Steal My Prompt Vol. 48: The Incident Escalation Editor

Something broke, and the write-up is yours. It travels to people who were not in the room and will read it once, and how it reads decides what happens next. Most first drafts swing between covering and confessing, and neither earns help.

Entry five in the Editor series runs on the same mechanic that carried the Memo Editor: the model reads your draft as skeptically as the leader receiving it, refuses the weak patterns by name, and demands what is missing. The weak patterns in an escalation are familiar ones. Blame dressed as analysis. A story with no timeline. A root cause that stops at a person. The fix underneath all three is what I call the Symptom-to-System pattern, and once it clicks, you will point it at post-mortems, audit responses, and every review that has to explain a failure without staging a trial.

I built this one because the write-up after a failure is the most consequential document nobody teaches you to write, and it is a learnable one. Vol 40 gave the memo an adversarial editor. This entry points the same discipline at the moment that needs it most.

What You Can Use This For

• The process failure your team owns and leadership needs to hear about from you first
• A system outage or vendor miss you have to summarize two levels up
• The customer-impacting mistake where your account of it shapes the response
• A clinical or safety event communication to leadership, after the official report is filed
• A compliance or audit finding you are self-reporting
• The project post-mortem that has to name causes without naming culprits

How to Use It

1. Open Claude. The prompt works the same in ChatGPT, Microsoft Copilot, and Gemini, all on free tier.
2. Turn on the reasoning option for this one. That is "Extended thinking" in Claude, the reasoning model in ChatGPT, "Think Deeper" in Copilot, or "Deep Think" in Gemini. It catches more of the weak patterns on the first read.
3. Paste the prompt, then paste your draft write-up where marked. Generalize anything sensitive first, so names become roles and systems become plain descriptions.
4. Answer the interview one question at a time. The questions are the point, because they surface what your draft assumed the reader already knew.
5. Read the Pass 1 table before you accept the Pass 2 rebuild. The table is where the lesson lives.

Pro tip: Run your draft through twice, once answering the interview and once typing "skip." Comparing the two outputs shows you exactly what the interview extracted, which teaches you what to put in your first draft next time.

The Prompt

Edit my incident write-up for the upward read. The problem: something went wrong on my watch, and this document travels to people with no context who will read it once and decide whether I get help or blame. Written defensively it reads as cover. Written raw it reads as panic.

What a strong write-up does: states the facts on a timeline, names the conditions that made the failure possible instead of the people nearest to it, says plainly whether the problem is contained, and makes one specific ask. What a weak one does: blame dressed as analysis, a story with no timeline, "human error" offered as a root cause, and corrective actions that amount to a stern reminder. Edit my draft against that standard, and read it as skeptically as the leader who will receive it.

Here is my draft: [PASTE YOUR DRAFT WRITE-UP]

Before you edit anything, interview me. Ask one question at a time, wait for my answer, and never stack two questions in one message. Ask at most four, and stop early once you have what you need:

1. Who receives this, and what do they already know about the incident? 2. What is the current state: fully contained, still active, or at risk of recurring? 3. What do I need from leadership: awareness only, a decision, resources, or air cover? 4. Only if my draft leaves it unclear: what actually happened, in one sentence?

If I type "skip," proceed using clearly stated assumptions and label each one. Do not produce the edited write-up until the interview is complete or I skip.

Then do two passes.

Pass 1, interrogate the draft. Flag every line that fits any of these patterns: (1) it names a person where it should name a condition, (2) it offers a cause that stops at "someone made a mistake" without the condition that made the mistake possible, (3) it describes an event with no time attached, (4) it proposes a corrective action that cannot be verified as done. For each flagged line, output a three-column table: THE LINE | THE PROBLEM | THE VERSION THAT SURVIVES THE UPWARD READ. If a line is already strong, leave it alone. A short table is a good sign, not a failure.

Pass 2, rebuild it for a reader with two minutes. Produce the write-up in exactly four labeled sections: WHAT HAPPENED (timeline, facts only), WHY IT HAPPENED (contributing conditions, not culprits), WHERE IT STANDS (contained, active, or recurring risk), WHAT I NEED (the specific ask, from question 3 or your labeled assumption). Keep it under 300 words.

Close with one line naming the single sentence in my original draft most likely to trigger blame instead of help. If nothing qualifies, say so.

One boundary: if this incident involves harm to a person or a regulatory reportable event, this prompt edits the leadership communication only. File the official report through your organization's required process first.

Transparency and Notes

• Built and tested in Claude. Works in ChatGPT, Copilot, and Gemini on free tier, no special features required.
• Strip names, patient details, and confidential specifics before pasting. Use roles instead of names and plain descriptions instead of system or vendor names.
• If the incident involves harm to a person or a regulatory reportable event, file through your organization's official process first. This prompt edits the leadership communication, not the official record. Educational content, not legal or compliance advice.
• Entry #5 in the Editor series. The prior entry is The Memo Editor.