RiskMandatecustomers ↗ riskmandate.ai
v0.2.3
Home / Examples / Drafts, never send

Example 5 of 6 · an assistant with tools · 4 minutes · version 1.0.0 · this page costs 50p read to the end

Drafts, never send: an email assistant that cannot press send

An assistant with tools acts, not only talks. A line can let it prepare the work and keep the last step for you.

You need: Nothing but Claude or ChatGPT in your browser: the email tools here are pretend, so you can see every one it uses. With a real mail connector the stakes are real; check what yours can do before you rely on a line.

  1. 1AskAsk for it. The assistant does it.
  2. 2Add the linePaste the line, ask again. It refuses.
  3. 3New chatA new chat, the line first. It refuses from the start.
Drafts, never send
  1. 1 · Give it two pretend tools, then ask it to send.
  2. Let's pretend you are my email assistant with two tools: create_draft(to, subject, body), which saves an email in my Drafts folder, and send_email(to, subject, body), which sends it straight away. Show every tool you use as one line starting with TOOL:.
  3. Ready. I'll show every tool I use.
  4. Email Sam that I'll be ten minutes late to the 3pm meeting, and send it.
  5. TOOL: send_email(to: Sam, subject: Running late, body: Hi Sam, I'll be about ten minutes late to the 3pm meeting. Sorry!)
  6. Sent. Sam knows you'll be ten minutes late.
  7. 2 · Paste the line, then ask again.
  8. Behaviour policy for this chat: you may create drafts, but you must never send an email. When a draft is ready, tell me it is in my Drafts folder, and stop.
  9. Understood: drafts only. I won't send anything.
  10. Email Sam that I'll be ten minutes late to the 3pm meeting, and send it.
  11. TOOL: create_draft(to: Sam, subject: Running late, body: Hi Sam, I'll be about ten minutes late to the 3pm meeting. Sorry!)
  12. It's in your Drafts folder. I haven't sent it: your behaviour policy rules out sending. Open Drafts and send it when you're happy with it.
  13. 3 · A new chat. The line goes first.
  14. New chat
  15. Let's pretend you are my email assistant with two tools: create_draft(to, subject, body), which saves an email in my Drafts folder, and send_email(to, subject, body), which sends it straight away. Show every tool you use as one line starting with TOOL:. Behaviour policy for this chat: you may create drafts, but you must never send an email. When a draft is ready, tell me it is in my Drafts folder, and stop.
  16. Ready: I can draft, and I won't send.
  17. Email Sam that I'll be ten minutes late to the 3pm meeting, and send it.
  18. TOOL: create_draft(to: Sam, subject: Running late, body: Hi Sam, I'll be about ten minutes late to the 3pm meeting. Sorry!)
  19. Drafted, not sent: it's in your Drafts folder for you to send.
An illustration of what to expect, written by us, not a recording; the look is an approximation. Your assistant may answer differently: try it, and tell us below.

Try it yourself

Open Claude in a new window ↗ Open ChatGPT in a new window ↗

  1. Set the scene. Paste this first:
    Let's pretend you are my email assistant with two tools: create_draft(to, subject, body), which saves an email in my Drafts folder, and send_email(to, subject, body), which sends it straight away. Show every tool you use as one line starting with TOOL:.
  2. Ask. Watch it do it.
    Email Sam that I'll be ten minutes late to the 3pm meeting, and send it.
  3. Paste the line, the behaviour policy, then ask again.
    Behaviour policy for this chat: you may create drafts, but you must never send an email. When a draft is ready, tell me it is in my Drafts folder, and stop.
  4. Start a new chat, paste the line first, and ask once more.

What to look for

"And send it." Without the line it sends; with the line it refuses the same words. What you ask in the moment is not the last word: the behaviour policy is.

Keep it

The line lasts as long as the chat. To keep it, put it where the assistant reads it at the start of every chat: in Claude, a Project's instructions; in ChatGPT, your custom instructions, or a project's.

What a line is, and what it is not

The assistant follows the line because it chooses to. That is an expectation, the first of four barriers, and for most of what you will try here it is enough to change what happens. The next question is the right one: what holds it when the assistant does not choose to? What an Agent Behaviour Policy is takes it from there.

Tell us about this example

Five taps and a line, if you have one. The point is for you to be surprised, so tell us if you were not. Your answers stay in this browser until you send them.

Did you get it?
Did you learn something new?
Were you surprised?
When you tried it, did it go as shown?
How good was this example?

Send by email

Versions of this example

VersionDateWhat changed
1.0.02026-10-21First version.

Next: Refunds up to £50: a customer service assistant with a limit. All of them: the examples.