mailbox

Benign intent indicators are strong (local filesystem IPC, no network/exfiltration primitives, no obfuscation/dynamic execution). However, the module has a notable security flaw: agent ids are used verbatim to form filesystem paths, enabling directory traversal/path injection that can escape the intended mailbox root and lead to arbitrary file read/write (and unintended archiving) if ids/paths are attacker-controlled. Treat this as a security alert for any deployment where callers are not fully trusted.