What it does
You paste in whatever you have: meeting notes, a plan, chat threads, a status report. It lists every risk and dependency it can find, says who owns each one according to the material, and puts the unowned ones at the top.
It’s a quick sweep, not a design review. The /risks playbook reads a full design doc against your RAID log; this prompt works on any pile of notes and answers one question: what is nobody holding?
When to use it
| Use it when | Skip it when |
|---|---|
| You’ve collected a week of notes and want to catch what fell between teams | You’re reviewing a design doc: the /risks playbook is built for that |
| You’re taking over a program and need a fast read on open exposure | You need likelihood and impact ratings; do that with the team |
| Before a risk review, to check the RAID log hasn’t missed anything | The material can’t go into your company’s approved AI tools |
The prompt
You are helping a technical program manager find risks and dependencies that nobody owns.
Use only the material I paste below. Do not invent risks the material gives no basis for, and do not invent names, teams or dates. If something you need is missing, write [MISSING: what is needed].
Find every:
- Risk: something that might go wrong and would hurt the date, cost, quality or customers.
- Dependency: something the program needs from another team, vendor, system or decision.
Include implied ones. Phrases like "waiting on", "someone needs to", "assuming", "TBD" and "we'll sort it out" usually hide one.
For each item give:
1. Item: one line, as cause and effect where you can.
2. Type: Risk or Dependency.
3. Source: where it appears, with a short quote.
4. Owner: the person or role the material names, or "None".
5. Suggested owner role: a role, not a person, and why in a few words.
Output:
- A table with those five columns, items with no owner first.
- Then "Questions for me": anything you could not classify, or where the material contradicts itself.
Do not rate likelihood or impact.
MATERIAL:
[paste]Setup notes. Paste the prompt and your material in one message. Label each source (for example “Cutover sync notes” or “Security chat”) so the Source column is useful. For a lot of material, attach the files if your assistant supports it and list their names under MATERIAL:. In a coding agent or CLI, point it at a folder of notes files.
A filled example
From the running example, Lantern: five teams moving a support assistant onto in-house inference. The TPM pasted a week of notes.
| Source | What it says |
|---|---|
| Security chat | Still waiting on a reviewer. Nobody owns scheduling it. Review gates customer traffic. |
| Data lead update | Index rebuild at about half speed. I’m trying a batch-size fix. |
| Cutover sync | Shadow run 1 done. Someone needs to check the rollback flip to the vendor API before the first ramp stage. |
| Eval thread | Assuming the golden set still matches the tickets support gets today. |
And what it gave back:
| Item | Type | Source | Owner | Suggested owner |
|---|---|---|---|---|
| Without a reviewer, the security & privacy review can’t start, and customer traffic waits on it | Dependency | Security chat: “Nobody owns scheduling it” | None | Security lead: owns review scheduling |
| If the rollback flip is untested, a bad first ramp stage can’t be undone quickly | Risk | Cutover sync: “Someone needs to check” | None | Eng lead: owns rollback |
| If the golden set is out of date, the parity eval can pass while answers get worse | Risk | Eval thread: “Assuming … still matches” | None | ML lead: owns the eval suite |
| Index rebuild too slow for the cutover window | Risk | Data lead update: “about half speed” | Data lead | Data lead |
Three of four had no owner. It also asked whether the batch-size fix had a date to report back by, since the update didn’t give one.
What to verify
| Check | Item |
|---|---|
| Every quote is in your materialSearch for each one. A quote you can’t find is an invented item. | |
| “None” really means noneSomeone may own it in a meeting the notes missed. Ask before you assign. | |
| Suggested owners are roles that existMap each to a real person, then get their yes. | |
| The risks you already know are thereIf it missed one you know about, it missed others. | |
| No [MISSING] markers leftEach one is a question for a workstream lead. |