Quick guardrails checklist

Toolkit 🧰

Type: Checklist ✅ · Theme: Rails 🛤️

Purpose

A five-question pre-flight check before you hand something important to an agent. Designed to be fast enough that you’ll actually use it.

How to use

Before you paste in data or ask for advice on a higher-stakes topic, scan down this list.

If you hit “Yes” on any question, slow down, adjust your prompt, or choose a different workflow.

The checklist

  • 1. People & harm

    Could someone be materially harmed if this goes wrong?

    (e.g. health, safety, finances, mental health, legal status, job security)

    If Yes: This should not be decided by an agent alone. Reframe as “help me think” not “tell me what to do”, and make sure a qualified human is responsible for the final decision.

  • 2. Sensitive data

    Am I about to paste in information that is private, confidential, or identifiable?

    (e.g. full names + contact details, internal financials, HR data, medical details)

    If Yes: Remove or anonymise details first. Use placeholders (“Person A”, “Company X”) and avoid uploading anything you wouldn’t email to a stranger.

  • 3. Legal, compliance, or HR

    Does this touch on law, contracts, HR decisions, or formal policy?

    If Yes: Treat the agent as a brainstorming / drafting partner only.

    Add a line like: “You are not a lawyer. I will have a qualified professional review this.”

    Never skip human review.

  • 4. Reputation & voice

    Is this something that represents you in public – like a client proposal, strategy doc, press post, or homepage copy – as opposed to a low-stakes rough note or quick internal email?

    If Yes: Let the agent draft and restructure, but you decide what actually ships.

    Always review tone, facts, and implications.

    Watch for over-confident claims, made-up statistics, or flattery that doesn’t feel like you.

  • 5. Over-confidence & Panic Mode

    Am I asking the agent to fix something I don’t fully understand myself – and hoping it just sorts it out?

    🚩 If Yes: That’s a red flag for Panic Mode.

    Pause.

    Ask the agent to explain the situation back to you first, then work together on options.

    If you still feel out of your depth, involve a human expert.

  • If you have multiple "Yes" answers

    – Slow down

    – Narrow the task

    – Switch to “help me think”

    – Make sure a qualified human owns the final decision