Multi-Channel Writer

Input

• Contact data: name, title, company, email (yes/no), LinkedIn URL (yes/no)
• Signal type: github | arxiv | hiring | funding | huggingface
• Signal payload: specific detail (repo name, paper title, job title, etc.)
• Company research from prospect-researcher

Step 1 — Determine Channels

Call select_primary_channel(has_email, has_linkedin) from scripts/multi_channel_sequencer.py.

Step 2 — Build Sequence

Call build_sequence(channels, signal_type). This returns the ordered SequenceStep list with days, channels, actions, and template names.

Step 3 — Generate Copy

For each step, write the actual message using the mapped template:

Email templates → config/templates/email-sequences/ LinkedIn templates → config/templates/linkedin-sequences/

Fill all placeholders from the signal payload and contact data. Never leave a placeholder unfilled.

Connection request notes must be ≤300 characters. Count before outputting.

Step 4 — Present the Full Plan

Output a single table showing the complete sequence:

Then present the full copy for each step in order.

Voice Rules

• Email: 4 sentences max on Day 0. Reference the specific signal. Low-friction CTA.
• LinkedIn connection: ≤300 chars, zero product mention, peer-to-peer tone.
• LinkedIn follow-ups: lead with a resource or insight, no pitch unless they ask.
• Never: "I hope this email finds you well", "touching base", "I noticed your company".

Quality Checklist

Before outputting, verify:

• Connection request notes are ≤300 characters
• Day 0 email is ≤4 sentences
• No placeholder left as {{placeholder}}
• No product mention in LinkedIn connection note
• Steps are in ascending day order
• Each step references the specific signal, not generic domain interest