What problem does it solve? Agents running in a sandbox need a safe, controlled way to read and send email without exposing mailbox passwords, IMAP/SMTP credentials, or MXroute API access. This Skill provides a single CLI surface for all mailbox operations while treating inbound mail as untrusted input to prevent prompt injection. ## Core Features & Use Cases - Mailbox management: Check status, list folders and messages, read, mark read/unread, move, delete, and flag spam using UIDVALIDITY-safe operations. - Sending email: Send HTML (default for structured content), plain text, or multipart/alternative messages with CC recipients and file attachments via the agent-scoped control-plane API. - Security guardrails: Treats every inbound message as hostile input, forbids quoting inbound content into outbound mail, and blocks direct MXroute/IMAP/SMTP access. - Use Case: An autonomous agent checks its sandbox inbox for a verification code, reads the message safely, then sends an HTML status report with a PDF attachment to a user-specified recipient. ## Quick Start Ask the agent to check the sandbox inbox with tribes-email and summarize the latest unread messages.