Cover of "AI Agent Guardrails" by Charles Chong: a seven-stage tool-call guardrail pipeline — prompt policy, tool-call gate, input validation, execution sandbox, output filter, confirmation check, and audit & logging — between an AI agent and an approved action. Published by SigmaShake.

Book · $5, one-time

AI Agent Guardrails

Containing Autonomous Coding Agents on Windows, macOS, and Linux

A practitioner’s handbook on containing autonomous coding agents at the tool-call, process, kernel, and network boundary, across Windows, macOS, and Linux. 24 chapters, 136,146 words. It does not promise a safe agent — guardrails reduce blast radius and buy detection time when one goes wrong, not more than that — and the book is written to match that honestly, with runnable labs and real rule files throughout.

$5 once. Not a subscription, and not part of the $5/month SSG plan — the book is billed and delivered separately by book.sigmashake.com.

24
Chapters
136,146
Words
93
Code listings
24
Hands-on labs
254
Citations verified
10
Diagrams

What’s inside

Part I
The Problem

Why autonomous coding agents need boundaries, and what happens without them.

Part II
Interception

Gating tool calls before they execute, at the hook and MCP boundary.

Part III
Containment

Process, filesystem, and kernel-level blast-radius limits once a call is allowed.

Part IV
Perimeter

Network egress and the boundary an agent can actually reach.

Part V
Measurement and Response

Benchmarking containment honestly, and reacting when it fails.

Part VI
Practice

Running interception, containment, and perimeter controls together on a real machine.

Free sample: preface + Chapter 1

Read the preface and the first chapter before you buy, at book.sigmashake.com.

Read a free sample

Purchase and download are handled by book.sigmashake.com. Published by SigmaShake, a sole-operator individual entity, USA.