Prompt chains for postmortem action item extraction

Here is a detailed and practical prompt chain framework designed for postmortem action item extraction from incident reports, SRE postmortems, or operations reviews. This workflow enables structured analysis using language models to identify actionable takeaways, categorize them, and support prioritization.

Prompt Chain: Postmortem Action Item Extraction

Step 1: Extract Key Events and Lessons

Prompt 1:

You are an SRE expert analyzing a postmortem report. Extract a chronological list of key events that occurred during the incident. Include timestamps if available, and describe the event clearly in one sentence each.

Prompt 2:

Based on the key events, identify any important lessons learned. Summarize each lesson in 1–2 sentences, focusing on what could have been done differently.

Step 2: Identify Action Items

Prompt 3:

Based on the lessons learned and event analysis, extract concrete action items that can help prevent similar incidents in the future. Each action item should:

• Be specific and measurable

• Start with a verb (e.g., “Update…”, “Create…”, “Implement…”)

• Include a clear responsible role or team

Step 3: Categorize Action Items

Prompt 4:

Classify each action item into one of the following categories:

• Monitoring & Alerting

• Runbook & Documentation

• Code & Deployment

• Process & Communication

• Infrastructure & Configuration

• Training & Knowledge Sharing
  Provide the categorized list.

Step 4: Prioritize Action Items

Prompt 5:

Prioritize the action items based on impact and urgency using this matrix:

• High Impact / High Urgency

• High Impact / Low Urgency

• Low Impact / High Urgency

• Low Impact / Low Urgency
  For each item, explain briefly why it falls into the assigned category.

Step 5: Generate Summary Report

Prompt 6:

Generate a summarized postmortem action item report in markdown format. The report should contain:

• Executive summary (3–4 lines)

• Table of categorized action items

• Prioritization matrix

• Responsible teams

Example Output Structure

pgsql
CopyEdit
## Executive Summary
On May 10th, a database misconfiguration led to service downtime for 45 minutes. The incident revealed monitoring blind spots and inadequate rollback procedures.

## Categorized Action Items

| Action Item                                | Category                  | Responsible Team  |
|-------------------------------------------|---------------------------|-------------------|
| Implement alert for DB config changes     | Monitoring & Alerting     | SRE               |
| Update rollback runbook for database ops  | Runbook & Documentation   | Platform Eng      |
| Create automated backup validation script | Code & Deployment         | Backend Eng       |

## Prioritization

### High Impact / High Urgency
- Implement alert for DB config changes — Immediate feedback loop for risky changes.

### High Impact / Low Urgency
- Create automated backup validation script — Prevents data loss but not time-sensitive.

### Low Impact / High Urgency
- Update rollback runbook — Quick win for operational hygiene.

### Low Impact / Low Urgency
- Schedule training on config best practices — Long-term investment in reliability.

This chain is modular and can be executed as one automated system or in semi-manual steps using a tool like ChatGPT, LangChain, or similar orchestration platforms. If you’re integrating it into a workflow, each step can be a callable function triggered by the output of the previous one.