Gatewatch reports security events. It does not replace SSH hardening, Fail2ban, firewall rules or an external availability monitor.

• Nodes use individual random 256-bit credentials. The hub stores SHA-256 digests, not recoverable node keys.
• A compromised node can forge its own events or flood its own identity until revoked. It cannot authenticate as another node without that node's key.
• The Telegram bot token stays on the hub. Bot commands require both the configured group ID and an explicit owner user ID. Anonymous admin messages are not authorized.
• Remote ingress requires HTTPS with certificate verification. The default listener is loopback-only. Deploy a TLS reverse proxy; do not put credentials in URLs or proxy logs.
• Batch size, body size, queues and deduplication records are bounded. An authenticated flood can exhaust shared capacity; use ingress rate limits and isolate hostile tenants in separate hubs.
• Account names and node names are HTML-escaped. Sources discard raw log messages. Password capture is off by default; an optional PAM integration can attach attempted passwords to confirmed SSH password failures.
• With PAM capture enabled, even a mistyped legitimate password may reach the hub and Telegram. Successful credentials are matched and discarded in memory. The C hook always returns PAM_IGNORE and uses nonblocking local IPC; custom PAM stacks require review. Backups and private queues can contain captured failed passwords.
• SQLite and configs contain personal/network metadata. Protect backups and local files. Linux installer uses an unprivileged system account and systemd filesystem restrictions.
• The same service user can read node configs on a host running both roles. For stricter tenant isolation, use separate hosts/accounts.
• Windows file modes do not provide Unix ACL isolation. Use your account's protected workspace for temporary configs.
• Delivery retries cannot guarantee exactly-once Telegram messages. Network failures after provider acceptance may create duplicates.
• Local source access can expose system logs. The installer grants journal/adm group access; source plugins are trusted code.
• Optional actions use a root-owned, socket-activated helper restricted to ban/unban in sshd. It checks peer UID and protects configured admin/server addresses. Only configured bot owners can request actions. Revoking a node stops new fetches; an executing command cannot be cancelled retroactively.

Report vulnerabilities privately to studio.rooni@gmail.com. Include affected version, reproduction and impact. Do not post real credentials or server addresses in public issues.