agents

Shared rules, commands, and skills for AI coding assistants. Uses rulesync to generate configuration for multiple targets from a single source.

Supported Targets

• Cursor
• Kilocode
• Claude Code
• Codex CLI

Structure

.rulesync/                    # Source of truth - edit files here
  rules/overview.md           # Project rules and guidelines
  commands/prime.md           # Prime command
  commands/review.md          # Code review command
  subagents/planner.md        # Planner subagent
  skills/code-review/SKILL.md # Code review skill
  .aiignore                   # Files for AI to ignore

.claude/settings.json         # Claude Code settings (not managed by rulesync)
rulesync.jsonc                # Configuration
sync-rules.js                 # Generate script (cross-platform)

Prerequisites

Node.js is required to run the sync script.

macOS:

brew install node

Windows:

scoop install nodejs

Usage

Edit files in .rulesync/, then regenerate:

node sync-rules.js

New Project Setup

To initialize rulesync in a new project:

npx rulesync init

Then copy the .rulesync/ contents from this repo as a starting point.

Generated Output

Target	Output
AGENTS.md	AGENTS.md
Cursor	AGENTS.md, .cursor/commands/, .cursor/skills/, .cursorignore
Kilocode	.kilocode/rules/, .kilocode/workflows/, .kilocode/skills/, .kilocodeignore
Claude Code	.claude/CLAUDE.md, .claude/commands/, .claude/agents/, .claude/skills/, .claudeignore
Codex CLI	AGENTS.md, .codex/skills/ (via sync script)

Customization

The script runs npx rulesync generate and performs customization steps.

• AGENTS.md
  • I prefer AGENTS.md over tool-specific rules subdirectories (e.g., .cursor/rules/) since it's a universal format supported by multiple tools.
  • The sync script removes .cursor/rules/ after generation because Cursor reads AGENTS.md as a fallback.
• Codex skills
  • rulesync doesn't currently support codex project skills, so the script copies the skills to .codex/skills/.

Adding Content

Rules

Edit .rulesync/rules/overview.md to change project guidelines.

Commands

Add .md files to .rulesync/commands/ with frontmatter:

---
description: 'Command description'
targets: ["*"]
---

Skills

Create a directory in .rulesync/skills/<name>/ with a SKILL.md file:

---
name: skill-name
description: What the skill does
targets: ["*"]
claudecode:
  allowed-tools: ["Read", "Grep", "Glob", "Bash"]
---

Subagents

Add .md files to .rulesync/subagents/ with frontmatter:

---
name: subagent-name
description: What the subagent does
targets: ["*"]
claudecode:
  model: inherit
---

Claude Code Settings

.claude/settings.json contains Claude Code-specific configuration (not managed by rulesync):

• Permissions: deny list for dangerous commands, allow list for common dev tools
• Environment: extended timeouts for long-running commands
• Sandbox: enabled with Docker socket access
• Status line: token usage via ccusage