Add image-spam automod, server management, and infrastructure

[max-bromberg]

Summary

This is a major feature release that adds a comprehensive image-spam automod system, server-management tooling, database persistence, and Docker deployment support. The bot evolves from a simple tag-posting utility to a full-featured community moderation and automation platform.

Key Changes

Image-Spam Automod

• Perceptual-hash and metadata-signature based detection of spam images across channels
• Burst detection (same-channel rapid-fire) and fan-out detection (same image across channels)
• Blocklist system with moderator-confirmed spam fingerprints persisted to Postgres
• Auto-action on high-confidence signals (message deletion + timeout)
• Interactive moderation console with one-click action buttons for staff review

Server Management & Automation

• Member join/leave logging with invite-source attribution
• Auto-crosspost (publish) messages in announcement channels
• Role-select buttons for opt-in notification roles
• Message-link quoting (embeds linked messages inline)
• /say command for staff to broadcast rich embeds as the bot

Help Forum Integration

• /solved command and button to mark help posts solved and credit helpers
• Auto-posted "Mark Solved" button in configured help forums
• Tag suggestion system that prompts users with relevant tags based on error patterns

Database & Persistence

• Prisma ORM with Postgres adapter for blocklist and analytics persistence
• Automatic migrations on startup
• In-memory fallback when no database is configured (all features still work)
• Member analytics (join/leave events) recorded to database

Infrastructure

• Docker Compose setup with bundled Postgres service
• Automatic database initialization and migration on container startup
• Environment-driven feature flags (each feature self-disables when unconfigured)
• Comprehensive .env.example with sensible defaults

Code Organization

• Extracted shared embed builder to utils/embed.ts
• New utils/automod/ module with tracker, incidents, blocklist, and console logic
• New utils/resolveTag.ts for tag resolution and templating
• Interaction handlers for button-based moderation and role selection
• Listeners for message analysis, member events, and server automation

Documentation

• Completely rewritten README with feature highlights, quick-start guides, and configuration table
• Added badges for license, discord.js version, Sapphire framework, TypeScript, and Node.js

Notable Implementation Details

• Perceptual hashing: Uses difference hashing (dHash) via Jimp to detect re-encoded/resized image variants within a configurable Hamming distance
• Metadata signatures: Cheap, download-free fingerprints built from content-type, byte size, and dimensions for exact-match detection
• Incident tracking: In-memory ephemeral store of open moderation alerts; incidents expire on bot restart (buttons gracefully report expiry)
• Privilege checks: Automod actions and moderation buttons require ManageMessages permission; /say requires ManageGuild
• New member heuristic: Stricter burst threshold for accounts younger than newMemberWindowMs to catch join-then-spam patterns
• Invite attribution: Snapshots invite use counts on ready and diffs on join to determine which invite a member used
• Feature flags: Every optional feature (MOD_LOG_CHANNEL_ID, JOIN_LEAVE_LOG_CHANNEL_ID, etc.) self-disables when unset; no configuration errors

Version

Bumped from 0.3.0 to 1.0.0 to reflect the significant feature expansion and production-readiness of the automod system.

---

Update — text automod + helper-assist (commits acc7ab4–6d809b9)

Text-based automod

• Text-flooding detection — flags a user fragmenting one thought across many tiny messages (default 5 messages ≤ 25 chars in 15s). Alert-only by default; optional auto-timeout via AUTOMOD_FLOOD_AUTO_TIMEOUT. New utils/automod/flood.ts.
• Cross-channel question-spam detection — two tenure-aware signals: near-identical fan-out (token-overlap similarity across N+ channels) auto-deletes duplicate copies keeping the first; new-member content-agnostic "spread" alerts only. New utils/automod/crosspost.ts.
• Both surface in the same mod-log console with the existing action buttons.

Helper-assist (Phase 2)

• Unified suggestion engine handles keyword→tag, unformatted-code, and low-effort-ask detection with one reply and one per-user cooldown. Tag triggers co-located in tags.ts; coverage expanded from 3 to 9 topics.
• Auto-needinfo on thin help posts, stale-post nudge & auto-archive, /openposts ephemeral digest.
• Help channels generalised to accept forum or text channels via HELP_CHANNEL_IDS.

---

Update — pHash, deployment, and rollout hardening (commits fd89c40–c5cbcea)

DCT-based pHash (fd89c40)

Replaces dHash with a proper perceptual hash: 32×32 greyscale input → 2D DCT → top-left 8×8 low-frequency block compared against its own mean. Significantly more robust against re-encoding, recompression, watermarking, and minor crops — the exact transforms spammers apply to evade blocklists. Cosine table precomputed at module load; thumbnail request bumped to 64×64. Output format (16 hex chars, Hamming-comparable) and all downstream consumers are unchanged.

Docker-compose env coverage + Dockge setup guide (e80acb4)

The compose file was missing every env var added during this branch — all helper-assist, flood/crosspost automod, phash threshold, and server-management IDs. Variables not listed in the environment block are silently ignored even when present in .env, so none of the new features could be configured without this fix. Migration command switched from npx prisma to node_modules/.bin/prisma. New SETUP.md is a step-by-step Dockge deployment guide with Discord application setup, privileged-intent requirements, ID collection table, and a full env-var reference grouped by feature.

Helper-assist rollout hardening (74effe8, c5cbcea)

• Role-based suggestion exemption — tag/code/ask suggestions are suppressed for members holding any role listed in SUGGEST_IMMUNE_ROLE_IDS (Trusted, Knowledgeable, Helper, Moderator, etc.). Self-assignable notification roles are intentionally excluded so those members still receive suggestions. Uses explicit role ID matching rather than a blanket "any role" check.
• SUGGEST_IGNORE_CHANNEL_IDS — comma-separated channel IDs where all suggestions are suppressed (staff channels, announcements, etc.). Listing a forum channel's ID silences all its threads.
• HELP_AUTO_NEEDINFO defaults to false — must be explicitly opted in after observing the Mark Solved rollout.
• Tighter thin-post detection — the needinfo check now uses thread title + body combined (forum titles are often the whole question), plus four independent "has substance" signals: code block, inline backtick code, image attachment, any URL. Any one of these bypasses needinfo regardless of length.
• HELP_NEEDINFO_MIN_CHARS default raised to 120 (combined title + body, was 60 body-only).
• Single combined message — when auto-needinfo fires, the concise checklist and Mark Solved button are sent in one message instead of two. The full /tag needinfo content used by helpers via the context menu is unchanged.