Work · Community report
Turn a to-do list into a prompt for another coding agent
@alliekmiller reports spending an evening talking through a 19-item to-do list with Instinct, then asking it to write one large prompt for Codex and email that prompt back. This page turns that to-do list handoff into a repeatable request, and says plainly what the post leaves out.
What you can get
- Prepare
The to-do list, or a conversation where you have talked it through
- Ask and review
Get the list out of your head first, in whatever form you like. Note anything that needs a decision from you before an agent starts.
- Check the result
One self-contained prompt that covers every item on a to-do list, plus a short second prompt that points the other agent at it, with nothing sent before you read both.
The source case
@alliekmiller describes talking nonstop to Instinct about everything on a list, covering drafting, prioritizing and strategizing. Before bed, by their account, they had Instinct create a massive prompt for Codex to execute the entire 19-item list and email that prompt to them, then had it write the prompt they gave Codex to search their email for the note. The post is the first of three tweets in a thread; the other two are not in the evidence. It does not show the prompt, the list items, what Codex did with it, or whether any item was finished.
A public report by someone outside this library. Not independently reproduced here.
From the source
Source excerpt
Before I went to bed, I had Instinct create a massive prompt for Codex to execute the entire 19-item list and then email that prompt to me.
Quoted from the original posts. Not written by this guide, and not a result produced here.
Suggested prompt
Bring The to-do list, or a conversation where you have talked it through, An email address the prompt should go to, A second agent you can paste the prompt into, A list of actions that agent must never take without asking. Source community report, checked 2026-10-09.
Written by this guide as a starting point, not quoted from the source. Your edits stay in this browser.
What you need
You supply
- The to-do list, or a conversation where you have talked it through
- An email address the prompt should go to
- A second agent you can paste the prompt into
- A list of actions that agent must never take without asking
How to try it
Editorial, not from the source
- Get the list out of your head first, in whatever form you like. Note anything that needs a decision from you before an agent starts.
- Ask for one prompt per run, not a running conversation: each item with its context, a definition of done and a boundary on what the agent may touch.
- Read the generated prompt before it leaves. Cut or rewrite any item that would send, buy, delete or publish something.
- Have the prompt emailed to yourself, as the author describes, and check that the email arrived with the full text intact.
- Paste the second, short prompt into the other agent only when you are ready for it to start, and watch the first item or two.
- Afterwards, compare what came back against the list and mark each item done, partial or not started.
What the author describes
@alliekmiller calls it "Perhaps my most unhinged workflow". The day's work was a conversation with Instinct about a list: drafting, prioritizing and strategizing. At the end of it they asked for a single large prompt that Codex could run against all 19 items, delivered by email so it would be waiting.
The second step is the unusual one. Rather than copy the prompt across by hand, they had Instinct write the short instruction they gave Codex, which told it to search their email and find the note Instinct had left. The post stops there, at the first of three tweets.
Making a list safe to hand off
A list that made sense in conversation often does not stand alone. An agent reading it cold has no idea which items you meant literally, which were reminders to yourself, or which depend on a choice you hadn't made. Ask the first assistant to flag those before they reach the second.
Nineteen tasks in one run is a lot to review after the fact. Splitting the list into a first batch of low-risk items and a second batch that touches accounts or sends messages keeps a bad guess from spreading across everything at once.
Open Instinct to try it. That is the product's own site; this guide is not part of it.
Limits and confirmation points
- The 19 items, the prompt and the email are not shown, so there is no way to judge how well the list was converted.
- The post gives no result from Codex; whether any item was completed is unreported.
- The thread's later tweets are not in the evidence, so the story may continue past what this page covers.
- How the email was found or opened by the second agent is not described beyond the instruction to search it.
- This is an attributed community report, not reproduced by this guide. The suggested workflow and checks are editorial guidance, not a recovered conversation or guaranteed product capability.
Editorial, not from the source
- The second agent cannot find the email
- Check the subject line and sender, and that the message is not in spam. Paste the prompt directly if the search fails.
- The generated prompt drops items or merges them
- Number the items in your request and ask for the prompt to keep the same numbers, then check each one against your list.
Common questions
Can Instinct write a prompt for Codex?
@alliekmiller reports having it create a large prompt for Codex covering 19 items and email the prompt to them. The post does not show the prompt or Codex's result, and this guide has not reproduced it.
Why email the prompt instead of pasting it?
The post does not say. In practice an email gives you a saved copy you can read and edit before another agent sees it.
Is it safe to let a second agent run a whole list unattended?
The post does not show a run. Treat it as a risk: limit what the agent can touch, and review the first results before letting it continue.
Next
Related to this task
More in Work
Get recurring task recommendations — ask your own agent to check for relevant updates.