@genex-ai/cli-demo 0.48.0-dev.78 → 0.48.0-dev.80
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/package.json +1 -1
- package/templates/skills/genex-ai-hud/SKILL.md +37 -1
- package/templates/skills/genex-ai-hud/references/stage2-prompt-template.md +5 -1
- package/templates/skills/genex-ai-menu/SKILL.md +51 -4
- package/templates/skills/genex-threejs-game-ui/SKILL.md +33 -6
- package/templates/skills/genex-threejs-skill-router/SKILL.md +7 -3
- package/templates/skills/genex-threejs-skill-router/references/routing-map.md +58 -16
package/package.json
CHANGED
|
@@ -28,6 +28,34 @@ ring) that no hand-written CSS can fake.
|
|
|
28
28
|
the CSS placeholder, finished to `$genex-threejs-game-ui`'s Tier-1 floor,
|
|
29
29
|
ships as the HUD.
|
|
30
30
|
|
|
31
|
+
## Three layers — plates are DOM, chrome is sprites, data is DOM
|
|
32
|
+
|
|
33
|
+
- **Plate (DOM).** The backing surface behind a widget is built in CSS — in
|
|
34
|
+
whatever register the brief and concept frame call for: smoked glass
|
|
35
|
+
(`background: rgba(<darkest hue>, 0.35–0.6)` + `backdrop-filter: blur(6–12px)`
|
|
36
|
+
and its `-webkit-` twin) when the style wants translucency, a solid painted
|
|
37
|
+
plate, a subtle gradient — or no plate at all for an outline-led look.
|
|
38
|
+
**Glass is a technique, not a default** — do not reach for translucency
|
|
39
|
+
because this skill mentions it; reach for it when THIS game's brief does.
|
|
40
|
+
What is fixed is WHERE plates are built — the DOM — because
|
|
41
|
+
semi-transparency physically cannot survive sprite extraction: a composited
|
|
42
|
+
panel's pixels are a blend of panel and background, and no cutout can
|
|
43
|
+
un-blend them. Plates are NEVER baked into sprites.
|
|
44
|
+
- **Chrome (sprites — what THIS pipeline generates).** Opaque frames, corner
|
|
45
|
+
brackets, ornaments, emblems, icons, medallions — hard-alpha art laid over
|
|
46
|
+
the glass. The Stage-2 sheet contains ONLY chrome; never a panel with its
|
|
47
|
+
fill or glass surface baked in.
|
|
48
|
+
- **Data (DOM).** Numbers, labels, and bar fills are DOM text and masked
|
|
49
|
+
fills (the fillBox pattern below) — always over the glass, never
|
|
50
|
+
rasterized into art.
|
|
51
|
+
- **Micro-elements are designed, not defaulted.** The sheet excludes
|
|
52
|
+
primitive-only shapes (plain crosshair ticks, dots, pips) because CSS
|
|
53
|
+
renders them crisper — but CSS never means default: the reticle and every
|
|
54
|
+
cue, toast, and counter still get the brief's treatment (stroke weight,
|
|
55
|
+
glow, state changes — hit, low-ammo, empowered). An ornamented,
|
|
56
|
+
identity-bearing reticle (engraved ring, sculpted brackets) is chrome —
|
|
57
|
+
put it on the sheet like any other asset.
|
|
58
|
+
|
|
31
59
|
## Framing rules — internalize before prompting
|
|
32
60
|
|
|
33
61
|
The HUD overlays a **live 3D scene**. The pixels between widgets show the game.
|
|
@@ -36,6 +64,11 @@ The HUD overlays a **live 3D scene**. The pixels between widgets show the game.
|
|
|
36
64
|
fabric, wood board) as part of the HUD. Material aesthetics belong on the
|
|
37
65
|
widget FRAMES — the HP bar frame is parchment, the minimap ring is leather —
|
|
38
66
|
never on a screen-filling sprite.
|
|
67
|
+
- **No plates in sprites.** Panel plates in the mockup — translucent or
|
|
68
|
+
solid — are welcome as the LOOK, but the Stage-2 asset list requests only
|
|
69
|
+
their opaque chrome (frame, brackets, ornament); the plate itself is
|
|
70
|
+
rebuilt at runtime in CSS matching the mockup's tint (see the three layers
|
|
71
|
+
above).
|
|
39
72
|
- **4–7 widgets.** Fewer doesn't read as a HUD; more clutters the screen and
|
|
40
73
|
burns generations.
|
|
41
74
|
- **Every widget needs internal contrast** — a panel fill, outline stroke, or
|
|
@@ -92,7 +125,10 @@ and [references/stage2-prompt-template.md](references/stage2-prompt-template.md)
|
|
|
92
125
|
**Order of work: kick Stage 1 off as your FIRST action after the UI plan
|
|
93
126
|
gate**, then write the widget layout and wiring code (placement, masked-fill
|
|
94
127
|
scaffolding, plain-CSS placeholder bars) while it renders — the CSS HUD keeps
|
|
95
|
-
the game playable until the sprites land.
|
|
128
|
+
the game playable until the sprites land. When the gate produced a concept
|
|
129
|
+
frame, anchor Stage 1 to it — `--edit <concept-url>` holds the game's
|
|
130
|
+
register far better than a text-only prompt; single-sprite regenerations
|
|
131
|
+
anchor the same way. Generate 2 mockup candidates in one
|
|
96
132
|
call (`--candidates 2`) and pick the better one: a re-roll costs the whole
|
|
97
133
|
serial chain, a second candidate costs nothing extra in wall-clock.
|
|
98
134
|
|
|
@@ -15,6 +15,10 @@ Fill `[ASSET_LIST]`, then pass the whole text as the prompt to
|
|
|
15
15
|
identity labels);
|
|
16
16
|
- **states what to REMOVE** (`runtime_static` labels and `runtime_dynamic`
|
|
17
17
|
values, fills, items);
|
|
18
|
+
- **requests chrome only** — never a panel's plate/backing surface,
|
|
19
|
+
translucent or solid: plates are rebuilt at runtime in CSS (rgba,
|
|
20
|
+
backdrop-filter, or a solid fill — whatever matches the mockup), so ask
|
|
21
|
+
for the panel's frame/brackets/ornament, not the panel surface itself;
|
|
18
22
|
- **requests the annotated duplicate** for every bar/meter frame (the
|
|
19
23
|
template's progress block below covers the mechanics — your item just says
|
|
20
24
|
"output the clean frame + annotated duplicate pair").
|
|
@@ -61,7 +65,7 @@ DO NOT include in the output sheet:
|
|
|
61
65
|
- Simple dot indicators, status pips, blip markers
|
|
62
66
|
- Fills inside bars, items inside slots, map terrain inside minimap rings (these are added at runtime by CSS / game code)
|
|
63
67
|
|
|
64
|
-
Those excluded elements will be recreated by runtime code (CSS shapes and DOM text) — much crisper than a downsampled PNG.
|
|
68
|
+
Those excluded elements will be recreated by runtime code (CSS shapes and DOM text) — much crisper than a downsampled PNG. Recreated does NOT mean default-styled: reticles, pips, and cues still follow the style brief (stroke weight, glow, state colors). An ornamented, identity-bearing reticle — an engraved ring, sculpted brackets — is NOT a simple primitive; include it in the asset list like any other chrome.
|
|
65
69
|
|
|
66
70
|
Pure white #ffffff background everywhere except where the actual UI assets sit.
|
|
67
71
|
```
|
|
@@ -48,6 +48,14 @@ might suggest. Two hard rules for the frame prompt:
|
|
|
48
48
|
perches", "fog drifts left then back right", "candle flames flicker".
|
|
49
49
|
Bad: "camera dollies forward into the gate" (one-way motion fights the
|
|
50
50
|
identical endpoints and produces a visible hitch).
|
|
51
|
+
**The seam is YOUR job, not the model's:** in practice the clip lands
|
|
52
|
+
*near* frame 0, not exactly on it, and a bare `video.loop = true` shows a
|
|
53
|
+
visible hitch every cycle. Always wire the video with the DOM crossfade
|
|
54
|
+
below, and watch one full cycle before calling the menu done. (Do not
|
|
55
|
+
reach for `--loop` here: it routes to a different model with a first-class
|
|
56
|
+
loop param but NO frame conditioning — the video won't match your key art —
|
|
57
|
+
and server-side `--frame` overrides `--loop` anyway, so passing both does
|
|
58
|
+
nothing.)
|
|
51
59
|
- **Two-frame** (`--first-frame` + `--last-frame`) only when the menu
|
|
52
60
|
genuinely needs a state change (rain starts and persists, a figure walks in
|
|
53
61
|
and stays). The loop will have a visible seam where the last frame snaps
|
|
@@ -71,6 +79,44 @@ npx genex wait <gen-id>
|
|
|
71
79
|
# -> https://assets.genex.technology/generations/<gen-id>/video-mp4
|
|
72
80
|
```
|
|
73
81
|
|
|
82
|
+
## Seamless in the DOM — the loop crossfade (mandatory wiring)
|
|
83
|
+
|
|
84
|
+
Never wire the clip as a bare `<video loop>` — the residual seam shows every
|
|
85
|
+
cycle. Two stacked `<video>` elements with the same src crossfade at the
|
|
86
|
+
cycle end; any seam disappears deterministically, no regeneration lottery:
|
|
87
|
+
|
|
88
|
+
```ts
|
|
89
|
+
/** Deterministic seamless loop: two stacked <video>s crossfade at cycle end. */
|
|
90
|
+
export function seamlessLoop(holder: HTMLElement, url: string, fade = 0.6): void {
|
|
91
|
+
const mk = (): HTMLVideoElement => {
|
|
92
|
+
const v = document.createElement("video");
|
|
93
|
+
v.src = url; v.muted = true; v.playsInline = true; v.preload = "auto";
|
|
94
|
+
holder.appendChild(v);
|
|
95
|
+
return v;
|
|
96
|
+
};
|
|
97
|
+
let front = mk(), back = mk();
|
|
98
|
+
front.style.opacity = "1";
|
|
99
|
+
void front.play();
|
|
100
|
+
const tick = (): void => {
|
|
101
|
+
if (front.duration > 0 && front.currentTime >= front.duration - fade && back.paused) {
|
|
102
|
+
back.currentTime = 0;
|
|
103
|
+
void back.play();
|
|
104
|
+
front.style.transition = back.style.transition = `opacity ${fade}s linear`;
|
|
105
|
+
back.style.opacity = "1";
|
|
106
|
+
front.style.opacity = "0";
|
|
107
|
+
const old = front; front = back; back = old;
|
|
108
|
+
window.setTimeout(() => back.pause(), fade * 1000 + 50);
|
|
109
|
+
}
|
|
110
|
+
requestAnimationFrame(tick);
|
|
111
|
+
};
|
|
112
|
+
requestAnimationFrame(tick);
|
|
113
|
+
}
|
|
114
|
+
```
|
|
115
|
+
|
|
116
|
+
Pause both videos when the menu phase hides (they're decode work), and
|
|
117
|
+
resume the front one when it returns. Watch ONE full cycle in the browser
|
|
118
|
+
before calling the menu done — that's the seam check.
|
|
119
|
+
|
|
74
120
|
**Work async — the menu must never block the game.** Generate the frame right
|
|
75
121
|
after the UI plan gate, enqueue the video with `--no-wait`, ship the CSS menu
|
|
76
122
|
(buttons + title over the frame IMAGE as a static backdrop), and swap the
|
|
@@ -89,9 +135,8 @@ copy-pasteable example:
|
|
|
89
135
|
```html
|
|
90
136
|
<div id="ui">
|
|
91
137
|
<div id="screen-menu" class="screen" data-phase="menu">
|
|
92
|
-
<video
|
|
93
|
-
|
|
94
|
-
autoplay muted loop playsinline></video>
|
|
138
|
+
<!-- seamlessLoop() (above) injects two crossfading <video>s here -->
|
|
139
|
+
<div class="menu-bg" id="menu-bg"></div>
|
|
95
140
|
<div class="menu-title stagger" style="--i: 0">EMBERFALL</div>
|
|
96
141
|
<button class="menu-btn stagger" id="menu-play" style="--i: 1; left: 38%; top: 52%; width: 24%; height: 8%;">PLAY</button>
|
|
97
142
|
<button class="menu-btn stagger" id="menu-options" style="--i: 2; left: 38%; top: 63%; width: 24%; height: 8%;">OPTIONS</button>
|
|
@@ -104,7 +149,9 @@ copy-pasteable example:
|
|
|
104
149
|
|
|
105
150
|
```css
|
|
106
151
|
#screen-menu { position: absolute; inset: 0; overflow: hidden; background: #000; }
|
|
107
|
-
.menu-bg { position: absolute; inset: 0;
|
|
152
|
+
.menu-bg { position: absolute; inset: 0; }
|
|
153
|
+
.menu-bg video { position: absolute; inset: 0; width: 100%; height: 100%;
|
|
154
|
+
object-fit: cover; opacity: 0; }
|
|
108
155
|
.menu-title {
|
|
109
156
|
position: absolute; left: 0; right: 0; top: 14%; text-align: center;
|
|
110
157
|
font-family: "Cinzel", serif; font-size: 9vh; color: #f4e3b8;
|
|
@@ -39,6 +39,15 @@ backdrop), not an afterthought `<div>`. When the menu decision is a reasoned
|
|
|
39
39
|
"no", every "menu backdrop" row above (pause, fail, win, lobby) grades over
|
|
40
40
|
the loader's key art instead.
|
|
41
41
|
|
|
42
|
+
The table lists screens; **elements are inventoried separately**. Walk the
|
|
43
|
+
whole loop in your head — loader → menu → spawn → action → pickup → damage →
|
|
44
|
+
death → retry → win — and write down EVERY on-screen element the player will
|
|
45
|
+
ever see: the reticle and each of its states, aim/interact cues, toasts,
|
|
46
|
+
damage numbers, kill feeds, timers, countdowns, pickup popups… Each element
|
|
47
|
+
gets a treatment from the brief. **Nothing ships looking like default
|
|
48
|
+
browser CSS** — an element styled like a bare `<div>` or a system button is
|
|
49
|
+
a bug wherever it appears, listed or not.
|
|
50
|
+
|
|
42
51
|
**2. One shared style brief — for the WHOLE game, not just the UI.** Write it
|
|
43
52
|
once — 4–5 named hues, materials, one display + one body font, mood — and
|
|
44
53
|
store it as a comment block near the UI code. `$genex-ai-menu`,
|
|
@@ -58,6 +67,13 @@ AAA games of the genre, and state in one line which structural conventions
|
|
|
58
67
|
you're borrowing. The bar is what those games ship, not "good enough for a
|
|
59
68
|
demo". Conventions only — never logos, exact layouts, or trade dress. The
|
|
60
69
|
same 2–3 names carry into the router's visual-direction gate for the scene.
|
|
70
|
+
Then turn references + brief into ONE **concept frame** — a single generated
|
|
71
|
+
image of what THIS game should look like mid-play (scene, mood, lighting;
|
|
72
|
+
a UI impression is welcome but optional):
|
|
73
|
+
`npx genex image "<one-sentence game look from the brief>" --aspect 16:9 --quality high --no-wait`,
|
|
74
|
+
enqueued FIRST of all art. Its URL goes into the style-brief comment as the
|
|
75
|
+
game's north star: every later 2D generation that accepts `--edit` anchors
|
|
76
|
+
to it, and the scene's lighting/grading is judged against it.
|
|
61
77
|
|
|
62
78
|
**4. Ask only when genuinely ambiguous.** If the concept pins the mood (a
|
|
63
79
|
"gothic horror dungeon crawler" pins it), decide and state the plan in one
|
|
@@ -70,12 +86,21 @@ it derives from the game type.
|
|
|
70
86
|
are examples, not defaults. Do not default to neon/cyberpunk/synthwave — or
|
|
71
87
|
any other single register — unless the concept calls for it.
|
|
72
88
|
|
|
73
|
-
**6. Close the gate out loud —
|
|
74
|
-
|
|
75
|
-
|
|
76
|
-
|
|
77
|
-
|
|
78
|
-
|
|
89
|
+
**6. Close the gate out loud — a fixed-format plan message.** Post this
|
|
90
|
+
message in chat, translated to the user's language, filling every line —
|
|
91
|
+
this exact structure, not a paraphrase (the scaffold's keep-it-short talk
|
|
92
|
+
rule explicitly does not apply to this one message):
|
|
93
|
+
|
|
94
|
+
```
|
|
95
|
+
UI plan
|
|
96
|
+
• Screens: <list with tiers — loader / menu / pause (Esc) / win/lose / …>
|
|
97
|
+
• Style: <4–5 named hues> · fonts <display> / <body>
|
|
98
|
+
• References: <2–3 AAA games — one line on what's borrowed>
|
|
99
|
+
• Concept frame: <generation id>
|
|
100
|
+
• Queued: HUD chrome <id> · menu video <id or "no menu: reason">
|
|
101
|
+
```
|
|
102
|
+
|
|
103
|
+
A message missing any line means the gate did not run — go back and run it.
|
|
79
104
|
|
|
80
105
|
## The tier ladder
|
|
81
106
|
|
|
@@ -268,6 +293,8 @@ architecture and consume the shared style brief.
|
|
|
268
293
|
- No pause screen, or a pause that isn't bound to Escape.
|
|
269
294
|
- A style brief whose fonts were never actually loaded (a system-stack display
|
|
270
295
|
font at runtime).
|
|
296
|
+
- A micro-element (reticle, cue, toast, damage number) left as default CSS
|
|
297
|
+
while the panels got the art treatment.
|
|
271
298
|
- A generic stat dashboard (rows of labels + numbers) instead of a designed
|
|
272
299
|
HUD — pick the 2–3 numbers that matter and style them by hierarchy.
|
|
273
300
|
- Hard-cut phase swaps, a menu whose elements just appear, numbers that
|
|
@@ -71,7 +71,9 @@ stack this game ships (bloom/AO/grade — at least ONE built render-pass effect
|
|
|
71
71
|
for EVERY game, justified by the shared style brief; the UI vignette div or a
|
|
72
72
|
CSS filter on the canvas does not count, and a plan that "deliberately skips
|
|
73
73
|
everything" is the stock default, not a look; `$genex-threejs-image-pipeline`
|
|
74
|
-
owns ordering when 2+ compose),
|
|
74
|
+
owns ordering when 2+ compose), one named **ambient-motion loop** that keeps
|
|
75
|
+
the scene alive at rest (emissive pulse, shimmer, drifting dust — shader
|
|
76
|
+
work, zero generations), and the lighting/atmosphere mood from that
|
|
75
77
|
same brief. Planning is not building — effects still land last in the
|
|
76
78
|
execution order.
|
|
77
79
|
|
|
@@ -150,8 +152,10 @@ concept-driven — a richer first build beats a grey-box one.
|
|
|
150
152
|
floors from the routing-map's acceptance gate: the generated sprite HUD wired in
|
|
151
153
|
(not the CSS placeholder), pause on Escape, the branded loader with its key-art
|
|
152
154
|
background, the brief's fonts actually loaded, the renderer baseline + at least one
|
|
153
|
-
built render-pass effect,
|
|
154
|
-
situ
|
|
155
|
+
built render-pass effect, every placed 2D/media piece screenshot-verified in
|
|
156
|
+
situ, no UI element left as default browser CSS, and no placeholder primitive
|
|
157
|
+
left where a generated asset was planned. These floors gate `npx genex publish`
|
|
158
|
+
and the final handoff of a session the
|
|
155
159
|
same way — they do not wait for the word "done" to be said.
|
|
156
160
|
- Entering an **existing or remixed project**: read before writing — learn the
|
|
157
161
|
current renderer choice, physics setup, and file conventions first, then
|
|
@@ -24,10 +24,12 @@ Three.js release or branch, and do not blindly copy demo architecture.
|
|
|
24
24
|
one shared style brief, a tier per screen — stated visibly in chat, never
|
|
25
25
|
decided silently. Enqueue the Tier-2 UI generations now with `--no-wait`
|
|
26
26
|
(pick them up later with `npx genex wait <id>`) — they render server-side
|
|
27
|
-
while you build the game.
|
|
28
|
-
|
|
29
|
-
|
|
30
|
-
|
|
27
|
+
while you build the game. FIRST of all art: the gate's **concept frame** —
|
|
28
|
+
the single image of the finished game's look that every later `--edit`-able
|
|
29
|
+
generation anchors to. Then the `$genex-ai-hud` Stage-1 mockup, enqueued
|
|
30
|
+
here for EVERY game; and the `$genex-ai-menu` video whenever the menu
|
|
31
|
+
decision is yes (the default for every game — "it's only a draft" is not a
|
|
32
|
+
reason to decide no). Skipping this enqueue is
|
|
31
33
|
the #1 way a finished game ships an ugly HUD — by step 12 there is nothing
|
|
32
34
|
to swap in.
|
|
33
35
|
4. Lock the visual direction — the same plan-first logic as the UI gate, in
|
|
@@ -47,7 +49,14 @@ Three.js release or branch, and do not blindly copy demo architecture.
|
|
|
47
49
|
owns the pass ordering;
|
|
48
50
|
- **references**: name 2–3 AAA games whose look this game borrows
|
|
49
51
|
(conventions, lighting mood, palette, post — never trade dress); the
|
|
50
|
-
same 2–3 the UI gate named, extended from the interface to the scene
|
|
52
|
+
same 2–3 the UI gate named, extended from the interface to the scene,
|
|
53
|
+
and embodied in the gate's concept frame — the image the scene is
|
|
54
|
+
judged against;
|
|
55
|
+
- **ambient motion**: name ONE subtle environmental motion loop that keeps
|
|
56
|
+
the scene alive at rest — an emissive pulse along edges, heat shimmer,
|
|
57
|
+
drifting dust, a slowly flowing texture. Shader/procedural, zero
|
|
58
|
+
generations, built with the scene — a world that is perfectly still
|
|
59
|
+
reads as a screenshot, not a place;
|
|
51
60
|
- **lighting/atmosphere mood** from the SAME shared style brief the UI gate
|
|
52
61
|
wrote — one art direction across scene and UI.
|
|
53
62
|
Planning is not building: effects still land LAST (steps 10–11); this step
|
|
@@ -61,15 +70,16 @@ Three.js release or branch, and do not blindly copy demo architecture.
|
|
|
61
70
|
architecture, planets, water, precipitation, clouds, or VFX. Show the
|
|
62
71
|
planned loader from the very first asset load — a player must never stare
|
|
63
72
|
at a black screen; the rest of the UI states come at step 12.
|
|
64
|
-
Then
|
|
65
|
-
|
|
66
|
-
screens
|
|
67
|
-
that should move)
|
|
68
|
-
|
|
69
|
-
|
|
70
|
-
|
|
71
|
-
|
|
72
|
-
|
|
73
|
+
Then make the world-dressing DECISION — a judgment call, not a quota:
|
|
74
|
+
does THIS world's fiction support 2D art (posters, signs, graffiti,
|
|
75
|
+
decals, banners, in-world screens — `$genex-ai-image`; `$genex-ai-video`
|
|
76
|
+
for anything that should move)? Say the decision in one line in the
|
|
77
|
+
visual-direction block. Name only the pieces that genuinely belong — a
|
|
78
|
+
garage wants grease posters and warning decals; a pristine void wants
|
|
79
|
+
none — and anchor their prompts to the concept frame. **Zero is a valid
|
|
80
|
+
answer with a stated reason; art placed just to satisfy this step is the
|
|
81
|
+
failure mode, not skipping it.** Enqueue what you named with `--no-wait`
|
|
82
|
+
alongside the other assets. Every placed piece is
|
|
73
83
|
screenshot-checked in situ — right scale, not stretched, unlit material
|
|
74
84
|
where it must glow, readable at gameplay distance
|
|
75
85
|
(`$genex-threejs-visual-validation` owns the capture discipline).
|
|
@@ -89,7 +99,10 @@ Three.js release or branch, and do not blindly copy demo architecture.
|
|
|
89
99
|
`$genex-threejs-game-ui` (HUD, pause on Escape, fail/retry, win, and the
|
|
90
100
|
full loading state grown from the step-6 loader), then `npx genex wait`
|
|
91
101
|
the step-3 UI generations and wire them in — the sprite HUD replaces the
|
|
92
|
-
placeholder CSS, the menu video replaces the still frame
|
|
102
|
+
placeholder CSS, the menu video replaces the still frame (wired with the
|
|
103
|
+
menu skill's loop crossfade), and the scene's planned generated models
|
|
104
|
+
replace their placeholder primitives — a shipped wave of enemies may not
|
|
105
|
+
be untextured boxes. This swap is a
|
|
93
106
|
completion gate, not an option: a game still on the placeholder CSS HUD,
|
|
94
107
|
or without a working Escape pause, is not done. Then run a feel pass via
|
|
95
108
|
`$genex-threejs-game-feel` (input response, camera, impact feedback,
|
|
@@ -97,6 +110,33 @@ Three.js release or branch, and do not blindly copy demo architecture.
|
|
|
97
110
|
13. Validate in a real browser with fixed seeds, captures, interaction checks,
|
|
98
111
|
and performance evidence.
|
|
99
112
|
|
|
113
|
+
## Parallelize with subagents (when the environment has them)
|
|
114
|
+
|
|
115
|
+
Generations already render server-side in parallel — the wall-clock savings
|
|
116
|
+
live in the ATTENDED chains. When the coding agent supports background
|
|
117
|
+
subagents (e.g. Claude Code's Agent tool), split these off; in environments
|
|
118
|
+
without them (Cursor), the same order simply runs sequentially and the
|
|
119
|
+
`--no-wait` pattern still hides most generation latency.
|
|
120
|
+
|
|
121
|
+
- **HUD chain worker.** The moment the Stage-1 mockup lands, hand the whole
|
|
122
|
+
`$genex-ai-hud` chain to one subagent: pick the better candidate → Stage-2
|
|
123
|
+
sheet → clean → extract → masks → write the sprite wiring. Its prompt must
|
|
124
|
+
be self-contained — the style brief, concept-frame and mockup URLs, the
|
|
125
|
+
output dir, and exactly which files it owns.
|
|
126
|
+
- **Asset shepherd.** One subagent runs `npx genex wait` on every queued ID
|
|
127
|
+
and wires each asset in as it lands (skybox, models, menu video, dressing),
|
|
128
|
+
swapping placeholders out and reporting each swap.
|
|
129
|
+
- **Verification runner.** Browser evidence — screenshots, control presses,
|
|
130
|
+
the menu loop-seam watch, in-situ checks of placed art — can run in a
|
|
131
|
+
subagent while the main agent keeps building.
|
|
132
|
+
|
|
133
|
+
Hard rules: **one writer per file** — each subagent owns a disjoint file set
|
|
134
|
+
(the HUD worker owns the hud modules and `public/assets/hud/`, the shepherd
|
|
135
|
+
owns asset-wiring modules), and the main agent stays the only writer of
|
|
136
|
+
gameplay/netcode. Merge points are explicit: a subagent reports what it
|
|
137
|
+
changed, the main agent reconciles. Never put two writers on the same
|
|
138
|
+
module — a corrupted entry point costs more than the minutes saved.
|
|
139
|
+
|
|
100
140
|
## Acceptance gate
|
|
101
141
|
|
|
102
142
|
**Game fast path:** for a game task that loaded no procedural/visual-system skill,
|
|
@@ -107,7 +147,9 @@ sprite HUD wired in — not the CSS placeholder — pause on Escape, the branded
|
|
|
107
147
|
loader with its key-art background, the brief's font pair actually loaded)
|
|
108
148
|
and the look floor (the step-4 renderer baseline + named post stack actually
|
|
109
149
|
built; every placed 2D/media piece — decals, posters, in-world screens —
|
|
110
|
-
screenshot-verified in situ
|
|
150
|
+
screenshot-verified in situ; no UI element left as default browser CSS; no
|
|
151
|
+
placeholder primitive left where a generated asset was
|
|
152
|
+
planned)**. These floors gate `npx genex publish` and the
|
|
111
153
|
final handoff of a session the same way — "I never said it was done" is not
|
|
112
154
|
an exemption. The list below applies to routed *visual-system* scenes, and each
|
|
113
155
|
system-specific item (debug views, seed manifests, tier knobs) applies only when
|