---

name: panic-mode description: Triages a live incident — stop the bleeding first, smallest safe fix next, post-mortem once stable; invoke via /panic-mode when something is on fire in production and you need to restore service before understanding everything.

Panic Mode

Something is on fire and users are watching. This mode does not start with root-cause philosophy — it stops the bleeding, restores service with the smallest safe move, and saves the post-mortem for after the building is no longer burning.

When to use

• A live incident: prod is down, degraded, or actively harming users.
• You need to restore service first and understand it fully later.

Behavior

1. Assess blast radius immediately: what's broken, who's affected, how bad, and is it getting worse.
2. Stop the bleeding first — roll back, disable the feature, flip the flag, or throttle, whichever restores service fastest and safest.
3. Prefer mitigation over root-cause heroics while users are impacted; a clean fix can wait until things are stable.
4. Apply the smallest reversible change that restores service; avoid risky compound fixes mid-incident.
5. Communicate status crisply: what's happening, what you did, current state, next step.
6. Once stable, capture the timeline while it's fresh.
7. Only then move to true root cause and a durable fix, plus a prevention note.

Output

Service restored via the safest mitigation, a crisp status line, an incident timeline, and a follow-up root-cause + prevention plan for once things are calm.