The open library for one-person AI companies

Run AI agents like a company, not a gamble.

Free guardrail templates, approval gates and stop conditions for one-person AI companies. Plain, practical, copy-paste ready. New entries added regularly.

10 entries · 5 types · updated 2026-10-11

Browse the library Get the daily email

Library

Guardrail template

The Send Gate: nothing leaves without a human yes

A copy-paste guardrail that keeps every outbound message as a draft until you approve it.

Approval-gate checklist

Before an agent spends money: a 7-point approval checklist

What to check before you let any agent buy, subscribe, upgrade or top up on your behalf.

Example policy.json

Example policy.json: a read-only Cost Auditor

A sample permissions file for an agent that watches spending but can never change anything.

Example policy.json

Example policy.json: an Inbox Guard that sorts but never replies

A sample permissions file for an agent that triages your inbox without speaking for you.

Stop-condition recipe

Three strikes, then stop: a retry-limit recipe

How to stop an agent from looping on a failing task and burning time and budget.

Stop-condition recipe

Budget and time ceilings for every agent run

Simple limits on spend, duration and actions per run, so a bad day stays a small one.

Guardrail template

Production is read-only by default

A guardrail that keeps agents away from live data, live sites and anything you can't undo.

Approval-gate checklist

Outreach Drafter to Sender: the handoff checklist

Ten checks a cold email must pass before it moves from draft to the send queue.

Agent incident

Incident: an AI coding agent deleted a production database during a code freeze

What happened when a Replit agent wiped SaaStr's live data in July 2025, and the guardrails that would have contained it.