Claude of Duty
Get updates here.
A first-person shooter built in the browser with Three.js r180 and WebGL2. Roughly 55k lines across 11 subsystems, written by a fleet of AI agents under orchestration.
There are no art assets. Every texture, mesh, animation and sound is generated
procedurally at load time from code. No models, no HDRIs, no image files, no audio
files. The only runtime dependency is three.
npm install
npm run dev # http://127.0.0.1:5173Click the canvas to lock the cursor. WASD move, mouse aim, LMB fire, RMB ADS, R reload, Shift sprint, Ctrl crouch, Space jump, Q/E lean, Esc release.
What's in it
| subsystem | what it does |
|---|---|
render |
HDR pipeline, cascaded shadow maps in a sampler2DArray with texel snapping and PCSS contact hardening, MRT depth/normal/velocity prepass, GTAO, TAA with YCoCg variance clipping, tile-dilated motion blur, Karis bloom pyramid, GPU EV100 metering, procedural 33³ grade LUT, AgX composite |
materials |
GPU texture forge: 19 procedural surfaces (concrete, brick, plaster, asphalt, sand, rusted/painted/brushed metal, wood, fabric, burlap, glass…), periodic noise so everything tiles seamlessly, Sobel height→normal, parallax occlusion mapping, triplanar projection, curvature-driven edge wear |
sky |
Atmospheric scattering, time of day, PMREM environment generation, volumetric fog and light shafts |
world |
~120×120 m market street: modular building kit with real wall thickness, enterable interiors, several hundred instanced props |
physics |
Written from scratch, no library. Binned-SAH BVH (29k tris → 14k nodes in 22 ms, 0.25 µs/raycast), swept-capsule character controller with a 5-plane crease stack, impulse rigid bodies with CCD, PBD ragdolls, multi-layer bullet penetration |
player |
Movement state machine, slide/mantle/lean, camera feel |
weapons |
Procedural weapon geometry, viewmodel rig, ADS, spring recoil, procedural reloads, ballistics with travel time and drop |
fx |
GPU particles, decals, tracers, muzzle flash, explosions |
ai |
Skinned soldiers, navmesh pathing, perception, cover behaviour, ragdoll death |
ui |
DOM/CSS HUD: crosshair, hitmarkers, minimap, compass, killfeed |
audio |
Web Audio synthesis — no sound files. Layered weapon fire, convolution reverb, HRTF spatialisation, occlusion |
ARCHITECTURE.md is the contract the agents worked against: subsystem interface,
directory ownership, the cross-subsystem event vocabulary, and shared surface types.
Tooling
The interesting part of this repo is arguably the harness, not the game.
| tool | purpose |
|---|---|
tools/capture.mjs |
Screenshot one named shot via GPU-backed headless Chromium |
tools/shotset.mjs |
All 11 shots in one session — fast review set |
tools/baseline.mjs |
Reproducible capture: each shot in an isolated page, fixed frame budget. Bit-identical across runs |
tools/imagediff.mjs |
Per-pixel gate. Exits non-zero if any pixel moved |
tools/profile.mjs |
Gameplay profiler at real device pixel ratio. Frame-time distribution and hitch attribution via per-frame WebGL program counts |
tools/playtest.mjs |
Scripted movement/fire smoke test |
Two findings worth recording, because both invalidated earlier measurements:
Median frame time hides the actual problem. A static-camera benchmark reported
94 fps while the game was unplayable. Real gameplay at Retina DPR (internal 3.34 MP,
not 2.07) ran 12–17 fps with 728–1236 ms stalls caused by 34+ WebGL programs
compiling lazily mid-frame. profile.mjs reports p50/p95/p99 and attributes each
hitch, which is what surfaced it.
Captures were not reproducible. shotset.mjs reuses one page across all 11
shots, so particle age, decal buffers and exposure state leak forward — two identical
runs differed on 10 of 11 shots. baseline.mjs isolates each shot in a fresh page,
which is bit-identical and is what makes imagediff.mjs a usable gate.
Performance
Measured on an Apple silicon laptop at 1512×982, DPR 2 (3.34 MP internal), ultra preset,
3 runs, gameplay in motion with AI and firing active:
| before optimization | after | |
|---|---|---|
| fps p50 | 12–17 | 28–30 |
| fps p99 | 4–9 | 14–17 |
| worst frame | 728–1236 ms | 66–82 ms |
| shader compiles during play | 34–35 | 0 |
| boot | ~9–12 s | 3.7–4.6 s |
The optimization pass was constrained to produce zero visual change, enforced by
imagediff.mjs rather than by assertion — the shipped build is bit-identical to its
pre-optimization reference across all 11 shots.
Shader pre-warm (src/core/prewarm.js) is what removed the stalls. Making it
provably pixel-neutral required first fixing subsystems that animated off
performance.now() instead of the engine clock, since any change to boot duration
otherwise shifted output.
Honest assessment
The goal was to match a modern Call of Duty. It does not.
Eleven independent adversarial critics scored the frames against that bar. Scores went 3.59 → 4.14 → 4.05 → 5.05 out of 10. Two shots reached "CLOSE"; the rest remain "AMATEUR". In a blind A/B, every critic in every round picked the real Call of Duty frame.
Where it falls short, specifically:
- Hands. Blocky finger slabs that don't convincingly grip the weapon.
- Material richness. Surfaces read as procedural noise rather than photographed reality at close range — the ceiling of generating texture from code.
- Characters. Enemies read as mannequins at distance.
- Indirect light. An approximation, not real GI.
- Frame rate. 28–30 fps at Retina. The art passes tripled geometry cost (5.9M → 11.3M triangles) and optimization recovered about half.
A known root cause remains unfixed: the viewmodel light rig in render/index.js
delivers roughly 20× the irradiance per unit albedo that the world does — a plain
black material in the view scene renders at L=110 against a background of 91,
purely from F0=0.04. Every weapon albedo is cheated to a third of physical to
compensate, which caps material separation on the most-looked-at object in the game.
Process note
Sequential single-owner passes beat parallel fan-out decisively. Three rounds of six agents each owning one directory moved the score +0.46 and left frame-ruining defects higher than they started (60 → 47 → 66), because tonemapping, sky and indirect light are one coupled system and isolated agents kept breaking each other's assumptions. One sequential pass with a single owner per coupled concern moved it +1.00 and cut defects 66 → 26.
The most valuable single result came from an agent contradicting its own brief. Every critic for three rounds reported the weapon as "untextured". It wasn't — it was specular-dominated, with the diffuse term measured at L=26 against a shipped L=67. Prior rounds had been crushing albedos to fight bright-part complaints, which killed diffuse and made it worse. The fix was the opposite of what was asked for.