry an optional `action`:
Any layer gains one optional field — an `onClick` action chain:
```js
// Jump to another card — target is a card id or a keyword.
{ "action": { "type": "goto-card", "target": "c_a1b2c3" } }
{ "action": { "type": "goto-card", "target": "next" } } // next|prev|first|last
// Open an external link.
{ "action": { "type": "open-url", "url": "https://example.com" } }
{
id: 'l_x1', type: 'shape', /* …normal layer fields… */
visible: true,
onClick: [ <action>, <action>, … ] // empty / absent = not interactive
}
```
- `goto-card.target` is a **card id**, never an index — reordering or
duplicating cards never breaks a link.
- A dangling target (the card was deleted) is a viewer no-op + a console
warning; the editor flags it.
- Unknown `action.type` values are ignored by the viewer — forward-safe
for the reserved set (`open-nostr`, `zap`, `toggle-layer`, …).
A **hotspot** is just a transparent shape layer (no fill, no stroke) with
an `onClick`. No new layer type — the editor offers an "Add hotspot"
convenience that drops a transparent, click-ready rect; otherwise any
text / image / shape layer can be given an `onClick`.
### Hotspots
### Triggers (v1)
No new layer type. A **hotspot** is just a shape (rect) layer that is
transparent — no fill, no stroke — carrying an `action`. The editor
offers an **"Add hotspot"** convenience that drops a transparent rect
ready for an action; otherwise any image / text / shape layer can be made
actionable.
| Trigger | Lives on | Fires when |
|---|---|---|
| `onClick` | a layer | the layer is clicked / tapped / Enter-activated |
| `onEnter` | a card | the card becomes visible |
| `onExit` | a card | the runtime leaves the card |
Named slots, not a generic trigger list — `onHover`, `onDoubleClick`,
etc. slot in later as additional named slots.
### Actions
An **action chain** is an ordered array of actions. v1 action types:
```js
{ "type": "goto-card", "target": "c_a1b2c3" } // or next|prev|first|last
{ "type": "go-back" } // pop the visit back-stack
{ "type": "open-url", "url": "https://example.com" }
{ "type": "show-layer", "target": "l_x1" } // layer id on this card
{ "type": "hide-layer", "target": "l_x1" }
{ "type": "toggle-layer", "target": "l_x1" }
```
**Execution rules:**
- A chain runs top to bottom.
- `goto-card` / `go-back` **end the chain** — actions after them are
ignored (you're on a different card now). The editor flags trailing
actions.
- `show` / `hide` / `toggle` `target` a layer **on the current card**;
cross-card layer manipulation is not v1.
- **`onEnter` / `onExit` chains may not contain `goto-card` or `go-back`**
— only `onClick` can navigate. This makes card-enter→navigate→card-enter
loops structurally impossible.
- Chains are capped at **20 actions**; the runtime ignores the excess.
- Unknown `action.type` values are silently ignored — forward-safe for
the Phase E logic actions (`set-variable`, `if`, …) and reserved ones
(`open-nostr`, `zap`, `play-audio`, …).
### Runtime layer visibility
Show/hide/toggle never mutate the stored design. The runtime keeps a
**per-card visibility map** (`layerId → bool`), initialised from each
layer's authored `visible`. Re-entering a card **resets** that card's map
to the authored state, then runs `onEnter` — so behaviour is predictable
and re-playable. (Persisting runtime state across visits is the Phase E
variables tier.)
A `composeCardForRuntime(card, visibilityMap)` helper — analogous to
`composeSlideForDeck` — yields the `{layers}` the renderer draws.
### Nostr
- Tag `casewrap-stack`; `d`-tag `stack:<title|id>` (or a random `d`-tag
when published privately, per `PRIVATE_PUBLISHING_FEATURE.md`).
- Forkable and editable-in-place via the existing NIP-33 flow.
- `naddr` shareable; encryptable.
- Tag `casewrap-stack`; `d`-tag `stack:<title|id>` (random when published
privately, per `PRIVATE_PUBLISHING_FEATURE.md`).
- Forkable, editable-in-place (NIP-33), `naddr`-shareable, encryptable.
---
@@ -197,85 +255,104 @@ The Deck Builder sorter, reused. A new **Stack** option in the Template-
tab Layout panel (beside Case cover / Disc labels / Slide deck). The
sidebar adds a **Stack** panel:
- **Aspect ratio** — 16:9 / 9:16 / 1:1 / 4:3 / custom. Chosen up front.
Changing it later **re-fits every card** by reusing the deck's existing
import-letterbox remap (`deckSlideFromImport`'s scale-to-fit + centre
math) — layers scale and centre into the new aspect instead of being
stranded off-canvas. The user confirms before a re-fit runs.
- **Aspect ratio** — 16:9 / 9:16 / 1:1 / 4:3 / custom. Chosen up front;
changing it later **re-fits every card** via the deck's import-letterbox
remap (scale-to-fit + centre), behind a confirm — layers are never
stranded off-canvas.
- **Add card** (blank / import a design / import from Nostr), duplicate,
delete, drag-reorder, inline-rename — all from the deck sorter. Imports
letterbox onto the **stack's** aspect, not a hardcoded 16:9 — the
existing remap is generalised to take a target width/height.
- **Start card** — which card the viewer opens on (default: first).
letterbox onto the **stack's** aspect, not a hardcoded 16:9.
- **Start card** — which card the runtime opens on (default: first).
- A **Stack theme** panel (font + background), reused from the deck.
- A per-card **onEnter / onExit** editor (an action-chain editor; see
surface 3) reachable from the card's context menu.
### 2. Card editor + the Action control
### 2. Card editor — drawing
Clicking a card opens it in the canvas editor (the deck edit round-trip,
unchanged). New in the **layer panel**: an **Interactivity** section for
the selected layer —
unchanged). All existing tools apply. New affordances:
- **Add hotspot** — drops a transparent, click-ready rect.
- A layer with a non-empty `onClick` shows an **action badge** in the
layer list and a subtle dashed outline on canvas, so interactive
regions are visible while authoring.
### 3. The action builder
The heart of the feature. A panel — opened for the selected layer
(`onClick`) or for a card (`onEnter` / `onExit`) — that edits one action
chain with **no JSON**:
```
Interactivity
Action: [ None ▾ ] None | Go to card | Open URL
→ Go to card: [ Card ▾ ] first | previous | next | last | <named card>
→ Open URL: [ https://… ]
When this layer is clicked: [+ Add action ▾]
⠿ 1. Hide layer → [ Intro Text ▾ ] ✕
⠿ 2. Show layer → [ Secret Panel ▾ ] ✕
⠿ 3. Go to card → [ Chapter 2 ▾ ] ✕
```
- Layers with an action get a badge in the layer list and a subtle dashed
outline on the canvas, so interactive regions are visible while editing.
- An **"Add hotspot"** button drops a transparent, action-ready rect.
- A dangling `goto-card` target is flagged in the panel.
- **+ Add action** menu: Go to card · Go back · Open URL · Show layer ·
Hide layer · Toggle layer.
- Each action row has a type-appropriate target picker — a **card
dropdown** (the keywords *first / previous / next / last* plus every
named card) for `goto-card`; a **layer dropdown** (layers on this card)
for show/hide/toggle; a URL field for `open-url`.
- Rows are **drag-reorderable** (⠿) and removable.
- The editor **warns** on: actions after a `goto-card`/`go-back`
(unreachable); a `goto-card` whose target card was deleted (dangling);
navigation actions placed in a card's `onEnter`/`onExit` (not allowed).
- `target`s are stored as **ids** (card id, layer id) — reordering or
renaming never breaks a chain.
### 3. The interactive viewer
### 4. The interactive runtime
The fullscreen runtime, built on the Presenter Mode engine:
The fullscreen player, built on the Presenter Mode engine:
- Renders the current card full-screen, letterboxed to the stack aspect,
themed (`composeSlideForDeck` reused), via `renderCustomArtPreviewDOM`.
- An **action overlay** sits above the card: for each layer with an
`action`, a positioned hit region matching the layer's (transformed)
box. The regions are real **`<button>`s**, so they are Tab-focusable,
Enter/Space-activatable, and screen-reader-announced — the viewer is
not mouse-only. Overlapping actionable layers resolve by **z-order**
(topmost wins); a hovered/focused region shows a pointer cursor.
- **Click routing:** `goto-card` → navigate (ids; or the keywords
`next` / `prev` / `first` / `last`, **clamped at the ends — no wrap**);
`open-url` → a confirm dialog showing the destination, then open in a
new tab with `rel="noopener noreferrer"`. Clicking empty (non-action)
space is a **no-op** — a stack wanting click-anywhere advance adds a
full-card hotspot.
- **Keyboard fallback** — arrows / space still do prev/next so a stack
with no on-card navigation is never a dead end; `Esc` exits. The viewer
owns the keyboard (capture-phase handler), like Presenter Mode.
- **History** — visited cards form a back-stack so a Back control / the
Backspace key can return, even across `goto-card` jumps.
- **Touch** — taps act as clicks; fullscreen is best-effort (matching
Presenter Mode). A 9:16 stack is the natural phone format.
- Launching **from the editor** persists the in-progress card edit first
(the `persistEditingDeckSlide` step), so the viewer always shows saved
state.
- Entry points: a ▶ **"View"** button in the Stack Builder tool palette,
and an **"Open Stack"** button on a stack's community-browser preview.
themed, with the runtime visibility map applied
(`composeCardForRuntime`).
- On entering a card: reset its visibility map → run `onEnter`.
- An **action overlay** sits above the card: for each layer with a
non-empty `onClick`, a positioned hit region matching the layer's
(transformed) box, rendered as a real **`<button>`** — Tab-focusable,
Enter/Space-activatable, screen-reader-announced. Overlapping
interactive layers resolve by **z-order** (topmost wins); a
hovered/focused region shows a pointer cursor.
- **Activating a layer** runs its `onClick` chain. `goto-card` runs the
current card's `onExit`, then navigates, then the new card's `onEnter`.
`open-url` shows a confirm with the destination, then opens a new tab
with `rel="noopener noreferrer"`. `show/hide/toggle` mutate the
visibility map and re-render the card. Clicking empty (non-action)
space is a **no-op**.
- **Navigation safety net** — arrows / space still do prev/next, a
visited-card **back-stack** powers `go-back` and Backspace, `Esc`
exits. So a stack with broken or missing on-card navigation is never a
dead end. `goto-card` keywords clamp at the ends (no wrap).
- The runtime **owns the keyboard** (capture-phase handler), like
Presenter Mode. Launching from the editor persists the in-progress card
edit first.
- **Touch** — taps act as clicks; fullscreen is best-effort.
- Two modes: **preview** (inside the editor — the ▶ View button in the
Stack Builder tool palette) and **published** (`/p/<naddr>`).
- Entry points: ▶ **View** in the Stack Builder palette; **Open Stack**
on a stack's community-browser preview page.
### 4. Public viewer route
### 5. Public viewer route
`/p/<naddr>` — a cold-loadable URL that resolves the addressable event
and boots straight into the interactive viewer (no editor chrome). Needs
a `vercel.json` rewrite (`/p/:id` → SPA), mirroring `/share/:id`.
`/share/<naddr>` of a stack continues to open the editor, as other
designs do. A `/p/` link to an **encrypted** stack the viewer can't
decrypt shows a "This stack is private" locked state, consistent with
`PRIVATE_PUBLISHING_FEATURE.md`.
and boots straight into the runtime (no editor chrome). Needs a
`vercel.json` rewrite (`/p/:id` → SPA), mirroring `/share/:id`.
`/share/<naddr>` of a stack still opens the editor. A `/p/` link to an
**encrypted** stack the viewer can't decrypt shows a "This stack is
private" locked state, per `PRIVATE_PUBLISHING_FEATURE.md`.
### 5. Community feed
### 6. Community feed
A published stack is a normal `kind:30078` event:
- **Feed-card thumbnail** — the start card's mini-render with a
card-count badge (e.g. `8 cards`), reusing the deck preview renderer.
- **Feed-card thumbnail** — the start card's mini-render with a card-count
badge (e.g. `8 cards`), reusing the deck preview renderer.
- **Publish-confirm preview** — likewise the start card.
- **Preview page** — carries the **Open Stack** button (surface 3).
- **Preview page** — carries the **Open Stack** button (surface 4).
- `stack` joins `MODE_LABELS`, the dedup fingerprint, the mode filters,
and gets a `stack` ("Interactive page") category.
@@ -283,64 +360,86 @@ A published stack is a normal `kind:30078` event:
## Security
- **No executable content.** The schema is declarative; the viewer only
ever interprets the two known action types and ignores the rest. No
HTML injection, no script, no iframes, no remote CSS.
- **`open-url` is the one outward vector** — always route it through a
confirm dialog that shows the real destination URL, and open with
`rel="noopener noreferrer"` in a new tab. Never auto-navigate.
- **Image layers stay remote URLs**, as everywhere in Artstr — no binary
embedded, payload size bounded to layer JSON.
- A stack imported / forked from the community is untrusted; the same
rules apply when *editing* it (e.g. surface dangling/odd actions).
- **No executable content.** The schema is declarative; the runtime only
ever interprets the known action types and ignores the rest. No HTML
injection, no script, no iframes, no remote CSS, no `eval`.
- **`open-url` is the one outward vector** — always behind a confirm
dialog showing the real destination; new tab; `rel="noopener
noreferrer"`; only `http(s):` / `mailto:` schemes — never `javascript:`
or `data:`.
- **Action chains are bounded** — capped length, no recursion (a chain
cannot call another chain), and `onEnter`/`onExit` cannot navigate, so
there is no loop or runaway-execution surface.
- **Image layers stay remote URLs**, as everywhere in Artstr — payload
bounded to layer/action JSON.
- A forked/imported stack is untrusted; `sanitizeStackJson()` drops
unknown action types and clamps chain lengths on load.
---
## Phased delivery
> **A + B + C together are the minimum genuinely-useful unit** — a stack
> with no actions and no viewer is just a free-aspect deck. Ship them as
> one arc, gated individually.
> with no behaviour is just a free-aspect deck; behaviour with no runtime
> can't be played. Ship them as one arc, gated individually.
### Phase A — Stack format
- `templateMode: 'stack'` reusing the deck engine via `isDeckLike()`.
- The `stack` data model; per-stack aspect ratio (presets + custom).
- Stack option in the Template-tab Layout panel; the Stack + theme
sidebar panels; the card sorter (add / import / duplicate / delete /
reorder / rename).
- Stable card `id`s.
reorder / rename); stable card `id`s.
- Save / load / publish / fork as `casewrap-stack`; feed card + count
badge; `stack` in `MODE_LABELS` / categories / mode filters.
**Ship gate:** build a 5-card 9:16 stack, reorder it, edit a card, save,
reload, publish — it round-trips and shows in the feed with its count.
**Ship gate:** build a 5-card 9:16 stack, reorder, edit a card, save,
reload, publish — round-trips; shows in the feed with its count.
### Phase B — Interactive layers
- The `action` field on layers; the layer-panel Interactivity control.
- "Add hotspot" convenience; actionable-layer badges + canvas outline.
- Dangling-target detection.
### Phase B — Behaviour authoring
- Layer `onClick` + card `onEnter` / `onExit` action chains in the data
model; `sanitizeStackJson()` validation on load.
- The action set: `goto-card`, `go-back`, `open-url`, `show-layer`,
`hide-layer`, `toggle-layer`.
- The **action-builder panel** (surface 3): add / reorder / remove
actions, id-based target pickers, the warnings.
- "Add hotspot"; action badges on interactive layers.
**Ship gate:** add a hotspot with a `goto-card` action and a button with
an `open-url` action; both persist through save / reload.
**Ship gate:** give a hotspot an `onClick` of *hide A → show B → go to
card 2*; give a card an `onEnter` that shows a layer; both round-trip
through save / reload / publish; dangling targets are flagged.
### Phase C — Interactive viewer
- The fullscreen viewer: action overlay, click routing, `open-url`
confirm, keyboard fallback, visited-card back-stack.
### Phase C — Interactive runtime
- The fullscreen runtime: `composeCardForRuntime`, the action overlay,
chain execution, per-card visibility, `onEnter`/`onExit`, the
navigation safety net, `open-url` confirm.
- ▶ View entry point in the Stack Builder palette; Open Stack on the
community preview.
**Ship gate:** play a stack end to end by clicking hotspots; `goto-card`
jumps and Back works; `open-url` confirms then opens.
**Ship gate:** play a stack end to end — clicking hotspots reveals/hides
layers and jumps cards; `onEnter` fires; Back + keyboard + `Esc` always
work; `open-url` confirms then opens.
### Phase D — Public viewer route + polish
- `/p/<naddr>` route + `vercel.json` rewrite; cold-boot into the viewer.
- Profile-page "feature this stack" pinning; share-link affordances.
- `/p/<naddr>` route + `vercel.json` rewrite; cold-boot into the runtime.
- Profile-page "feature this stack" pinning.
### Later
- Blossom / NIP-B7 media upload.
- More actions: `open-nostr`, `zap`, `toggle-layer`.
### Phase E — Stack logic *(the scripting tier)*
- `state` variables on the stack; actions `set-variable`,
`increment-variable`; a conditional action `{ type:'if', condition,
then:[…], else:[…] }`. Local-only state (`localStorage`), reset on
runtime start.
- A condition builder in the action panel.
The v1 action model already tolerates these (unknown types are ignored),
so Phase E is additive — no schema rewrite.
### Later / backlog
- Audio / video layer types and `play`/`pause` actions; `onMediaEnd`.
- More triggers: `onHover`, `onDoubleClick`, `onTimer`, `onKeyPress`.
- `open-nostr` (npub / note / naddr) and `zap` actions; Nostr embeds.
- Card transitions; stack templates (zine / portfolio / comic / landing).
- Multi-event stacks (manifest + per-card events) for very large stacks.
- Blossom / NIP-B7 media upload; an asset library.
- Modal overlays; static-HTML export; multi-event stacks.
---
@@ -348,56 +447,62 @@ jumps and Back works; `open-url` confirms then opens.
| Risk | Mitigation |
|---|---|
| "Pages on Nostr" invites expectations of real websites (HTML/JS/SEO/hosting) | Name and frame it as *interactive stacks/pages*, not websites; the declarative-only schema is a hard line. |
| `open-url` is a phishing / malware vector in shared stacks | Mandatory confirm dialog showing the destination; new tab + `noopener`; never auto-navigate. |
| `goto-card` targets break on reorder / delete | Targets are stable card `id`s, never indices; dangling targets are editor-flagged and viewer no-ops. |
| A stack with only on-card navigation could trap the viewer | Keyboard prev/next + a back-stack + `Esc` always work, regardless of the stack's own buttons. |
| Aspect-ratio change strands existing layers off-canvas | Changing aspect re-fits every card through the existing import-letterbox remap (scale-to-fit + centre), behind a confirm — layers are never silently stranded. |
| Stack vs Deck duplication | One shared card-container engine via `isDeckLike()`; only aspect, actions, and the viewer branch. |
| Big stacks → large events (relay size caps) | Images are URLs; warn past ~100 KB / ~25 cards; Blossom offload + multi-event stacks deferred. |
| Layer click hit-testing | The overlay reuses the renderer's box math; a CSS `transform:rotate` hit region catches clicks in its true rotated shape, so **rotation is exact**. A *clip-masked* layer hit-tests on its bounding box, not the mask shape — documented limitation; authors wanting a precise non-rect hotspot use a rotated/sized rect. |
| Viewer is mouse-centric | Action regions are focusable `<button>`s — Tab / Enter / Space and screen readers work; keyboard prev/next is always available. |
| Action chains become a back-door scripting language | v1 is a fixed, finite action set; no variables/branching until Phase E; chains can't call chains; the runtime interprets, never `eval`s. |
| `onEnter` → `goto-card` → `onEnter` infinite loop | `onEnter`/`onExit` chains structurally cannot contain navigation actions; only `onClick` navigates. |
| `open-url` is a phishing / malware vector | Confirm dialog showing the destination; new tab + `noopener`; scheme allowlist; never `javascript:`/`data:`. |
| Targets break on reorder / delete / rename | Actions reference stable card / layer **ids**, never indices or names; dangling targets are editor-flagged and runtime no-ops. |
| A stack traps the viewer (bad on-card nav) | Keyboard prev/next, a back-stack, and `Esc` always work regardless of the stack's own buttons. |
| Aspect-ratio change strands layers | Changing aspect re-fits every card via the import-letterbox remap, behind a confirm. |
| Runtime visibility drifts / becomes unpredictable | Per-card map reset to authored `visible` on every (re-)entry; show/hide/toggle never mutate stored data. |
| Stack vs Deck code duplication | One shared card-container engine via `isDeckLike()`; only aspect, behaviour, and the runtime branch. |
| Layer click hit-testing | The overlay reuses the renderer's box math; a CSS `transform:rotate` hit region catches clicks in its true rotated shape (rotation is exact); a clip-masked layer hit-tests on its bounding box — documented limitation. |
| Viewer is mouse-centric | Action regions are focusable `<button>`s — Tab / Enter / Space and screen readers work. |
| Big stacks → large events (relay size caps) | Images are URLs; warn past ~100 KB / ~25 cards; Blossom + multi-event stacks deferred. |
---
## Open questions
1. **Mode** — confirm `templateMode: 'stack'` reusing the deck engine,
vs. folding actions + free aspect into `deck` itself.
2. **v1 action set** — `goto-card` + `open-url` only — confirmed?
3. **Aspect presets** — 16:9 / 9:16 / 1:1 / 4:3 / custom — the right set?
4. **Viewer** — extend the Presenter Mode runtime with an interactive
mode, or a separate viewer module? (Lean: extend it.)
5. **Route** — `/p/<naddr>` as proposed, or a different prefix?
6. **Category** — one `stack` category, or split (e.g. `page`, `zine`)?
1. **Mode** — confirm `templateMode: 'stack'` reusing the deck engine.
2. **v1 action set** — the six above — confirmed? (`open-nostr` deferred.)
3. **Aspect presets** — 16:9 / 9:16 / 1:1 / 4:3 / custom — right set?
4. **Runtime** — extend the Presenter Mode engine, or a separate module?
(Lean: extend it.)
5. **Route** — `/p/<naddr>` as proposed?
6. **Category** — one `stack` category, or split (`page`, `zine`)?
7. **Re-entry semantics** — confirm a card resets its layer visibility on
every entry (vs. remembering it). v1 plan: reset.
---
## Acceptance summary
### Phase A
- [ ] Stack is a Template-tab layout; per-stack aspect ratio works, and
changing it re-fits existing cards (no stranded layers).
- [ ] Stack is a Template-tab layout; per-stack aspect works, and
changing it re-fits existing cards.
- [ ] Card sorter: add / import / duplicate / delete / reorder / rename;
cards have stable ids; imports letterbox onto the stack's aspect.
- [ ] Save / load / publish / fork as `casewrap-stack`; feed card shows a
card-count badge.
### Phase B
- [ ] Any layer can be given a `goto-card` or `open-url` action; a
transparent hotspot can be added in one click.
- [ ] Actions round-trip through save / reload / publish; dangling
`goto-card` targets are flagged.
- [ ] A layer can be given an `onClick` chain and a card an
`onEnter`/`onExit` chain via the no-JSON action builder.
- [ ] Action chains support goto-card / go-back / open-url /
show / hide / toggle-layer; rows reorder; targets are id-based.
- [ ] Chains round-trip through save / reload / publish; dangling
targets, trailing-after-navigation, and navigation-in-onEnter are
all flagged.
### Phase C
- [ ] The interactive viewer renders cards full-screen and routes clicks
on actionable layers.
- [ ] `goto-card` (ids + keywords, clamped at the ends) navigates; a
back-stack + keyboard + `Esc` always work; `open-url` confirms then
opens in a new tab.
- [ ] The viewer owns the keyboard; theme is composed in.
- [ ] Action regions are focusable buttons — Tab / Enter reach and fire
them; empty-space clicks are no-ops.
- [ ] The runtime renders cards full-screen and executes `onClick`,
`onEnter`, `onExit` chains.
- [ ] show / hide / toggle change layer visibility live; visibility
resets on card re-entry.
- [ ] `goto-card` (ids + clamped keywords) and `go-back` navigate; a
back-stack + keyboard + `Esc` always work; `open-url` confirms.
- [ ] Action regions are focusable buttons; empty-space clicks are
no-ops; the runtime owns the keyboard; theme is composed in.
### Phase D
- [ ] `/p/<naddr>` cold-boots a published stack into the viewer.
- [ ] `/p/<naddr>` cold-boots a published stack into the runtime.