Your sales agent booked a meeting
while you were in one.

Check suppressions and fatigue

Before you write a single word, check if you're even allowed to email this person. Suppressions catch bounces, unsubscribes, and complaints. Fatigue checks cross-agent send history. If either says no, stop here.

molted suppressions list --email jake@stackline.dev
# → { "suppressed": false, "entries": [] }

molted analytics fatigue --contact jake@stackline.dev
# → { "recommendation": "safe_to_send", "recentSends": 0,
#     "windows": { "24h": 0, "7d": 0, "30d": 0 } }

Propose the cold email

This is where AI does the writing. molted send propose takes a recipient, a template, and a context blob describing what you know about the prospect. It returns a draft subject line and body that your agent can review, tweak, or send as-is. The context matters here — the more specific you are about why you're reaching out (their recent Series A, the DevTools space, a specific pain point), the less the email reads like a mail merge. Think of it as giving the AI a brief, not a blank canvas.

molted send propose \
  --to jake@stackline.dev \
  --template cold-outreach \
  --mailbox sales-agent \
  --context '{
    "prospect_name": "Jake",
    "company": "Stackline",
    "industry": "DevTools",
    "funding": "Series A, $12M",
    "pain_point": "scaling outbound without spam complaints",
    "sender_name": "Your Name",
    "sender_role": "Head of Partnerships"
  }'
# → { "subject": "Quick question about Stackline's outbound",
#     "body": "Hey Jake, saw Stackline just closed the Series A — congrats.
#       We work with a few DevTools companies scaling outbound and keep
#       hearing the same thing: deliverability tanks once volume goes up.
#       Worth a 15-min chat to see if we can help?",
#     "confidence": 0.87 }

Simulate and send

Dry-run the message through all policy rules first. If it passes, send it with a deterministic dedupe key. The pattern for outbound is outreach-{contact}-{sequence}-{step} — if your agent crashes and retries, or a network timeout causes a double call, the same key prevents a duplicate. No prospect should ever get the same cold email twice.

molted send simulate \
  --to jake@stackline.dev \
  --template cold-outreach \
  --mailbox sales-agent
# → { "status": "allowed", "rulesEvaluated": 22,
#     "passed": 22, "blocked": 0 }

molted send \
  --to jake@stackline.dev \
  --template cold-outreach \
  --dedupe-key "outreach-jake-stackline-cold-1" \
  --mailbox sales-agent \
  --reason "Series A DevTools company, matches target segment" \
  --payload '{
    "name": "Jake",
    "company": "Stackline",
    "sender_name": "Your Name"
  }'
# → { "status": "sent", "requestId": "req_def456" }

Schedule follow-ups

If Jake doesn't reply in 48 hours, send follow-up one. Another 72 hours, follow-up two. The --cancel-on-reply flag is the whole point: the moment Jake responds, every pending follow-up in the sequence gets cancelled. No awkward "just bumping this" after he already said yes.

molted threads followup req_def456 \
  --contact-email jake@stackline.dev \
  --delay-hours 48 \
  --cancel-on-reply

molted threads followup req_def456 \
  --contact-email jake@stackline.dev \
  --delay-hours 120 \
  --cancel-on-reply

Handle the reply

Jake replies: "Yeah, we're struggling with this actually. Can you do Thursday?" Three CLI calls handle the rest. First, classify his intent. Then check the recommended next action. Then respond. The interesting part is what happens when the intent isn't a meeting request. If Jake says "not right now, maybe next quarter," the next-action recommends a 90-day snooze instead of a reply. If he says "please stop emailing me," it recommends suppression — and your agent should honor that immediately.

molted classify \
  --subject "Re: Quick question about Stackline's outbound" \
  --body "Yeah, we're struggling with this actually. Can you do Thursday?"
# → { "intent": "interested", "confidence": 0.95 }

molted next-action --contact jake@stackline.dev
# → { "action": "reply", "reason": "prospect expressed interest,
#     respond with calendar link within SLA" }

molted threads reply thr_sales_001 \
  --body "Thursday works great. Here's a link to grab 30 minutes:
  https://cal.com/yourteam/sales-intro"

# What if Jake said "not right now"?
# molted classify → { "intent": "not_now", "confidence": 0.88 }
# molted next-action → { "action": "snooze", "delayDays": 90 }

# What if Jake said "stop emailing me"?
# molted classify → { "intent": "unsubscribe", "confidence": 0.97 }
# molted next-action → { "action": "suppress", "reason": "explicit opt-out" }

Record the outcome

Jake takes the meeting. Two weeks later, the deal closes. Record both. Molted traces the revenue back to the cold email that started the thread, so you can measure whether your sales agent is actually generating pipeline or just sending mail.

molted outcomes ingest \
  --contact jake@stackline.dev \
  --event-type meeting_booked \
  --event-name "Intro Call Booked" \
  --metadata '{ "source": "cold_outreach", "days_to_book": 1 }'

molted outcomes ingest \
  --contact jake@stackline.dev \
  --event-type deal_closed \
  --event-name "Stackline - Starter Plan" \
  --revenue 2400 \
  --metadata '{ "plan": "starter", "deal_cycle_days": 15 }'

molted outcomes journey-impact --mailbox sales-agent
# → { "meetings_booked": 34, "deals_closed": 8,
#     "revenue": 52800, "attribution": "last-touch",
#     "avg_deal_cycle_days": 18 }