Five small rule files that tell your AI coding assistant to stop guessing, stay inside your architecture, keep changes small, verify its work, and protect your secrets.

What you get (13 files in one zip)

- 5 Cursor rules in .mdc format, for .cursor/rules/

- 1 CLAUDE.md with all five rules, for Claude Code

- 5 Claude Code rule files in .md format, for .claude/rules/ (an alternative to CLAUDE.md)

- README with install steps and a quick check that the rules are loaded

- CHANGELOG

The five rules

- No guessing: separate facts from assumptions, never invent APIs, say "not run" for anything not run.

- Architecture boundaries: match the existing structure, no new dependencies without asking.

- Small diffs: plan first when a change touches more than 2 files, one concern per change.

- Verify before done: run the project's own tests, lint and type check, and report exactly what ran.

- Secrets and safety: never print or commit secrets, ask before destructive actions.

Each rule is about 20 lines, with Do, Avoid and Why sections.

How to install

- Cursor: copy the five .mdc files into .cursor/rules/

- Claude Code: copy CLAUDE.md to your project root, or copy the five rule files into .claude/rules/. Use one option, not both.

- Fill in the Commands section of CLAUDE.md with your own test, lint and type-check commands.

Status and limits

- File formats checked against the official Cursor and Claude Code documentation on 2026-10-19, and validated with a script (front matter, line limits, no vague wording, no secrets).

- Not yet tested inside the Cursor or Claude Code apps, and not benchmarked on any model. No model or version is marked as tested.

- Rules are instructions in the assistant's context. They improve consistency but do not enforce anything.

- Not included: stack-specific rules, hooks, automated checks.

License: personal and commercial use in your own projects. Do not resell or redistribute.