Field guides · Setup
Configure your coding agent
Every agent reads a global memory file before it does anything (~/.claude/CLAUDE.md, AGENTS.md, and friends), and that file shapes how it responds to you. Ours took months of trial and error. Here's the distilled version so your first session starts where our hundredth left off.
Get it
Download the starter file and drop it in ~/.claude/CLAUDE.md (or your agent's equivalent). Then read it once, top to bottom. It's short.
Source of truth: creative2/ai-skill-exchange on Bitbucket, under global/. Found a better rule? PR it there and everyone gets it. While you're there, browse the shared skills in the gallery.
What's inside
Safety rails. The rules that prevent the expensive accidents: a list of database commands the agent must never run without your explicit say-so (looking at you, migrate:fresh), test-isolation checks so a test suite can't wipe your dev data, and hard protection for .env files, which are un-backed-up secrets you cannot get back.
Framework conventions. Tailwind and Laravel rules that keep agent output matching how we actually build: no dynamically assembled class names, utilities over custom CSS, Laravel Boost for version-aware answers, Filament icon checks.
Tooling habits. Our commit message format, always-venv Python, and when to reach for curl versus a page-reading tool.
Working style. One well-worn setup: terse answers, take stated facts at face value, no speculative pre-checking, debug forward. This section is taste. Keep what fits, rewrite what doesn't — the point is that you decide once, in a file, instead of correcting the agent every session.