---
name: community_management
description: Run an online community (Discord server, forum, social group) end to end: design welcome/onboarding flows, write enforceable community guidelines, run a moderation ladder with an audit trail, build engagement loops and rituals that drive retention, and de-escalate conflict with BIFF/EAR techniques. Triggers when someone asks to start, grow, moderate, or fix an online community.
metadata: { "includeInPrompt": true }
---

# Community Management

## Purpose
Turn a group of strangers into a self-sustaining community. This skill gives an agent the full operating playbook: onboarding that converts lurkers into contributors, guidelines short enough to be read, a graduated moderation ladder (educate → moderate → remove) with a written audit trail, engagement loops and rituals that create retention, and de-escalation responses for heated moments. Informed by Higher Logic's moderation best practices, the oss-wishlist moderation strategy rubric, and Bill Eddy's BIFF/EAR conflict framework.

## Workflow

### 1. Set the foundation (before launch)
- Write the one-page guidelines first. Cap at ~500 words: expected behaviors, who the community is for, banned content, banned behaviors (harassment, bullying, defamation, intolerance), usage rules (no ads, no thread hijacking, no spam, stay on-topic, post in the right place), and how members should handle complaints. Put legalese below, guidelines above. Force acceptance at signup; link it from the homepage/footer. See `references/guidelines-template.md`.
- Design structure that scales: a channel/category taxonomy that works at 50 members must not break at 5,000. Start with FEW channels — add more only when a topic repeatedly floods an existing one.
- Assign moderation roles before you need them: who triages what, a rotation so nobody burns out, and one escalation contact for serious issues.

### 2. Run the welcome flow (member journey: visit → lurk → contribute → champion → lead)
1. Auto-welcome within minutes of joining: greeting + one link to guidelines + ONE simple first action ("introduce yourself in #intros" or "pick your role in #start-here"). One ask, not five — choice paralysis kills signups. See `references/welcome-flows.md` for templates.
2. Check back at 24–48h: if the newcomer hasn't posted, a human-feeling nudge ("saw you joined — anything you're hoping to find here?") often turns a quiet joiner into a first-time poster. No guilt, no pressure.
3. At 7 days with zero activity: final light-touch check-in, then let them be. Lurkers are fine members; most communities run on a small active core.
4. Escalate involvement deliberately: first reply → contributor recognition → champion (regular) → leadership (helper/mod). This is the engagement loop: **prompt participation → respond meaningfully → recognize contributions → escalate involvement**. Members who lead rarely churn.

### 3. Build engagement loops and rituals (retention beats acquisition)
- **Fixed schedule = automatic attendance.** Set rituals on a calendar and never skip: Weekly Wins, Monday Goals, Friday Highlights, Spotlight Sunday. Consistency matters more than complexity — a simple weekly check-in maintained for a year beats an elaborate event series that dies in month two.
- **Recognition is retention.** Celebrate milestones publicly: 10th event, 6-month anniversary, member wins in newsletters. Dedicate a segment of every weekly check-in to shoutouts. Specific format matters less than consistency.
- **Close every feedback loop visibly:** collect input → act on it → announce what changed. Collecting without acting reads as performative. Run structured feedback cycles quarterly; quick pulse checks after events.
- **Seed micro-communities:** niche groups by interest, region, role, or seniority. Smaller groups = stronger bonds. Hand mature members real ownership (ambassador programs, peer mentorship, member-run rituals) — member-driven rituals signal a healthy, decentralized community.
- **Each event is a stepping stone:** at the close, promote the next one, open early registration, encourage post-event discussion, and highlight attendee-generated content. See `references/engagement-loops.md`.

### 4. Moderate (the ladder)
Use the graduated ladder below for every violation. Log every action in `assets/strike_log.csv` — who, what, when, which strike, evidence link. The log is the audit trail; no log, no enforcement.
1. **Strike 1 — Educate.** Written notice explaining what happened, why it violated the guideline, and a positive redirect. Assume genuine ignorance on the first offense (most first-timers truly didn't know). Example: someone posts business promo → DM explains the rule, links a relevant thread where they can share expertise without selling.
2. **Strike 2 — Moderate.** Second written notice + offer to talk it through on a call. Place on full moderation (posts held for review) until trust is rebuilt. Still hear them out — two moderators should sanity-check a strike 2 before it lands.
3. **Strike 3 — Remove.** Suspend access + final notice. Three strikes, out.
- **Never delete negative feedback unless it breaks the rules** — address criticism respectfully, in the community's voice. Deleting criticism reads as censorship; steering conversation is good, silencing users is bad.
- **Severity shortcut:** threats, doxxing, hate speech, or illegal content skip the ladder — remove immediately, log it, escalate to the community owner and, if needed, the platform.
- Keep a shared internal FAQ / moderation cheat sheet so two moderators handle the same comment the same way. Response chaos (one ignores, one over-apologizes, nobody logs) fractures community trust. See `references/moderation-playbook.md`.

### 5. De-escalate (BIFF + EAR)
When a member is heated but not yet violating the ladder:
- **BIFF written response** (Bill Eddy, High Conflict Institute): **B**rief — short, no essay. **I**nformative — facts only. **F**riendly — civil tone, no warmth required. **F**irm — clear boundary that closes the loop; no questions that reopen debate. Before sending, strip it of advice, admonishments, and unnecessary apologies; ask yourself how they'll react to it.
- **EAR for live conflict** (Eddy's in-person de-escalation): **E**mpathy — "I can hear you're really frustrated" (acknowledges feeling, concedes nothing). **A**ttention — show you're present, not defensive. **R**espect — treat them as worthy of dignity even when their behavior isn't. Empathy ≠ agreement.
- **Limits, not ultimatums.** "If X continues, you'll be muted for 24h" is a limit. "Do this or else" is a threat that invites a power struggle you can't win. See `references/de-escalation.md` for templates and worked examples.

### 6. Watch community health
Track weekly: daily/weekly active members, post frequency, first-response time, sentiment trend, churn rate. Watch for dropoff points (where newcomers stall) and engagement peaks (what to double down on). Moderators' pattern-spotting feeds the feedback loop — they see problems before the metrics do.

## Output Contract
When this skill runs, it produces one or more of:
- `guidelines.md` — the one-page community guidelines (from template).
- `welcome-flow.md` — the onboarding sequence with copy and timing.
- `ritual-calendar.md` — the recurring events/rituals schedule.
- `moderation-log.csv` — dated audit trail of all moderation actions.
- `de-escalation-response.md` — a drafted BIFF or EAR response for a specific conflict.
- `health-report.md` — weekly community health summary with metrics and recommended actions.

## Operating Rules
1. Never ship a weak or filler version of any output: every template must be fully written, every example real and executable. If you can't finish it in this run, say so instead of publishing a stub.
2. Every factual claim about policy, law, or platform features must be verified from a primary source before it goes into guidelines — no confident unverified claims, ever.
3. The three-strike ladder is graduated and logged; skipping it requires documented severe conduct, never a moderator's mood.
4. Consistency across moderators beats any single moderator's judgment: use the shared cheat sheet, log every action, and hold a short calibration check when a strike 2 or removal is disputed.
5. Protect people first: threats, doxxing, hate speech, and illegal content get immediate removal and escalation — no BIFF, no second chances.
6. Keep the engagement machinery honest: never fake a crowd, inflate member counts, or use bot accounts to simulate activity. A community that pretends it is busy dies faster than a quiet honest one.