跳过主要内容
B3D
Starter Kit免费

ForgeSkills

20 gated game-development agent skills (orchestration, verification, Blender headless, procedural modeling, textures, animation, VFX, Godot and Unity pipelines) plus Python tools that measure the results. Apache-2.0.

GitHub 星数
9
(GitHub 星数,统计于 2026-10-24)
许可证
Apache-2.0
最近提交
2026/10/23
作者
sekaikx
下载 ZIP

信息来自作者的 README。 GameSkill 还没完整跑过。支持的 Agent 和安装步骤以作者文档为准。

安装

Claude Code (user level)
git clone https://github.com/sekaikx/ForgeSkills.git
cd ForgeSkills
cp -r skills/gamedev-* ~/.claude/skills/
Cursor and other agents (project level)
cp -r ../ForgeSkills/skills/gamedev-* .agents/skills/

Cursor also reads .claude/skills/, so the Claude Code project install works unchanged.

命令照搬项目 README。在本机运行前,请先看一眼仓库。

包含内容

  • Workflow core: gamedev-orchestrator, -verification, -visual-iteration, -debugging, -game-testing
  • Assets: gamedev-reference-reconstruction, -blender-headless, -procedural-modeling, -modular-kit, -texture-authoring
  • Motion and feel: gamedev-animation, -locomotion, -vfx-impact
  • World: gamedev-lighting, -camera, -world-from-data
  • Systems: gamedev-gameplay-systems, -performance, -godot-pipeline, -unity-pipeline
  • Python tools: reference-vs-render comparison, texture map derivation, tiling seam check, texel density, Blender build/render helpers
  • A 58-entry technique knowledge base and tools/validate_skills.py

适合

  • Agents that drive Blender and need measurable 'done' criteria
  • Rebuilding a concept sheet as a real asset with a discrepancy ledger
  • Fixing weightless combat, floaty movement or flat lighting

不适合

  • Proven Godot/Unity automation: the README says those two skills were written but not executed
  • Chatting about game design; it is built for agents that write code and ship assets

概述

ForgeSkills packs 20 agent skills around one loop: measure → build → render → compare → fix one thing → repeat. gamedev-orchestrator turns a one-line brief into a six-line contract and refuses to start while any line is blank; gamedev-verification catches the ways AI game work reports false success.

What sets it apart is the tooling. Instead of asking the agent to judge its own render, Python scripts compare reference and render, derive PBR maps from one albedo, detect tiling seams and measure texel density, and Blender helpers script builds and renders.

It is a brand-new, low-star repo (one commit on 10-23), but it is unusually honest: the README states which skills were executed, which were only written, and where its own test failed.

环境要求

  • Python 3.10+ with Pillow and numpy for the measuring tools
  • Blender 4.2+ or 5.x for tools/blender
  • Godot or Unity only if you use those pipeline skills

相关主题

常见问题

搭配使用

作者标注的相关 Skills,或更适合相邻任务的 Skills。

相似 skills