Inbound spam/trust filtering

[longevityboris]

Problem

No spam filtering exists. All inbound email is treated equally. This is a production blocker for any public-facing agent inbox.

Risks:

• Spam counts against Resend's inbound quota (100/day free)
• Prompt injection via email body is a first-class threat for AI agents
• Autoresponders, list mail, and bounce notices look like real work

Proposal (layered)

Minimum viable (v1)

• Parse Authentication-Results headers (SPF/DKIM/DMARC pass/fail)
• Detect Auto-Submitted, List-Id, Precedence: bulk headers
• Allowlist/blocklist by sender domain
• Quarantine folder for suspicious messages

Future

• Integration with Rspamd or SpamAssassin
• ML-based classification
• Prompt injection detection in email bodies

Priority: HIGH — critical for public-facing agent inboxes