Sentinel Guide

Core Concepts

Guardrails

Guardrails are permanent rules that control your bot's behaviour. They live in a file called AGENTS.md and are enforced every single time, across every conversation.

How Guardrails Work

Every bot has a configuration file called AGENTS.md. This file contains the rules and boundaries your bot follows. Unlike a verbal instruction in chat that might be forgotten over time, rules in AGENTS.md are permanent — they are checked and enforced on every single interaction.

Pre-configured defaults

Your bot comes with sensible default rules already in place — industry-appropriate behaviour, professional tone, and standard safety checks. These work well out of the box.

Your risk tolerance

Every user has a different comfort level. Some want their bot to act autonomously, others want approval at every step. Guardrails let you set the boundaries that are right for you.

Always enforced

Rules in AGENTS.md persist forever. They are not suggestions — they are hard rules your bot must follow in every conversation, without exception.

Want a new rule? Just tell your bot.

Any time your bot does something you'd rather it checked with you first, tell it the rule and ask it to add it to AGENTS.md. That rule will persist forever — across every future conversation.

YouAdd a rule to your AGENTS.md: always ask me for approval before sending any email. Never send without my explicit confirmation.
BotDone. I've added this to my AGENTS.md. From now on, I'll always draft emails for your review before sending.

Example Guardrails

Here are some common rules users add to their bots. You can add any rule — these are just ideas to get you started.

Email approval

“Always show me a draft and get my approval before sending any email.”

Client confidentiality

“Never include client names or personal details in email subject lines.”

Cron job safety

“Always confirm with me before creating, modifying, or deleting any cron job.”

Communication style

“When drafting client emails, always use a formal and professional tone.”

Web browsing control

“Do not browse the web unless I specifically ask you to.”

Managing Your Guardrails

Add a rule

Tell your bot the rule and ask it to save it to AGENTS.md. It takes effect immediately.

View your rules

Ask your bot: “What rules do you have in your AGENTS.md?” It will list all current guardrails.

Remove a rule

Tell your bot to remove a specific rule from AGENTS.md. It will stop enforcing it immediately.

Modify a rule

Ask your bot to update an existing rule. For example: “Change my email rule to only require approval for external emails.”

Start simple, adjust as you go

Your bot's default rules are a great starting point. As you use your bot, you'll naturally discover situations where you want more or less control. Add rules when you need them — over time, your bot's behaviour will be perfectly tailored to how you work.