Agent RuleZ

A deterministic, auditable, local-first AI policy engine for Claude Code.

Agent RuleZ provides a robust hook system that replaces fragile JSON configurations with readable YAML, enabling developers to enforce coding standards, inject context dynamically, and block dangerous operations with sub-10ms latency.

---

Table of Contents

• Overview
• Why Use Agent RuleZ?
• Architecture
• Installation
• Quick Start
• Configuration Guide
• The Mastering Hooks Skill
• Governance & Policy
• Development
• CLI User Guide
• Agent Skill User Guide

---

Overview

Claude Code’s native hook system relies on raw JSON configuration and shell commands, which can be opaque, hard to maintain, and lack conditional logic.

Agent RuleZ solves this by introducing a two-part system:

1. The RuleZ Binary: A high-performance Rust executable that acts as the runtime hook handler.
2. The Mastering Hooks Skill: An intelligent assistant that analyzes your project and generates configurations automatically.

Core Philosophy: LLMs do not enforce policy; they are subject to policy. Agent RuleZ is the policy authority.

---

Why Use Agent RuleZ?

---

Architecture

1. RuleZ Binary (rulez)

• Role: Deterministic runtime engine.
• Language: Rust.
• Performance: <10ms execution time (cold start <5ms).
• Function: Intercepts Claude Code events (e.g., PreToolUse, PermissionRequest), evaluates YAML rules, and returns actions (inject context, block, or run scripts).

2. Mastering Hooks Skill

• Role: Configuration assistant.
• Function: Bridges the gap between the binary and the user. It discovers project skills, parses CLAUDE.md for rules (MUST/MUST NOT), and generates the hooks.yaml file.

---

Installation

Method 1: Pre-built Binaries

Download the appropriate artifact for your platform (macOS, Linux, Windows) from the releases page.

Method 2: Cargo (Rust)

cargo install --git https://github.com/SpillwaveSolutions/agent_rulez

Method 3: Skill Assistance

If you have the Mastering Hooks skill enabled, simply ask Claude:

“Install RuleZ” The skill will detect your OS and handle the installation and verification.

---

Quick Start

1. Initialize Configuration Scaffold a new configuration with example templates:

   rulez init
   

   This creates .claude/hooks.yaml.

2. Install to Claude Code Register Agent RuleZ as the handler in Claude Code’s settings:

   rulez install --project
   

   Note: Use --user for global installation.

3. Verify Ensure your configuration is valid:

   rulez validate
   

---

Configuration Guide

Configuration is stored in .claude/hooks.yaml. It supports event-based triggers and additive matching.

Example hooks.yaml

version: "1"

# Event: Before a tool is executed
PreToolUse:
  # Rule: Enforce CDK guidelines
  - match:
      tools: ["Edit", "Write"]
      directories: ["infra/**", "cdk/**"]
    action:
      inject: "docs/cdk-guidelines.md"

  # Rule: Prevent committing "WIP" code
  - match:
      tools: ["Bash"]
      command_match: "git commit.*WIP"
    action:
      block: true
      message: "Commits with 'WIP' are forbidden on this branch."

  # Rule: Run custom Python validator
  - match:
      extensions: [".py"]
    action:
      run: ".claude/validators/security-check.py"

Supported Events

• SessionStart / SessionEnd
• PreToolUse / PostToolUse
• PermissionRequest (Force Claude to explain commands)
• UserPromptSubmit

Actions

• inject: Add Markdown context to the conversation.
• block: Stop the operation immediately.
• run: Execute a script (Python, Bun, Bash) for custom logic.

---

The Mastering Hooks Skill

The Mastering Hooks skill allows you to configure hooks using natural language.

Triggers:

• “Set up hooks for this project”
• “Enforce the ‘no console.log’ rule with a hook”
• “Why isn’t my hook blocking WIP commits?”

Capabilities:

• Project Analysis: Scans .claude/skills and CLAUDE.md to suggest relevant hooks.
• Conflict Detection: Identifies contradicting rules.
• Audit Trail: Keeps a record of installations.

---

Governance & Policy

Agent RuleZ acts as a policy engine with advanced governance features.

Policy Modes

Rules can be set to different enforcement levels:

• enforce (Default): Blocks operations if conditions are met.
• warn: Injects a warning context but allows the operation.
• audit: Logs the event without blocking or injecting (useful for silent monitoring).

Metadata & Provenance

Every rule can track its origin for auditability:

metadata:
  author: "mastering-hooks-skill"
  reason: "Enforce infrastructure coding standards"
  ticket: "PLAT-3421"
  confidence: "high"

Logging

Logs are stored in ~/.claude/logs/rulez.log in JSON Lines format, supporting “Explainable Policy” auditing.

---

Development

Prerequisites

• Rust (Cargo)
• Python 3 (for testing validators)
• Bun (for testing TypeScript validators).

Build

cargo build --release

Testing

Agent RuleZ includes unit, integration, and end-to-end tests.

cargo test

---

License

MIT License.