Add moderation system with deterministic detection and strike ladder

Deterministic-only detection (no LLM): word list with precompiled
leetspeak variation matching, text/image flood, wall of text, repeated
messages (Levenshtein similarity), mention flooding, link rate limiting
for new members, invite flooding, join/leave cycling detection.

Three-strike response ladder: warn + redact → temp mute → permanent ban.
Strikes expire after configurable period. Admin room notifications with
context cards. DMs over public callouts.

Admin commands: !mod warn/mute/unmute/ban/strikes/forgive/history/reload/
status/test. All require ADMIN_USERS membership.

Feature-flagged: disabled by default, set FEATURE_MODERATION=true to
enable. All thresholds configurable via env vars. New member grace
period with stricter thresholds. Word list hot-reload via !mod reload.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

.env.example

@@ -50,6 +50,34 @@ MISSING_EXCLUDE_USERS=        # comma-separated user IDs to never list as missin
 # ---- Encryption ----
 QUOTE_ENCRYPTION_KEY=         # Base64-encoded 32-byte AES-256 key (required for !quote)
 
+# ---- Moderation (disabled by default) ----
+FEATURE_MODERATION=            # set to "true" to enable the moderation system
+MOD_WORDLIST=                  # path to newline-delimited word list file
+MOD_WORDLIST_VARIATIONS=true   # check leetspeak/spaced variants
+MOD_STRIKE_EXPIRY_DAYS=30      # strikes expire after N days
+MOD_MUTE_DURATION_MINUTES=60   # temp mute duration on strike 2
+MOD_MAX_STRIKES=3              # strikes before permanent ban
+MOD_FLOOD_MESSAGE_COUNT=5      # messages within window = flood
+MOD_FLOOD_MESSAGE_WINDOW_SECONDS=10
+MOD_FLOOD_IMAGE_COUNT=3        # images/files within window = flood
+MOD_FLOOD_IMAGE_WINDOW_SECONDS=30
+MOD_MAX_MESSAGE_LENGTH=2000    # max chars per message (0 = disabled)
+MOD_REPEAT_COUNT=3             # same message N times = strike
+MOD_REPEAT_WINDOW_SECONDS=60
+MOD_REPEAT_SIMILARITY_THRESHOLD=0.85
+MOD_MENTION_MAX=5              # max unique @mentions per message
+MOD_MENTION_FLOOD_COUNT=3
+MOD_MENTION_FLOOD_WINDOW_SECONDS=30
+MOD_LINK_RATE_NEW_MEMBER=3     # max links per window (new members only)
+MOD_LINK_RATE_WINDOW_SECONDS=60
+MOD_INVITE_MAX_PER_HOUR=5
+MOD_JOIN_LEAVE_COUNT=3         # join/leave cycles = flag (no auto-strike)
+MOD_JOIN_LEAVE_WINDOW_MINUTES=10
+MOD_NEW_MEMBER_DAYS=14         # accounts newer than N days = stricter thresholds
+MOD_NEW_MEMBER_FLOOD_MULTIPLIER=0.5
+MOD_ADMIN_ROOM=                # dedicated admin room for mod notifications
+MOD_DM_ON_ACTION=true          # DM users when action is taken
+
 # ---- Rate Limits (per user per day) ----
 RATELIMIT_WEATHER=5
 RATELIMIT_TRANSLATE=20

README.md

@@ -29,7 +29,8 @@ Written in Go using [mautrix-go](https://github.com/mautrix/go) for encryption a
 
 - **E2EE that actually works** - mautrix-go with goolm (pure Go). Crypto state lives in SQLite so device keys survive restarts. Cross-signing bootstraps on first run — the bot self-verifies its own device.
 - **No CGo, no system deps** - builds to a single static binary. Cross-compile to whatever you want.
-- **37 plugins** with dependency injection and ordered registration
+- **38 plugins** with dependency injection and ordered registration
+- **Moderation system** (optional) - deterministic detection only, no LLM. Word list with leetspeak variation matching, text/image flood, repeated messages, mention flooding, link rate limiting, invite flooding, join/leave cycling. Three-strike ladder (warn → mute → ban). Admin room notifications, DMs over public callouts.
 - **Passive tracking** - XP, stats, streaks, achievements, markov corpus, keyword alerts, all running silently
 - **Scheduled posts** via [robfig/cron](https://github.com/robfig/cron) - WOTD, holidays, game releases, birthdays, anime/movie releases, concert digests, esteemed members
 - **LLM integration** (optional) - Ollama-powered sentiment analysis, roast profiles, room vibes, tarot readings, conversation summaries
@@ -136,6 +137,39 @@ Everything is configured through environment variables or a `.env` file.
 | `FEATURE_TRIVIA` | Set to `false` to disable trivia (default: enabled) |
 | `FEATURE_ESTEEMED` | Set to anything to enable satirical esteemed member posts |
 | `ESTEEMED_ROOM` | Room ID for esteemed member posts (separate from broadcast rooms) |
+| `FEATURE_MODERATION` | Set to `true` to enable the moderation system (disabled by default) |
+
+### Moderation
+
+All moderation settings are optional. The system is disabled unless `FEATURE_MODERATION=true`.
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `MOD_WORDLIST` | | Path to newline-delimited prohibited word list |
+| `MOD_WORDLIST_VARIATIONS` | `true` | Check leetspeak/spaced variants |
+| `MOD_STRIKE_EXPIRY_DAYS` | `30` | Days before strikes expire |
+| `MOD_MUTE_DURATION_MINUTES` | `60` | Temp mute duration on strike 2 |
+| `MOD_MAX_STRIKES` | `3` | Strikes before permanent ban |
+| `MOD_FLOOD_MESSAGE_COUNT` | `5` | Messages in window = flood |
+| `MOD_FLOOD_MESSAGE_WINDOW_SECONDS` | `10` | Text flood window |
+| `MOD_FLOOD_IMAGE_COUNT` | `3` | Images/files in window = flood |
+| `MOD_FLOOD_IMAGE_WINDOW_SECONDS` | `30` | Image flood window |
+| `MOD_MAX_MESSAGE_LENGTH` | `2000` | Max chars per message (0 = disabled) |
+| `MOD_REPEAT_COUNT` | `3` | Repeated messages before strike |
+| `MOD_REPEAT_WINDOW_SECONDS` | `60` | Repeat detection window |
+| `MOD_REPEAT_SIMILARITY_THRESHOLD` | `0.85` | How similar counts as "same" (0.0–1.0) |
+| `MOD_MENTION_MAX` | `5` | Max unique @mentions per message |
+| `MOD_MENTION_FLOOD_COUNT` | `3` | Mention-heavy messages in window |
+| `MOD_MENTION_FLOOD_WINDOW_SECONDS` | `30` | Mention flood window |
+| `MOD_LINK_RATE_NEW_MEMBER` | `3` | Max links per window (new members only) |
+| `MOD_LINK_RATE_WINDOW_SECONDS` | `60` | Link rate window |
+| `MOD_INVITE_MAX_PER_HOUR` | `5` | Max room invites per user per hour |
+| `MOD_JOIN_LEAVE_COUNT` | `3` | Join/leave cycles before flag |
+| `MOD_JOIN_LEAVE_WINDOW_MINUTES` | `10` | Join/leave window |
+| `MOD_NEW_MEMBER_DAYS` | `14` | Days before a member is no longer "new" |
+| `MOD_NEW_MEMBER_FLOOD_MULTIPLIER` | `0.5` | Multiply flood thresholds for new members |
+| `MOD_ADMIN_ROOM` | | Dedicated room for mod notifications |
+| `MOD_DM_ON_ACTION` | `true` | DM users when action is taken |
 
 ### Space Groups
 
@@ -337,6 +371,20 @@ Rep is earned when someone thanks you. The bot detects this automatically.
 | `!tarot [@user]` | Draw a tarot card + LLM reading |
 | `!tarotspread [@user]` | Three-card spread (Past/Present/Future) |
 
+### Moderation (admin only, requires `FEATURE_MODERATION=true`)
+| Command | Description |
+|---------|-------------|
+| `!mod warn @user [reason]` | Issue a manual warning (counts as a strike) |
+| `!mod mute @user [duration]` | Manual mute (does not consume a strike) |
+| `!mod unmute @user` | Remove mute |
+| `!mod ban @user [reason]` | Permanent ban |
+| `!mod strikes @user` | Show active strikes |
+| `!mod forgive @user` | Clear all active strikes |
+| `!mod history @user` | Full moderation history |
+| `!mod reload` | Reload word list from disk |
+| `!mod status` | Show current moderation config |
+| `!mod test @user` | Simulate next violation without taking action |
+
 ### Other
 | Command | Description |
 |---------|-------------|
@@ -526,6 +574,7 @@ gogobee/
 │   │   ├── tarot.go         # LLM-powered tarot readings
 │   │   ├── horoscope.go     # Daily horoscopes
 │   │   ├── esteemed.go      # Satirical esteemed member posts
+│   │   ├── moderation.go   # Moderation system (strikes, word list, flood detection)
 │   │   └── ratelimits.go    # Rate limiting
 │   └── util/
 │       ├── auth.go          # Matrix login, token check

internal/db/db.go

@@ -621,6 +621,31 @@ CREATE TABLE IF NOT EXISTS api_cache (
 	cached_at INTEGER DEFAULT (unixepoch())
 );
 
+-- Moderation: strikes
+CREATE TABLE IF NOT EXISTS mod_strikes (
+	id           INTEGER PRIMARY KEY AUTOINCREMENT,
+	user_id      TEXT NOT NULL,
+	room_id      TEXT NOT NULL,
+	issued_at    DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP,
+	expires_at   DATETIME NOT NULL,
+	reason       TEXT NOT NULL,
+	issued_by    TEXT NOT NULL,
+	active       BOOLEAN NOT NULL DEFAULT TRUE
+);
+CREATE INDEX IF NOT EXISTS idx_mod_strikes_user ON mod_strikes(user_id, issued_at);
+
+-- Moderation: action log
+CREATE TABLE IF NOT EXISTS mod_actions (
+	id           INTEGER PRIMARY KEY AUTOINCREMENT,
+	user_id      TEXT NOT NULL,
+	room_id      TEXT NOT NULL,
+	action       TEXT NOT NULL,
+	reason       TEXT,
+	taken_at     DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP,
+	taken_by     TEXT NOT NULL
+);
+CREATE INDEX IF NOT EXISTS idx_mod_actions_user ON mod_actions(user_id, taken_at);
+
 -- Space groups (rooms with overlapping membership)
 CREATE TABLE IF NOT EXISTS space_groups (
 	room_id    TEXT PRIMARY KEY,

main.go

@@ -71,6 +71,10 @@ func main() {
 
 	// ---- Register plugins in dependency order ----
 
+	// Moderation (runs first in dispatch order)
+	modPlugin := plugin.NewModerationPlugin(client)
+	registry.Register(modPlugin)
+
 	// Foundation (passive tracking)
 	xpPlugin := plugin.NewXPPlugin(client)
 	registry.Register(xpPlugin)
@@ -154,10 +158,15 @@ func main() {
 
 	syncer := client.Syncer.(*mautrix.DefaultSyncer)
 
-	// Auto-join on invite
+	// Auto-join on invite + moderation member tracking
 	syncer.OnEventType(event.StateMember, func(ctx context.Context, evt *event.Event) {
+		mem := evt.Content.AsMember()
+		if mem == nil {
+			return
+		}
+
+		// Auto-join invites for the bot
 		if evt.GetStateKey() == string(client.UserID) {
-			mem := evt.Content.AsMember()
 			if mem.Membership == event.MembershipInvite {
 				_, err := client.JoinRoomByID(ctx, evt.RoomID)
 				if err != nil {
@@ -166,7 +175,12 @@ func main() {
 					slog.Info("joined room", "room", evt.RoomID)
 				}
 			}
+			return
 		}
+
+		// Track join/leave/invite for moderation
+		targetUser := id.UserID(evt.GetStateKey())
+		modPlugin.OnMemberEvent(evt.RoomID, targetUser, mem.Membership)
 	})
 
 	// Message handler