6d3bac8d41
Playtest ask: "more detailed looking enemies... we have local trellis-2 and hunyuan, we can make
actual shaped pathogens". Two landed, keyed by ARCHETYPE so one mesh skins every instance of
that behaviour with no code change:
latcher -> Giardia trophozoite: teardrop body, broad front, ventral suction disc. 2232 tris.
Replaces a tetrahedron, and it is the one that matters most — a latcher ATTACHES
to your hull, so it is the only enemy the player ever sees close up.
driller -> H. pylori: thick corkscrew rod. 3231 tris. Replaces a cone.
Both prompts deliberately ask for BODIES and no flagella: the phage taught us image->3D webs
thin appendages into a lump, and at flat-emissive gameplay scale only the silhouette ships.
It worked — bbox aspect 3.80 (teardrop) and 2.12 (rod) against the phage's near-cubic 1.27:1.41:
0.94 failure, so these are genuinely shaped rather than blobs.
THE PIPELINE FIX that made them usable, worth recording: collapse-decimate FLOORS OUT on Trellis
output. Asking for 2500 tris gave 15871 and iterating moved it not at all; welding first made it
worse (12042 -> 78663); triangulating changed nothing. The answer is VOXEL REMESH first — it
rebuilds the surface as clean manifold topology, which is the right tool for a blobby organic and
finally gives decimate something it can collapse. 15439 -> 2232 and 78663 -> 3231, shapes intact,
4.4MB -> 91KB. Any future pathogen goes through the same path.
Not landed: seeker and turret. MODELBEAST returned 401 then 429 on both across two runs — the
guest token's active-job cap, not a content failure. The script is resumable; they need a quieter
window on the farm.
Honest limit, so nobody expects more than this delivers: at gameplay distance under flat emissive
these read mostly as silhouette, so the gain over the old primitives is real but modest. The place
it should actually show is a Giardia clamped to the hull. If we want enemies to read as DETAILED
rather than merely better-shaped, the lever is the rendering (a matcap or rim treatment on
entities), not more geometry — that is an ART_BIBLE decision, not a pipeline one.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|---|---|---|
| .claude | ||
| docs | ||
| pipeline | ||
| tools | ||
| web | ||
| .gitignore | ||
| README.md | ||
GUTS — a fantastic voyage through the alimentary canal
Browser flight-combat game (three.js, zero build step). You pilot ENDO-1, a microscopic probe, mouth → rectum through five anatomically-grounded biomes: ride peristalsis down the esophagus, cross the acid sea of the stomach, thread the villi forest of the small intestine, survive the gas swamps of the colon. Fantastic Voyage × Star Fox × Descent.
- Design: docs/GDD.md · Look: docs/ART_BIBLE.md
- Architecture + contracts: docs/TECH.md
- Asset generation: docs/PIPELINE.md
- How we work (lanes/rounds): docs/PROCESS.md — read this first if you are a lane
- Current round: docs/LANES/ROUND1_INSTRUCTIONS.md
Run it
cd web && python3 -m http.server 8140
# http://localhost:8140 — the game
# http://localhost:8140/?stub=1&dbg=1 — stub world + debug HUD (works from day 0)
No bundler, no npm, no build. ES modules + import map; three.js r175 vendored in web/vendor/
(never CDN — the game must run offline). Before you finish a session: bash tools/qa.sh.
The one-paragraph tech pitch
The whole canal is a spline. World position is (s, θ, ρ) — distance along the canal,
angle and radius in the cross-section. Tube walls are chunked extrusions along
parallel-transport frames; peristalsis and breathing are vertex-shader displacement
(cheap, squishy, everywhere); collision is an analytic radius test (no physics engine).
Caverns (mouth, stomach) switch the same controller to free 6DOF. Look is synthetic
scanner — dark endoscope void, SEM-style monochrome walls with FLUX-generated detail
textures, emissive-coded enemies. Everything is seeded + deterministic, and every asset is
optional at runtime — the game boots and plays asset-free with procedural fallbacks.
Hard laws (all lanes)
- Determinism. No
Math.randomoutsidejs/core/rng.js. Seeded RNG only. qa.sh greps. - Assets are optional. Missing texture/model/audio ⇒ procedural fallback, never a crash.
- Eyeball everything. Visual work isn't done until a screenshot is in
docs/shots/laneX/. - Perf budget: ≤300 draw calls, 60 fps on M-series / mid laptop. Measure via
window.DBG. - Free-first. Local MODELBEAST generation before any paid API; fal.ai is a gated fallback (John's go). See PIPELINE.md.
- Git: stage only your exact paths, commit + push each session, never
git add ..
Deploy
Live: https://partly.party/gutsy/ — bash tools/deploy.sh (QA gate → rsync web/ minus
dev/ → docker cp into forum-nginx → verify). No build step; web/ ships as-is.
Two things to know before you touch it, both documented at length in the script header:
- It's ephemeral. Content lives in the container's own layer (like blobbo/glytch), so a
docker compose up --force-recreateor nginx image bump wipes it — re-run the script. The durable fix is a bind mount inforum/docker-compose.yml(what cratewars/roguelike use), which costs a forum-nginx recreate and brief partly.party downtime. - A 404 there returns HTTP 200. nginx's catch-all
try_files ... /index.htmlserves the arcade index for any missing file. Verify on content, never on status codes.
Origin: ssh://git@100.71.119.27:222/monster/guts.git.