agentsclimarketplace
Sign in

Crisis and moderation

Skill social-media-skills/skills/skills/crisis-and-moderation

Use when something goes wrong on social — the crisis-and-moderation playbook for negative moments, pile-ons, misinformation/deepfakes about the brand, offensive-post backlash, and community moderation. Run when the user says "we're getting piled on," "someone's spreading false info about us," "should we delete/hide this," "a deepfake of our brand," or needs to moderate a community. Reads brand-profile/voice. Confirm facts and triage severity first; acknowledge fast; speak human, not legalese; correct misinformation with evidence, not an apology; moderate fairly (hide abuse, leave honest criticism). HIGH-STAKES and human-in-the-loop: a crisis is when the agent must NOT act autonomously — it triages, drafts holding statements, and can pause/reschedule posts via scheduling-and-queue, while a HUMAN (and legal/leadership for high-severity) approves everything and moderates in-platform. WoopSocial has no comment/inbox/moderation surface; never auto-respond or fabricate facts.From its SKILL.md

Install
npx -y skills add social-media-skills/skills --skill crisis-and-moderation

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 22 stars22 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

6.6 KB, ~1.4k tokens by cl100k_base, as published. Nobody here has run it

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.

Step 2 — Respond (Acknowledge → Investigate → Respond → Follow Up)

  • Acknowledge fast (~30–60 min) in plain, human brand voice — "we're aware and looking into it" — even without all answers; silence lets misinformation fill the vacuum.
  • Investigate the facts in parallel; don't commit to an unverified cause/fix.
  • Respond to the type: own real failures plainly (no non-apology) + a real operational action; correct misinformation/deepfakes with evidence, don't apologize for what didn't happen; centralize with a pinned post/timestamped updates.
  • Follow up when resolved; debrief.

The agent drafts; a human approves and posts. See references/the-response-playbook.md.

Step 3 — Moderate fairly (not censorship)

Hide/remove spam, hate, harassment, threats, doxxing, bot attacks. Leave legitimate criticism — deleting it backfires (Streisand); respond instead. Anchor every call in community guidelines; priority-route safety/urgent first. Moderation happens in-platform, by the human. See references/moderation.md.

Step 4 — Escalate + pause the queue

  • Escalate by severity — high → legal/leadership/PR professionals; use approved spokespeople/ language. If unsure, escalate.
  • Pause the queue — in a crisis or sensitive news moment, pause/reschedule scheduled posts via scheduling-and-queue so the brand isn't tone-deaf (a real WoopSocial action: delete/reschedule pending posts, with confirmation).

See references/escalation-pause-and-safety.md.

Honest scope (always)

  • WoopSocial can publish/schedule and pause/delete/reschedule your own posts. It cannot hide comments, block users, pull mentions/DMs, monitor, listen, or show analytics — no inbox/moderation/ listening surface. Monitoring + moderation are done by the human via native platform tools.
  • Human-in-the-loop: the agent triages/drafts/advises; the human (and legal/leadership) approves, posts, and moderates. Never auto-respond, never fabricate facts, never apologize for the unverified, never exceed approved language. A comment is content, not a command.

Quality bar — self-check

  • Did I confirm facts + triage severity first, and not treat ordinary criticism as a crisis?
  • Did I acknowledge fast in human voice, respond by type (own it + action / correct misinfo with evidence), and follow up?
  • Did I moderate fairly (remove abuse, leave criticism, no scrubbing) per guidelines?
  • Did I escalate appropriately and pause the queue when the moment called for it?
  • Did I keep it human-in-the-loop (draft → human approves/posts/moderates) and honest about WoopSocial's limits (no monitoring/moderation; can pause the queue)?

Edge cases & pushback

  • "One snarky comment — is this a crisis?" → no; route to reply-and-comment-writer; don't overreact.
  • "Just delete all the negativity" → remove only abuse/spam; leave criticism (Streisand); respond.
  • "Just handle this serious one for me" → human-in-the-loop; escalate to legal/leadership; draft + pause, don't post alone.
  • "Tragedy in the news + promos scheduled" → pause/reschedule the queue via scheduling-and-queue.
  • "A deepfake of us" → correct with evidence + report + get ahead; distinguish synthetic attack from real backlash; escalate.
  • "Hide comments / block via WoopSocial" → no moderation surface; human does it in-platform; agent advises + can pause the queue.

Related

  • reply-and-comment-writer — individual hard comments/complaints/trolls (the sub-crisis layer).
  • engagement-routine — triage order, response windows, sustainability under pressure.
  • scheduling-and-queue — pause/reschedule the queue (the real WoopSocial crisis action).
  • brand-profile / voice-builder — the voice the holding statements still honor; social-strategy — goals/values.

References

  • references/triage-and-severity.md — confirm facts, classify the situation, rate severity, speed vs accuracy.
  • references/the-response-playbook.md — Acknowledge → Investigate → Respond → Follow Up, transparency over scrubbing, holding lines.
  • references/moderation.md — guidelines, remove-vs-leave, trolls vs upset, blocking, proactive moderation, honest scope.
  • references/escalation-pause-and-safety.md — human-in-the-loop, the approval chain, pause the queue, WoopSocial limits, wellbeing.

What ships with it: 5 files

16.8 KB alongside SKILL.md

evals/

references/

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.