Duplicate Message

• Shuts down copy-paste advertising — the same crypto scam pasted across five channels trips the buffer once the user hits your threshold.
• Catches raid shouts and slogan-spam, where a group pastes an identical line over and over to drown out a channel.
• Stops single-user “say my line” repetition without needing a regex or word list.
• Very low false-positive rate on normal conversation — nobody types the exact same sentence four times by accident.

A single user posting the same exact message content enough times within their recent message buffer. Each message is fingerprinted with MD5 and dropped into a per-user deque that holds the last window_size hashes — once more than max_duplicates matching hashes sit in that deque, the rule fires on the offending send.

• Near-duplicates — adding a period, a trailing space, an emoji, or changing one character breaks the hash. "hello" and "hello " do not match each other.
• Case variants — comparison is case-sensitive. "Hello" and "hello" are different fingerprints.
• Cross-user duplicates — the buffer is keyed per user. Two accounts pasting the same text do not contribute to each other's counts.
• Cross-guild duplicates — state is per-guild. A user spamming the same text in two servers is tracked separately in each.
• Duplicates across a bot restart — the hash deque lives in memory only and resets on restart.
• Duplicate embeds, attachment contents, stickers, or reactions — only the message's content string is hashed.

A balanced baseline for most community servers:

{
  "max_duplicates": 3,
  "window_size": 10
}

Pair with action: delete and a per-user cooldown of 30s. Start in Mode: Log Only so you can see which users and phrases trip the rule before any message is removed. Flip to Live once the Event Log looks clean.

• One-word replies like yes, no, lol, or gg — enthusiastic users can hit the threshold during a hype moment.
• Greeting loops — gg, wp, ez across match endings can stack up quickly on a competitive server.
• Command retries typed as messages (rather than slash commands), like !play repeated because the bot is lagging.
• Emoji-only reactions typed as text, e.g. 😂 😂 😂.
• Translation or bridge bots that echo messages back into the same channel on behalf of a user.

• Current behavior: there is no time cutoff. The rule compares each new message against the user's last window_size messages — regardless of how long ago those messages were sent. A slow spammer who paces their duplicates over hours (or days) will still trip the rule as long as the copies stay inside the deque. window_size bounds message count, not elapsed time.
• Current behavior: the operator is strict greater-than. max_duplicates=3 means the 4th identical copy fires the rule — the 1st, 2nd, and 3rd silently pile up in the buffer. This matches every other volumetric trigger (spam_detection, mention_spam, caps_spam, emoji_spam, attachment_spam).
• Current behavior: content comparison is exact, including whitespace and case. "Hey" and "hey" are treated as different messages. "hey" and "hey " (trailing space) are also different. Determined users can bypass the rule by appending a random character each time.
• Current behavior: MD5 is used as the message fingerprint. This is a non-security use — MD5 is chosen for speed, not collision resistance. Duplicate detection does not need cryptographic strength.
• Current behavior: the hash deque is shared across all duplicate_message rules for the same (guild, user). There is one deque per user per guild. If you run two duplicate_message rules with different window sizes, the first rule evaluated for that user owns the deque's maxlen — the second rule's larger window will not retroactively grow it. In practice, running a single duplicate_message rule per server is the cleanest setup.
• Current behavior: minor variations bypass the match. Adding a punctuation mark, flipping one letter's case, inserting an emoji, or tacking on a space all produce a new hash and reset the streak. Stack with Spam Detection to catch fast-but-varied floods.
• Current behavior: the buffer is in-memory only and does not persist across bot restarts. Restarting Arkanis clears every user's duplicate-message deque, so a spammer mid-flood briefly gets a fresh count.

Pair this trigger with any of the five AutoMod actions. delete removes only the latest duplicate, not the earlier copies that piled up in the buffer — moderators may still want to sweep the earlier messages by hand, or rely on an escalated action like mute or ban to stop further sends.

• Mention Spam — another volumetric trigger. Use when harassment takes the form of repeated pings rather than repeated text.
• Spam Detection — catches message-rate floods regardless of content, so it covers the “fast but varied” case that Duplicate Message misses.
• Caps Spam — volumetric trigger for shouting. Pairs well with Duplicate Message against raid behaviour.
• Emoji Spam — volumetric trigger for emoji walls. Another layer on the same content type.
• AutoMod Actions — full reference on delete, warn, mute, strike, and ban.