Heathen Engineering

Working with Claude Code and AI Coding Agents

Introduction #

Coding agents like Claude Code, Cursor, and GitHub Copilot are now a normal part of how a lot of developers work, and Heathen’s tools are built with that in mind. Every Foundation and Toolkit ships a SKILL.md file (see the companion article, Heathen’s SKILL.md Files) specifically so an agent working in your project can ramp up on the tool it’s looking at without you having to explain it by hand.

Best Practices #

  • Point your agent at the SKILL.md file explicitly. Don’t assume it’ll find it on its own, tell it to read the file at the root of the package or Gem first.
  • Check your license terms before using paid Toolkit source with an AI tool. Our license doesn’t bar AI use, but it does limit it: you can’t use an AI tool in a way that leaks our IP to a non-licensed party, including the AI platform holder itself. Most tools these days let you configure what they spy on, sorry, “share”, so make sure that’s actually turned off. See the section below.
  • Let the agent read the actual shipped source, not just the KB. A Foundation’s source is real and complete; the KB is orientation, the code is the ground truth.

What a SKILL.md file gives your agent #

A dense, self-contained reference dropped next to the code your agent is already looking at: what the package does, its real API surface, how it’s structured, and pointers to more detail if it needs it. It’s written to be token-efficient for an agent to read, not marketing copy, and it ships inside the installed package itself (the UPM package folder in Unity, the copied plugin folder in Unreal, res://addons in Godot, the Gem folder in O3DE), so it’s there even for an agent with no web access working entirely inside your IDE.

AI use and your license, read this before you paste in paid source #

If you’re using a paid Toolkit (not a free FOSS Foundation), the Heathen Standard License has real, specific terms around AI use, not a vague “use responsibly” clause. In short: you may use the Licensed Technology with an AI system, including a third-party or cloud-hosted one, only if that AI service is configured (contractually and technically) so that it doesn’t retain your inputs/outputs beyond serving your immediate request, doesn’t use them for model training, and doesn’t expose them to any third party beyond what’s strictly needed to run the service. Consumer or free-tier AI products (free-tier ChatGPT, GitHub Copilot Individual, and similar) don’t meet this bar and are explicitly called out as prohibited, regardless of how reasonable your own use feels. The license also separately prohibits using an AI system to reverse-engineer or reconstruct the Licensed Technology as a way of getting around the license itself.

This isn’t unique to Heathen, it reflects a real, current risk with cloud AI tools generally: your code can end up retained or used for training somewhere you didn’t intend. Check your AI tool’s actual data-handling terms (enterprise/business tier, zero-retention agreement, or equivalent) before pointing it at paid source, not just its marketing page.

A practical starting point #

If you’re setting an agent up in a new project that uses Heathen tools: install the package normally, have your agent read the SKILL.md file(s) that came with it, and let it explore the real source from there. For anything the SKILL.md doesn’t cover, the Knowledge Base is the next place to point it, most articles are written clearly enough for an agent to parse directly.

Rate This Article!

Contents