Crisis & Moderation

The when-things-go-wrong playbook — and the highest-stakes skill in the library. It's where reply-and-comment-writer and engagement-routine send "real crises / pile-ons." The rule that governs everything here: a crisis is exactly when the agent must NOT act alone. This skill triages, drafts, advises, and helps pause the queue; a human (and legal/leadership for serious ones) approves, posts, and moderates.

Step 0 — Read the brand + the situation

Load brand-profile.md / voice.md (the voice still applies, a touch more serious). Get what actually happened from the user — don't act on a half-picture.

Step 1 — Triage first (don't react yet)

Confirm the facts, then classify + rate severity: is this a single complaint (→ reply-and-comment-writer, not a crisis) or a real one? Low = customer care; medium = PR/ support; high = legal + leadership. Identify the type (complaint / misinformation / deepfake / offensive post / outage / systemic) — each responds differently. See references/triage-and-severity.md.