rpa-gen-rules: project rules for agents

Install

/plugin marketplace add EvilFreelancer/rpa-skills
/plugin install rpa-gen-rules@rpa-skills

This is third-party code. Review the repository files before installing.

What it does

Using the repository's code, specs and docs, the skill creates or refreshes agent rules: .cursor/rules/*.mdc, CLAUDE.md with .claude/rules and a root AGENTS.md. The rules require building the project in layers, from modules with no internal dependencies upward, and working test-first. Every rule gets a mandatory sync step so a change for one agent lands in all trees in the same commit. For Codex it installs a hook bridge that attaches Cursor rules by glob.

Who it is for. Teams and developers who use several agents in one repository and want the same rules for all of them.

Example request

/rpa-gen-rules generate rules for this repository

Limitations

Generated rules are written in English by default. Testing templates are pytest-oriented. After the Codex bridge is installed you must run /hooks once to trust the hook.

How to disable. Remove the plugin via /plugin in Claude Code or delete the rpa-gen-rules folder from the agent's skills directory.

Security check

• Writes a Python hook for Codex into the project that runs during sessions

README in short

The skill creates agent rules for Cursor, Claude Code and Codex following a layered-cake model and BDD style. No separate user brief is needed; the agent infers rules from the repository. Codex gets a hook bridge that reads .cursor/rules directly. The skill is part of Pavel Rykov's rpa-skills collection and is packaged from the cursor-vibe-prompts prompt set. MIT license.

SKILL.md

---
name: rpa-gen-rules
version: 1.3.0
description: >
  Run when the user invokes /rpa-gen-rules or asks to create or refresh agent project rules (Cursor .mdc,
  Claude Code CLAUDE.md and .claude/rules, Codex .codex hook bridge). Infers from specs, docs, and code.
  Rules use a layered-cake model (implement inner layers with no dependencies first, then the next layer)
  and BDD-style delivery. Generated rules and AGENTS.md are written in English unless the user asks for
  another language. Generated rules include a mandatory Rules Sync step so changes to one agent's tree are
  mirrored to every other agent's tree (Cursor <-> Claude <-> Codex <-> AGENTS.md) in the same commit.
  For Codex the skill installs a hook bridge that auto-attaches Cursor rules by glob, like Cursor and
  Claude Code do natively. No separate user brief required.
---

# RPA generate project rules (layered + BDD, multi-agent)

## Goal

Produce or update **versioned** instructions so the agent stays inside architecture, style, and **BDD-style** discipline (behavior specified in tests, tests before new behavior, full suite before calling work done).

- **Layered cake** - rules must tell the agent to build **by layers**: first units that depend on nothing inside the project, then layers that depend only on lower layers, and so on. Mirror this in `architecture` and `implementation-order` style files.
- **Cursor** - `.cursor/rules/*.mdc` ([MDC](https://github.com/nuxt-content/mdc), `globs`, `alwaysApply`, `@` links).
- **Claude Code** - `CLAUDE.md` and `.claude/rules/*.md` with optional YAML `paths:` ([memory](https://code.claude.com/docs/en/memory#organize-rules-with-claude/rules/)).
- **Codex** - root `AGENTS.md` plus a `.codex/` hook bridge (`hooks.json`, `hooks/attach_rules.py`, `rules.md` index) that injects `.cursor/rules/*.mdc` into Codex sessions: `alwaysApply` rules at `SessionStart`, glob-matched rules on `PreToolUse` - the same behavior Cursor and Claude Code have natively.

## Language of generated files: English by default

Everything this skill writes into the target project is **written in English** unless the user explicitly asks for another language:

- top-level briefs - `AGENTS.md`, root `CLAUDE.md`, `.claude/CLAUDE.md`, `.github/copilot-instructions.md`;
- rule files and their frontmatter - `.cursor/rules/*.mdc`, `.claude/rules/*.md`, and any other agent tree;
- headings, bullets, examples, and comments inside generated code snippets.

Related