# Incident Response

> Stabilize first. Then find the blast radius, the trigger, and the fix that will not make it worse. Use when the user asks to incident.

- Skill: `delzarsolutionsllc/incident-response` (Agent Skill)
- Install (CLI): `npx skillmds@latest add delzarsolutionsllc/incident-response`
- Raw SKILL.md: https://api.skillmd.com/api/skills/delzarsolutionsllc/incident-response/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: delzarsolutionsllc (https://skillmd.com/u/delzarsolutionsllc)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/delzarsolutionsllc/incident-response

---


# Incident Response

Priority is user impact, then stop the bleeding, then understand, then fix, then write it down.

## Stabilize
1. Is it still happening? Who is affected? Since when?
2. Roll back or disable the last change if that is the faster restore. Fix-forward only when rollback is impossible or worse.
3. Do not deploy unrelated changes during the incident.

## Diagnose
- Symptom, first bad version, first bad config, first bad query.
- Blast radius: which tenants, regions, tables.
- One timeline with timestamps. Guessing is labeled as a guess.

## Close
- User-facing status if that is your job.
- Fix + regression test or guard.
- Follow-ups with owners. "We should monitor this" is not a follow-up.

Do not write a witty postmortem. Write a dull, true one.