Safe Email (Privacy-First, Extraction-Only)

Use this skill to safely process forwarded emails from a dedicated inbox via IMAP.

This skill is extraction-only:

• Read newest forwarded email (when explicitly asked)
• Extract structured details
• Return suggested next actions
• Do not write to calendar/reminder or other external systems automatically

What users must know first

1. Use a dedicated inbox (recommended: a brand-new Gmail account).
2. Forward target emails to that dedicated inbox before running this skill.
3. This skill does not auto-check inboxes; it runs only on explicit user instruction.

Required secrets/config (declared in metadata)

• SAFE_EMAIL_IMAP_USERNAME
• SAFE_EMAIL_IMAP_APP_PASSWORD

Policy:

• Provide secrets through secure runtime configuration.
• Never store plaintext credentials inside the skill package.

Security rules

1. Never auto-check email without explicit user instruction.
2. Read minimally: only newest relevant message for the request.
3. Deletion is optional and requires explicit consent.
4. Ask before ambiguous or destructive actions.

Setup guide (Gmail + IMAP)

1. Create a dedicated Gmail account for automation.
2. Enable 2-Step Verification.
3. Generate an App Password.
4. Configure IMAP client (example: Himalaya):
   • IMAP: imap.gmail.com:993 (TLS)
   • SMTP (optional if sending needed elsewhere): smtp.gmail.com:587 (STARTTLS)

Execution workflow

Step 0 — Require explicit trigger

Proceed only if user explicitly asks, e.g.:

• "I forwarded an email, process it."
• "Read the latest forwarded email."

Otherwise, stop.

Step 1 — Read newest relevant email only

• List recent inbox messages.
• Open only the newest relevant candidate.
• Do not bulk-read old/unrelated messages.

Step 2 — Extract structured details

Extract as available:

• sender
• subject/title
• date/time window (and timezone if present)
• location
• links
• key notes (confirmation numbers, seats, participants, etc.)
• actionable items

If time/timezone is ambiguous, ask user for confirmation.

Step 3 — Return extraction + suggested next actions

Return:

1. Structured summary
2. Confidence/ambiguities
3. Suggested next actions (examples):
   • "Create a calendar event"
   • "Create a reminder/task"
   • "Draft a reply"
   • "Archive/delete email"

Do not execute those actions unless user explicitly asks.

Step 4 — Optional email deletion (consent required)

If and only if user explicitly requests deletion:

1. Move processed email to Trash
2. Permanently expunge when supported
3. Report deletion status

If not requested, leave email untouched and state so clearly.

Failure handling

• If parsing fails: provide partial extraction + clarification questions.
• If deletion fails: report exact status and ask whether to retry.

Default privacy posture

• Explicit trigger only
• Minimal access scope
• No background surveillance behavior
• No automatic downstream writes
• Optional deletion with explicit consent