Lovelyn

Shipped customs for @NousResearch Hermes Agent. Most security tools are built to detect and report suspicious activity. customs takes a slightly different approach: it lays traps. Think of marked banknotes that are used to identify stolen money when it resurfaces. customs borrows that same decoy principle, but it doesn't wait for a resurface. It plants fake credentials around your machine and waits. When an agent or script reaches for one, customs catches the attempt and blocks the tool call before execution. There's also a blend feature that plants decoys in randomized locations, using existing folder structures on your machine. Those locations are chosen locally, so an attacker can't simply inspect the source code to find a list of fixed decoy paths. customs also reads scripts before they run and checks what Hermes is about to execute. It doesn't need the AI model to recognise the attack or decide whether to stop itself. customs is entirely local and makes no network calls by default. Decoys, detection and incident records stay on your machine. Telegram alerts are available, but by opt-in. One important distinction: customs catches attempts that reach its decoys. It doesn't watch your real credentials. It's built to catch the hunt, not just report what happened after the fact. Try it today: hermes plugins install pgreyy/customs Code: github.com/pgreyy/customs Hermes catalog submission: github.com/NousResearch/herm…

Shipped customs for @NousResearch Hermes Agent. Most security tools are built to detect and report suspicious activity. customs takes a slightly different approach: it lays traps. Think of marked banknotes that are used to identify stolen money when it resurfaces. customs borrows that same decoy principle, but it doesn't wait for a resurface. It plants fake credentials around your machine and waits. When an agent or script reaches for one, customs catches the attempt and blocks the tool call before execution. There's also a blend feature that plants decoys in randomized locations, using existing folder structures on your machine. Those locations are chosen locally, so an attacker can't simply inspect the source code to find a list of fixed decoy paths. customs also reads scripts before they run and checks what Hermes is about to execute. It doesn't need the AI model to recognise the attack or decide whether to stop itself. customs is entirely local and makes no network calls by default. Decoys, detection and incident records stay on your machine. Telegram alerts are available, but by opt-in. One important distinction: customs catches attempts that reach its decoys. It doesn't watch your real credentials. It's built to catch the hunt, not just report what happened after the fact. Try it today: hermes plugins install pgreyy/customs Code: github.com/pgreyy/customs Hermes catalog submission: github.com/NousResearch/herm…