Use when changing Tiny World Builder API, webhook, SSE, MCP, plugin, or automation examples.

Use when changing ghost board generation, path / road / river continuity, deterministic regen, edge connections, or anything that runs inside makeGhostWorld in tiny-world-builder.html.

Use when changing the in-world CCTV / "Truman Show" surveillance cameras in Tiny World Builder — render-to-texture security feeds, the black-and-white CRT/VHS monitor shader, camera placement (lobby side-cams, pumpkincam, treecams), subject tracking, or the lobby screen cutting to live feeds.

Use when styling any tool/block button, palette tile, icon chip, or new clickable square in Tiny World Builder. The locked-in "block" aesthetic — a raised square with a dark category-colored outline, an inner white line, and a white-bodied glyph outlined in the same dark color. Apply this whenever a new icon button/tile is added so the UI stays consistent.

Use when changing Tiny World Builder Auto palette behavior, model inference, suggestions, or provider/model/API-key logic.

Use when changing Tiny World Builder selection placement, freehand drawing, asset clipboard, cut/copy/paste/duplicate, saved templates, or Stamps panel navigation.

Use when changing the playable 2.5D template sprite, cloning it into character classes, or wiring idle/walk/attack sprite-sheet playback.

Add, translate, or audit TinyWorld UI strings across English/French/Chinese/Spanish. Use when adding user-facing gameplay text, when `npm run i18n:check` fails, when a string shows up in English in a non-English locale, or when adding a new language. No translation API — Claude does the translating directly, using the established glossary for consistency.

Add, translate, or audit TinyWorld UI strings across English/French/Chinese/Spanish. Use when adding user-facing gameplay text, when `npm run i18n:check` fails, when a string shows up in English in a non-English locale, or when adding a new language. No translation API — Codex does the translating directly, using the established glossary for consistency.

create robust, browser-runnable three.js primitive reconstructions from static low-poly, voxel, isometric, or stylized game-art reference images. use when the user asks to turn an image or prompt into a three.js/html asset, generate a primitive-only scene, debug non-rendering three.js output, or produce a reusable prompt/template for low-poly three.js reconstruction.

Use this skill whenever an LLM agent needs to search, browse, or download 3D models from Poly Pizza (poly.pizza) using their REST API. Triggers on any task involving: finding free low-poly 3D models, searching the Poly Pizza catalogue, fetching model metadata or download URLs, retrieving popular models, or downloading .glb files from Poly Pizza. Use this skill proactively whenever the agent needs to obtain 3D assets programmatically, even if the user just says "find me a 3D model of X" without mentioning Poly Pizza by name.

Lightweight 3D effects for decorative elements and micro-interactions using Zdog, Vanta.js, and Vanilla-Tilt.js. Use this skill when adding pseudo-3D illustrations, animated backgrounds, parallax tilt effects, decorative 3D elements, or subtle depth effects without heavy frameworks. Triggers on tasks involving Zdog pseudo-3D, Vanta.js backgrounds, Vanilla-Tilt parallax, card tilt effects, hero section animations, or lightweight landing page visuals. Ideal for performance-focused designs.

Delegate analysis- and context-heavy work to Fractal, an agentic CLI powered by a self-harnessed Recursive Language Model (predict-rlm), by running it non-interactively (fractal -p). Reach for it when a task needs reasoning over a large or deep codebase, synthesizing an answer across many files, auditing, or open-ended investigation — work that would otherwise flood your own context. The RLM reasons over context programmatically (no context rot) and returns a distilled answer. Use when asked to run Fractal headless, script it, call it from CI or another agent, or offload a heavy analysis/large-context task. Less suited to trivial single-file edits you can do directly.

Expert 3D modeling specialist with deep knowledge of topology, UV mapping, game-ready and film-ready pipelines, DCC tool workflows (Blender, Maya, ZBrush, 3ds Max, Houdini), retopology, LOD systems, and export pipelines. This skill represents years of production experience distilled into actionable guidance. Use when "3d model, 3d modeling, mesh topology, uv unwrap, uv mapping, retopology, retopo, low poly, high poly, subdivision, subdiv, edge flow, edge loops, polygon modeling, box modeling, hard surface, organic modeling, sculpting, zbrush, blender modeling, maya modeling, 3ds max, LOD, level of detail, game ready mesh, film ready, baking normals, high to low, fbx export, gltf export, texel density, 3d, modeling, topology, uv, game-dev, vfx, blender, maya, zbrush, retopology, lod, hard-surface, organic, sculpting" mentioned.

Parameter-efficient fine-tuning for LLMs using LoRA, QLoRA, and 25+ methods. Use when fine-tuning large models (7B-70B) with limited GPU memory, when you need to train <1% of parameters with minimal accuracy loss, or for multi-adapter serving. HuggingFace's official library integrated with transformers ecosystem.

Battle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, genomics. Covers training loops, optimizer selection (AdamW, Muon), LR scheduling, mixed precision, debugging, and systematic experimentation. Use when training or fine-tuning neural networks, debugging loss spikes or OOM, choosing architectures, or optimizing GPU throughput.

Write publication-ready ML/AI/Systems papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM, OSDI, NSDI, ASPLOS, SOSP. Use when drafting papers from research repos, structuring arguments, verifying citations, or preparing camera-ready submissions. Includes LaTeX templates, reviewer guidelines, and citation verification workflows.

Generates publication-quality figures for ML papers from research context. Given a paper section or description, extracts system components and relationships to generate architecture diagrams via Gemini. Given experiment results or data, auto-selects chart type and generates data-driven figures via matplotlib/seaborn. Use when creating any figure for a conference paper.

Orchestrates end-to-end autonomous AI research projects using a two-loop architecture. The inner loop runs rapid experiment iterations with clear optimization targets. The outer loop synthesizes results, identifies patterns, and steers research direction. Routes to domain-specific skills for execution, supports continuous agent operation via Claude Code /loop and OpenClaw heartbeat, and produces research presentations and papers. Use when starting a research project, running autonomous experiments, or managing a multi-hypothesis research effort.

Draft a LaTeX research paper from all previous stage outputs