skill
ai-coding-rules
Opinionated rules & prompts for every major AI coding assistant (Claude Code, Cursor, Copilot, Windsurf, Continue, Cody). Drop-in files that enforce plan, verify, review discipline.
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/HidayetCanOzcan/ai-coding-rules itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 23, 2026ai-coding-rules
Opinionated rules and prompts for every major AI coding assistant — drop one file into your repo and hold your assistant to the same disciplined workflow in every session.
AI coding tools are capable but eager. Left to their defaults they'll happily declare a change "done" the moment it typechecks, quietly widen the scope of a diff, or invent a config key that looks right. This repo is a set of small, ready-to-use rules files that hold them to a stricter standard — the same one across Claude Code, Cursor, Copilot, Cody, Windsurf, and Continue.
Every file here encodes one idea:
A change isn't done when it compiles. It's done when you've watched it do the right thing and a skeptic couldn't poke a hole in it.
That's it. Copy the file for your tool, and you're set.
Why rules matter
A model's default behavior drifts — between sessions, between tasks, and especially between tools. One day it writes tight, scoped diffs; the next it refactors three files you didn't ask about. A rules file is how you stop re-explaining your standards every single prompt.
- Consistency across sessions. The discipline is applied on every request, not just when you remember to ask for it.
- Consistency across tools. These files all encode the same five gates, so switching from Cursor to Claude Code doesn't mean re-teaching your expectations.
- Fewer bad surprises. Plan first, keep diffs small, actually verify, review like a skeptic, and report honestly — the failure modes that waste your afternoon get caught before they reach you.
- It's just a file. Version-controlled, reviewable, editable. Tune it to your stack and it travels with the repo for the whole team.
The five gates, in one breath: plan before editing → match the house style → verify by exercising the change → self-review as an adversary → report honestly what you did and didn't check.
What's inside
Each tool gets its own file in its own folder. Copy it to the path in the last column, at your project's root.
| Tool | File in this repo | Where it goes in your project |
|---|---|---|
| Claude Code | claude-code/CLAUDE.md | CLAUDE.md (repo root) |
| Cursor | cursor/.cursorrules | .cursorrules (repo root) — or .cursor/rules/*.mdc |
| GitHub Copilot | github-copilot/copilot-instructions.md | .github/copilot-instructions.md |
| Sourcegraph Cody | cody/cody-rules.md | .sourcegraph/engineering.rule.md |
| Windsurf | windsurf/.windsurfrules | .windsurfrules (repo root) |
| Continue | continue/continue-rules.md | .continue/rules/engineering-discipline.md |
Each file carries a placement note — explaining exactly where it goes and how that tool loads it — and marks where your stack, commands, and conventions go, so the project-specific bits win over the general discipline. (Some notes sit at the top, some in a comment at the bottom.) Tool rule systems evolve, so the notes point you at the current docs in case a path or format changes.
Quick start
# Example: Claude Code
cp claude-code/CLAUDE.md /path/to/your-project/CLAUDE.md
# Example: Cursor
cp cursor/.cursorrules /path/to/your-project/.cursorrules
Then open the copied file and replace the <PLACEHOLDER> markers with your real
commands (install, test, build) and any rules specific to your codebase. Commit
it. Done.
Prompt library
Rules set the default behavior. For steering a specific task, there are tool-agnostic, copy-paste prompts that pair with the same discipline.
The prompts/ folder has four focused files:
plan-a-change.md— force a plan and surface the riskiest assumption before any code is written.find-root-cause.md— diagnose the actual cause of a bug instead of patching the symptom.review-this-diff.md— get an adversarial review of a diff, not a rubber stamp.write-tests.md— cover the behavior and the failure paths, in your existing test style.
For a wider set, PROMPTS.md is a single-file library of a dozen
prompts covering the same ground plus root-causing bugs, safe refactors, security
hardening, dependency upgrades, and PR descriptions.
Paste any of them into any assistant, fill in the bracketed placeholders, and go.
Go deeper
These rules are intentionally small. If you want more depth — especially on Claude Code:
- Guide: claude-code-guide — a free walkthrough of getting real work done with Claude Code.
- Starter: claude-code-starter — a free, MIT-licensed project template to build on top of these rules.
If you end up wanting the full setup, there's a Claude Code Pro Pack with skills, subagents, and workflows — a paid kit for people who've outgrown a single rules file. Everything in this repo stays free and is enough on its own; the Pro Pack is there only if you want it.
Contributing
Improvements welcome — a sharper rule, a fix for a tool whose config format changed, a new tool folder that follows the same five-gate structure. Keep it practical and keep it honest; no hype, no filler. Open an issue or a PR.
License
MIT. Use it, fork it, adapt it for your team. No attribution required.