Security & Safety

How mIRCscord protects servers, users, and data from malicious bots and abuse. Every layer is enforced server-side — bots cannot bypass these protections.

Permission System

Only server owners and admins (users with the Administrator permission bit) can:

Install or remove bots from a server
Configure bot settings (blocked words, welcome messages, AI prompts)
Set or remove API keys for AI bots
Enable or disable installed bots

Regular members cannot modify bot configuration. This is enforced server-side — the API rejects unauthorized requests with 403 Forbidden.

Bot Sandboxing

What bots CAN do:

Read messages in channels they're configured for
Send messages as bot responses
Delete messages (AutoMod only, for moderation)
React to messages

What bots CANNOT do:

Access other servers — bots are scoped per-server installation
Read DMs or encrypted messages
Modify server settings, roles, or channels
Access user passwords, tokens, or API keys from other bots
Make external network requests (except AI bots to their configured provider)
Execute code or access the filesystem
Mention @everyone or @here (stripped automatically)

AI Bot Safety

Input Sanitization AUTO-ENFORCED

All user messages are sanitized before being sent to the AI provider. Common prompt injection patterns are filtered:

ignore previous instructions
disregard previous
you are now
system: / assistant: role injections
[INST] / <<SYS>> template injections

Output Sanitization AUTO-ENFORCED

AI responses are capped at 2000 characters
@everyone and @here mentions are neutralized with zero-width spaces
Responses are posted as the bot user, never as a real user

Rate Limiting AUTO-ENFORCED

Max 1 AI response per 5 seconds per channel
Max 3 API key changes per hour per server
Every AI call is logged with timestamp and token count

API Key Security

Keys are stored encrypted in the database
Keys are never displayed in the UI after entry (only last 4 chars shown as hint)
Each server's key is isolated — one server's key cannot be used by another
Keys are validated on entry: format checks, minimum length, provider-specific prefixes
Usage is tracked per key (call count, last used timestamp)

AutoMod Safety

AutoMod never moderates server owners or admins — admin messages always pass through
Every moderation action is logged to the audit trail with: timestamp, channel, target user, reason
Config validation enforced: max 500 blocked words, each max 100 chars

Config Validation

All bot configuration is validated server-side before saving:

Field	Limit	Validation
`blocked_words`	500 items	Array of strings, each max 100 chars
`welcome_message`	1000 chars	String with {username} placeholder
`system_prompt`	4000 chars	String (AI bots only)
`active_channels`	20 channels	Array of valid UUID strings
`block_links`	—	Boolean only
`block_caps`	—	Boolean only
`api_key`	—	Min length, provider prefix validation
`provider`	—	Must be: openrouter, openai, anthropic, google, mistral, groq

Audit Trail

Every bot action is logged to bot_actions_log with:

Bot install ID (which bot on which server)
Action type (automod_spam, automod_blocked_word, ai_response, etc.)
Channel ID where the action occurred
Target user ID (who was moderated)
Details (JSON with specifics: word matched, caps ratio, etc.)
Timestamp

Server admins will be able to view the audit log in server settings (coming soon).

Platform-Level Security

These protections apply to the entire platform, including bots:

Per-IP rate limiting — Auth: 5/min, General: 120/min, Uploads: 10/min
WebSocket rate limiting — Max 30 messages/second per connection, 64KB max frame
Body size guard — 256KB max for non-upload requests
JWT authentication — All API calls require valid tokens (7-day expiry)
bcrypt password hashing — Default cost factor
Parameterized SQL — All queries use bind parameters (no SQL injection)
CORS restricted — Only mircscord.com and localhost origins
Security headers — X-Frame-Options, CSP, X-Content-Type-Options, Referrer-Policy