threejs-game-skills
AGENTS.md0 starsChanged 5 months agoMITView on GitHub
This repo contains Codex workflow assets for Three.js browser-game development.
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository9 other files it gives its agents.
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
23 rules. About 0 are common, 9 are distinctive, and 14 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
Distinctive · estimate
This author's own choices. Worth a look.
14
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 7
Prefer TypeScript, Vite, npm package imports, and Three.js modules.
- Very specific: line 8
Use three/addons/... for official controls, loaders, and post-processing helpers.
- Distinctive: line 9
Use simple custom collision for arcade triggers and prototype-scale interactions.
- Very specific: line 10
Use Rapier as the default robust physics engine for serious Three.js browser games with rigid bodies, sensors, balls, ramps, moving …
- Very specific: line 11
Use cannon-es only as a lightweight JavaScript fallback for small/simple rigid-body scenes when avoiding WASM matters.
- Very specific: line 13
Use lil-gui for local tuning when it materially speeds iteration.
- Very specific: line 14
Use a lightweight HUD or stats.js for frame diagnostics when performance is in scope.
- Very specific: line 15
Treat WebGPU as conditional.
- Very specific: line 24
When shell tools are available, run skills/threejs-game-director/scripts/audit_reference_report.py --premium <report.md> against the final …
- Distinctive: line 26
Do not stop at first playable slice when the user asked for premium, AAA, polished, complete, release-ready, or showcase quality.
- Distinctive: line 27
Keep scene setup, loop, input, systems, entities, UI, assets, and debug tools separated once the prototype grows beyond a single simple …
- Distinctive: line 30
Avoid decorative-only UI, generic purple gradients, particle clutter, and unchecked post-processing.
- Distinctive: line 31
Keep mobile input and resize behavior in the first implementation path, not as a final afterthought.
- Distinctive: line 32
Use screenshot-based art-direction critique when visual priorities are unclear.
- Very specific: line 33
Use threejs-aaa-graphics-builder when screenshots still look basic across multiple visual surfaces.
- Distinctive: line 34
Use focused UI passes for HUDs, menus, text fit, icon controls, and safe-area layout instead of treating interface craft as generic polish.
- Distinctive: line 35
Use focused procedural-model passes for scratch-built assets, prop kits, material variation, instancing, LOD, and renderer-count-aware …
- Very specific: line 37
Run the director credential probe before claiming 3D/image/audio generation credentials are unavailable.
- Very specific: line 40
Use threejs-3d-generator for high-value external AI-generated 3D assets, auto-rigging, animation, texture, conversion, and GLB/FBX game …
- Very specific: line 41
Use threejs-image-generator before threejs-3d-generator when 2D concept/reference images, T-pose/A-pose character sheets, texture …
- Very specific: line 42
Use threejs-audio-generator for game SFX, ambience loops, UI sounds, announcer/dialogue, voice conversion, and audio cleanup.
- Distinctive: line 43
Treat generic stat-card HUDs, cube obstacles, and skyline boxes as prototype placeholders unless the user explicitly wants that style.
- Very specific: line 64
Use the scaffold's npm run verify:visual and npm run inspect:canvas when available, or …
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Agent Instructions
This repo contains Codex workflow assets for Three.js browser-game development.
Default Technical Stack
- Prefer TypeScript, Vite, npm package imports, and Three.js modules.
- Use
three/addons/...for official controls, loaders, and post-processing helpers. - Use simple custom collision for arcade triggers and prototype-scale interactions.
- Use Rapier as the default robust physics engine for serious Three.js browser games with rigid bodies, sensors, balls, ramps, moving platforms, many contacts, or high-speed collisions.
- Use
cannon-esonly as a lightweight JavaScript fallback for small/simple rigid-body scenes when avoiding WASM matters. - Load
threejs-gameplay-systems/references/physics-engine-selection.mdbefore adding or changing physics-heavy gameplay. - Use
lil-guifor local tuning when it materially speeds iteration. - Use a lightweight HUD or
stats.jsfor frame diagnostics when performance is in scope. - Treat WebGPU as conditional. Use Three.js
WebGPURendereronly when the project benefits from it and keep a WebGL/WebGL2 fallback path.
Game Quality Bar
- For broad requests to build, upgrade, polish, or finish a Three.js game, route through
threejs-game-directorfirst. The user should not have to name every specialist skill. - Public Three.js skills are consolidated around the director plus specialist systems: gameplay, AAA graphics, UI, debug/profile, QA/release, 3D generation, image generation, and audio generation.
- In Claude-style skill runners, do not assume a director skill can literally invoke other skills. The director must attempt to load sibling public
SKILL.mdfiles first, report a skill-loading ledger, and use the bundled phase OS only for files that cannot be loaded. - For broad/premium director work, the director must load each phase's required
references/*.mdfiles at phase entry and report a reference ledger. A phase is not done if its required references were skipped. - Premium/AAA/showcase claims must include the filled visual scorecard from
threejs-aaa-graphics-builder/references/visual-scorecard.md, including average score and automatic failures remaining. Do not substitute an improvised rubric. - When shell tools are available, run
skills/threejs-game-director/scripts/audit_reference_report.py --premium <report.md>against the final evidence report before claiming premium, AAA, showcase, complete, release-ready, or "less basic" success. - Build a playable loop first. A static scene is not done.
- Do not stop at first playable slice when the user asked for premium, AAA, polished, complete, release-ready, or showcase quality.
- Keep scene setup, loop, input, systems, entities, UI, assets, and debug tools separated once the prototype grows beyond a single simple file.
- Tune movement, camera, collisions, feedback, and HUD through short playtest loops.
- For physics-heavy games, report engine choice, fixed timestep, collider strategy, sensors, CCD use, body/collider counts, and restart cleanup evidence.
- Avoid decorative-only UI, generic purple gradients, particle clutter, and unchecked post-processing.
- Keep mobile input and resize behavior in the first implementation path, not as a final afterthought.
- Use screenshot-based art-direction critique when visual priorities are unclear.
- Use
threejs-aaa-graphics-builderwhen screenshots still look basic across multiple visual surfaces. Do not split that broad graphics problem into isolated small polish tasks. - Use focused UI passes for HUDs, menus, text fit, icon controls, and safe-area layout instead of treating interface craft as generic polish.
- Use focused procedural-model passes for scratch-built assets, prop kits, material variation, instancing, LOD, and renderer-count-aware detail.
- For premium/AAA/showcase/high-fidelity/less-basic work with characters, vehicles, ships, weapons, buildings, signature props, skies, textures, decals, logos, icons, GUI art, SFX, ambience, or voice, load
threejs-3d-generator,threejs-image-generator, and/orthreejs-audio-generatorbefore deciding procedural/generated assets are unnecessary. - Run the director credential probe before claiming 3D/image/audio generation credentials are unavailable. A missing key is not a valid blocker unless the SET/MISSING probe output is reported.
- Report an external asset sourcing ledger for premium graphics work: procedural / threejs-image-generator / threejs-3d-generator / hybrid per high-value surface, plus task IDs/output paths or real blocker evidence.
- For premium hero surfaces, procedural-only is not a valid final answer unless credential probe output or attempted generation shows a real blocker.
- Use
threejs-3d-generatorfor high-value external AI-generated 3D assets, auto-rigging, animation, texture, conversion, and GLB/FBX game asset workflows. - Use
threejs-image-generatorbeforethreejs-3d-generatorwhen 2D concept/reference images, T-pose/A-pose character sheets, texture references, decals, logos, icons, skies, backgrounds, or GUI art would improve the result. - Use
threejs-audio-generatorfor game SFX, ambience loops, UI sounds, announcer/dialogue, voice conversion, and audio cleanup. - Treat generic stat-card HUDs, cube obstacles, and skyline boxes as prototype placeholders unless the user explicitly wants that style.
- Require active-play screenshot scoring for premium/AAA claims. A build is not premium if the visual scorecard has any category below 2.
Verification Bar
For meaningful Three.js changes, gather evidence before claiming success:
npm run build- a local dev or preview server opened in a browser
- browser console and page error check
- Playwright screenshot
- canvas nonblank pixel check
- at least one desktop and one mobile viewport
- interaction check for the main control path
- performance snapshot when draw calls, asset counts, shaders, or post-processing changed
- UI text-fit, overlap, safe-area, and touch-target evidence when interface layout changed
- renderer diagnostics when procedural model fidelity, material count, or repeated props changed
- external asset sourcing ledger when premium graphics include high-value 3D or 2D asset categories
- credential probe output plus real external asset outputs or blocker evidence for premium asset-category claims
- visual scorecard before/after when the user asks for premium, AAA, showcase, or "less basic" graphics
Use the scaffold's npm run verify:visual and npm run inspect:canvas when available, or skills/threejs-qa-release/scripts/inspect-threejs-canvas.mjs from this repo.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

