ZipGuard roadmap

Public and honest. "Live" means running in the pilot today. "Building" means code is being written this week. Nothing here is a promise of a date; the order is the order.

Live in the pilot

• Join screening from the join event only, every verdict explained rule by rule
• One-click challenge inside Discord for medium risk; hold for a moderator at high risk
• Moderator Approve, Kick and Ban buttons on every log entry; member appeals from inside Discord
• /my-data, /forget-me and /privacy for every member
• Shadow mode for the first 7 days; per-server thresholds and high-risk action
• Anti-nuke: three channel or role deletes by one account inside a minute are reverted and the account loses its roles
• Link defense: lookalike domains, scam wording, the same link in three channels inside a minute (needs the Message Content permission, each server turns it on)
• Weekly scorecard digest in the log channel
• Data statement generated from the database schema, with a deletion test in CI
• Private pilot application form and approval flow
• Raid mode: after a join burst (10 joins in a minute by default), every joiner is held for 10 minutes and invites are paused
• Right-click moderator actions on a member: show the verdict with its reasons, or hold for review
• Shared ban list across servers with the same owner, opt-in, stored as salted hashes
• Member reports (/report): three reports from established accounts inside 24 hours hold the member automatically
• Breach notification runbook: every affected server owner, member and Discord told within 72 hours
• Published review of the Discord Developer Policy and Terms against what the bot does
• Encryption at rest: the bot database lives on a LUKS-encrypted volume, and the bot refuses to start without it
• Read-only dashboard: /zipguard dashboard gives moderators a 24-hour link with counts, rule frequencies and settings, never member data

Building now

• Terms of service page

Planned before public launch

• Public repository under the MIT license, after an external security review
• External penetration test and a public bug bounty with real payouts

Later

• Opt-in trust proofs: a member may link Steam, GitHub or Twitch to skip review; only a salted hash is kept
• Dashboard rule editing
• A trained scoring model once moderator overrides exist, with the rules kept as guardrails
• A real scam classifier built on Jebadiah decision models (jebadiah.ai) instead of the phrase list, trained without Discord message content unless Discord grants permission

Never

• IP addresses, browser or device fingerprints, an external verification page
• Ads, a paid tier, data resale, people search