AI Agent|
Research

Agent Goph

Autonomous infrastructure auditing.

ReAct

Pattern

Claude

LLM

Read-only

Mode

Overview

Agent Goph uses a ReAct (Reason + Act) pattern powered by Claude to autonomously inspect cloud infrastructure. It observes state, reasons about findings, selects appropriate audit tools, and produces structured reports — all with human-in-the-loop approval for any actions. Safe by default: all operations are read-only unless explicitly run in dry-run mode.

Stack

Python 3.11+Claude APIReActMCPasynciopytest

Features

What it does.

Audit Capabilities

  • Security auditing: misconfigurations, weak policies, exposed resources
  • Cost optimization: unused resources, rightsizing, savings recommendations
  • Compliance checks: configurable benchmarks and policy enforcement

Agent Architecture

  • Claude-powered reasoning loop (ReAct pattern)
  • MCP-based tool integrations (Cloudflare, extensible to AWS/GCP)
  • Modular audit engines: security, cost, compliance
  • Structured JSON and markdown reporting

Safety

  • Read-only operations by default
  • Explicit dry-run mode for any mutations
  • Human-in-the-loop approval for actions
  • Full audit trail of agent decisions

Architecture

How it's built.

1

Agent Core

Claude-powered orchestration with ReAct reasoning loop

2

Tool System

MCP-based integrations for cloud provider APIs

3

Audit Engine

Modular auditors for security, cost, and compliance

4

Reporting

Structured JSON and markdown output with severity levels

Interested in this project?

Get in touch →
← Back to all projects