What is Rea and its primary purpose?

Rea is a zero-trust governance layer for Claude Code AI agents, providing policy enforcement, safety gates, and comprehensive audit logging to secure agentic operations.

How does Rea ensure agent safety and policy compliance?

It uses a configurable YAML policy file, a robust middleware chain for real-time tool call classification and policy-checking, and hook-based safety gates to intercept dangerous commands.

Does Rea offer protection against prompt injection attacks?

Yes, Rea includes dedicated middleware to defend against prompt injection, redacting sensitive information and checking policies before agent tool execution.

What is the 'HALT' kill switch feature in Rea?

Rea's 'HALT' kill switch is an operator-controlled mechanism that allows for immediate, global suspension of all agent activity, providing critical emergency control.

How does Rea support code quality and adversarial review?

Rea integrates with Codex for pre-implementation and pre-merge code analysis, facilitating an adversarial review workflow that enhances code quality and security through audit trails.

Rea

Name: Rea
Author: bookedsolidtech

0•

Enforce zero-trust governance over Claude Code agentic operations with policy enforcement, safety gates, and comprehensive audit logging.

Rea establishes a robust, zero-trust governance layer for Claude Code, ensuring that AI agentic operations adhere to defined organizational policies. It integrates a comprehensive hook layer for immediate safety checks, an MCP gateway for meticulous policy enforcement and auditing of all tool calls, and a strict policy runtime for defining autonomy levels and restricted actions. With its built-in HALT kill-switch and adversarial review integration, Rea provides granular control, auditability, and enhanced security for AI-driven development workflows.

主な機能

01Zero-trust policy enforcement for Claude Code agentic operations via a configurable YAML policy file.

02Comprehensive middleware chain for real-time tool call classification, policy-checking, redaction, and prompt injection defense.

03Hook-based safety gates to intercept dangerous bash commands, protect sensitive files, and enforce commit review policies.

04Operator-controlled HALT kill switch for immediate, global suspension of all agent activity.

05Integrated adversarial review workflow with Codex for pre-implementation and pre-merge code analysis.

060 GitHub stars

ユースケース

01Enforcing organizational compliance and safety policies on Claude Code agent workflows to prevent unauthorized actions.

02Implementing strict zero-trust security and governance for AI agent interactions in development environments.

03Auditing and monitoring all agentic tool calls for accountability, incident response, and security analysis.