Opinionated guidelines for collaborating with LLMs on software development.
A set of instructions you paste into your LLM's custom instructions or system prompt. It shapes how the LLM behaves when helping you write code.
These guidelines prioritize:
- Thinking before coding — discuss approaches before generating code
- Simplicity — avoid clever solutions and over-engineering
- Collaboration — the LLM asks before acting, offers alternatives, and waits for approval
Copy the contents of llm_code_collaboration_guidelines.md into:
- Claude: Project Instructions (in a Project) or Custom Instructions (in Settings)
- ChatGPT: Custom Instructions or GPT system prompt
- Gemini: Custom Instructions
- Other LLMs: System prompt or equivalent
Customize as needed. These are starting points, not rules.
This is how I like to work with LLMs. You may disagree with some choices. Fork it, modify it, make it yours.
Some opinions expressed:
- Imperative over OOP (usually)
- Over-comment rather than under-comment
- Exit loudly on errors rather than catch silently
- YAGNI over speculative architecture
If these don't fit your style, change them.
CC BY-SA 4.0 — Use it, modify it, share it, attribute it.