Skip to content

Tutorial

Jump to bottom
Daniel edited this page Feb 16, 2026 · 1 revision

ScamScreener Player Tutorial

This tutorial explains how to use ScamScreener in normal gameplay, step by step.

It is written for players (not developers) and focuses on practical in-game usage:

  • how to react to warnings
  • how to review and label chat lines
  • how to manage your blacklist
  • how to upload training data correctly
  • how to configure the mod for your playstyle

1. What ScamScreener Does

ScamScreener is a client-side Fabric mod for Hypixel SkyBlock chat safety.

It can:

  • detect suspicious chat patterns
  • warn you when risk is high
  • help you review messages manually
  • maintain a local blacklist
  • collect local training data for model improvements

Important:

  • It runs locally on your client.
  • It does not auto-ban players.
  • It does not replace your own judgment.

2. First Start (2 Minutes)

  1. Join a server where ScamScreener is active (for example Hypixel SkyBlock).
  2. Open chat and run:
    • /ss
  3. The main ScamScreener settings menu opens.

/ss is a full alias of /scamscreener.

If you prefer the long form, all commands also work with /scamscreener.

3. Quick Start Setup (Recommended)

Open /ss and configure these first:

  1. Alert Threshold
    • Start with MEDIUM for balanced warnings.
  2. AI Auto-Capture
    • Start with OFF or HIGH if you want conservative automatic sample capture.
  3. Auto /p leave on blacklist
    • Enable if you want instant party leave when a blacklisted player is detected.
  4. Mute Filter
    • Enable only if you already use mute patterns.
  5. Local AI Signal
    • Keep enabled unless you are troubleshooting.

4. Understanding Warning Messages

When ScamScreener detects risk, you will see a warning message in chat.

Typical warning content:

  • player name
  • risk level/score
  • action tags: [manage] and [info]

What to do:

  1. Click [info] to inspect why the alert triggered.
  2. Click [manage] to review and label relevant lines.

You may also see a [BYPASS] action when your own outgoing message is blocked for safety reasons (email, Discord invite, risky coop add flow).

5. Review Workflows (Core Feature)

ScamScreener has multiple review modes.

5.1 Alert Review ([manage])

Use this after a live warning.

In the review screen:

  • click a row to cycle its state:
    • [I] = ignored
    • [S] = scam
    • [L] = legit
  • top summary shows selected counts
  • Save writes selected labels
  • Save & Upload writes labels, then runs upload flow

In this alert-based mode, two optional checkboxes are available:

  • add player to ScamScreener blacklist
  • add player to Hypixel /block list

5.2 Logged Chat Review (/ss review)

Run:

  • /ss review

This opens a review screen for logged chat lines that are not saved yet.

Use it to label chat even if no alert tag is currently visible.

Behavior:

  • same [I]/[S]/[L] row interaction
  • no blacklist/block checkboxes (no player-name action mode)
  • Save / Save & Upload available

Help for legacy review command usage:

  • /ss review help

5.3 Player-Specific Review (/ss review player <playerName>)

Run:

  • /ss review player <playerName>

Autocomplete is available and filtered:

  • player must be online
  • player must have logged chat entries

Use this when you want to focus on one specific player's recent conversation context.

5.4 Training CSV Review (Menu)

Open:

  1. /ss
  2. click Review Training CSV

This screen shows existing rows from your training CSV directly.

Actions:

  • Open File opens the CSV file in your operating system
  • Save applies selected changes
  • Save & Upload applies changes and starts upload flow

Important CSV behavior:

  • duplicate messages are handled row-by-row
  • if a row is set to ignored and saved, that row is removed

6. Training Data and Upload

6.1 Where training data is stored

Main file:

  • config/scamscreener/scam-screener-training-data.csv

6.2 Labeling paths

You can create labels through:

  • alert manage review
  • /ss review (logged unsaved lines)
  • training CSV review screen
  • advanced command path: /ss ai flag <messageId> <legit|scam>

6.3 Upload flow

Run:

  • /ss upload

Or use:

  • Upload Training Data button in /ss menu
  • Save & Upload in review screens

Upload behavior:

  • first upload asks for ToS confirmation
  • data is archived before upload
  • webhook upload starts after local preparation

6.4 Large file reminder

If your active training CSV reaches at least 500 entries:

  • you get a reminder immediately
  • then every 5 minutes while size stays >= 500
  • reminder includes a clickable upload action

7. Settings Menu Guide

Open with:

  • /ss

7.1 Main buttons

  • Alert Threshold
  • AI Auto-Capture
  • Auto /p leave on blacklist
  • Mute Filter
  • Local AI Signal

7.2 Secondary screens

  • Rule Settings: enable/disable individual detection rules
  • Message Settings: toggle warning messages/sounds
  • Blacklist: edit entries, score, reason, remove, paging
  • AI Update: check/download and apply model update (Accept, Merge, Ignore)
  • Metrics: local funnel metrics summary + copy button
  • Debug Settings: advanced diagnostics (usually keep off)

7.3 Training actions in menu

  • Review Training CSV
  • Upload Training Data

8. Command Reference (Player-Focused)

You can use either /ss ... or /scamscreener ....

8.1 Daily commands

  • /ss -> open settings menu
  • /ss help -> command overview
  • /ss review -> review logged unsaved chat lines
  • /ss review help -> review command help
  • /ss upload -> archive + upload training data
  • /ss list -> show blacklist entries
  • /ss add <player> [score] [reason] -> add/update blacklist entry
  • /ss remove <player> -> remove blacklist entry

8.2 Review commands

  • /ss review manage <alertId>
  • /ss review info <alertId>
  • /ss review player <playerName>

8.3 Protection and behavior commands

  • /ss alertlevel [low|medium|high|critical]
  • /ss autoleave [on|off]
  • /ss mute [pattern]
  • /ss unmute [pattern]
  • /ss bypass <id>

8.4 AI and maintenance commands

  • /ss ai
  • /ss ai flag <messageId> <legit|scam>
  • /ss ai migrate
  • /ss ai update
  • /ss ai update force
  • /ss ai update notify [on|off]
  • /ss ai model <download|accept|merge|ignore> <id>
  • /ss ai metrics [reset]
  • /ss ai reset
  • /ss ai autocapture [off|low|medium|high|critical]

8.5 Utility commands

  • /ss settings
  • /ss version
  • /ss preview

9. Recommended Daily Routine

  1. Play normally.
  2. When a warning appears, inspect [info].
  3. Label with [manage] if you can classify confidently.
  4. Use /ss review between sessions to clear pending logged lines.
  5. Use Review Training CSV occasionally for cleanup.
  6. Upload when dataset gets large (or when reminder appears).

10. Troubleshooting

"No review lines appear"

  • Wait for actual player chat lines (system lines are filtered).
  • Use /ss review after chat activity.
  • For /ss review player, target must be online and have logged entries.

"Upload does not start"

  • Confirm upload ToS when prompted.
  • Check that scam-screener-training-data.csv exists.
  • Read chat error code/details if upload failed.

"Too many warnings"

  • Raise alert threshold (/ss alertlevel high).
  • Disable noisy rules in Rule Settings.
  • Review and clean training data quality.

"Not enough warnings"

  • Lower alert threshold (/ss alertlevel low or medium).
  • Ensure Local AI Signal is enabled in settings.
  • Confirm message warnings are enabled in Message Settings.

11. Privacy and Network Behavior

ScamScreener is client-side first.

Local by default:

  • detection
  • review
  • blacklist management
  • training data storage

Network is only used for optional features:

  • Mojang name/UUID lookups
  • AI model update check/download
  • training data upload when you explicitly trigger it

12. Final Tips

  • Label only what you are confident about.
  • Mixed, high-quality scam + legit labels improve model quality more than bulk noisy labels.
  • Use ignored when context is unclear.
  • Review often, upload periodically.

Clone this wiki locally