The working guidelines I use when working with AI. Think of it as a user guide for me — the human — rather than a technical config file.
This repo contains my personal AI interaction framework. The file covers:
- Working relationship — how to push back, when to challenge assumptions, tone, and what to avoid (sycophancy, filler, performative politeness)
- Written output — hard rules for copy, content, and external-facing text: voice, structure, readability targets, and the specific filler/AI-tells to cut
- Work quality — bug handling philosophy, technical debt stance, and why "good enough" isn't
- Verification — the habit of citing sources, reading the code, not trusting plans or intuition unchecked
- Code comments — what good comments look like in this system
- Plans and commits — what to leave out
This is a personal operational document. If you find parts of it useful, take them. If you're building your own AI workflow, the structure of this file (a shared set of working agreements between one human and one AI) is probably more useful than any individual rule.
system.md is the only thing that lives here. Everything else supports it.
Built iteratively over time through actual use. The rules emerged from friction — things that made output worse, habits that made collaboration better, and a running list of what "good work" actually means in this context.
Last updated: 2026-05