write-outreach

name	write-outreach
description	Write personalised cold outreach — email and LinkedIn — using Saber signal results and contact data as personalisation context.

Write Outreach

Use this skill to write a personalised cold email or LinkedIn message for a specific contact, using Saber signal results (company-level and contact-level) as the personalisation engine.

Goal

Produce 1–3 outreach variants (cold email, LinkedIn connection request, LinkedIn InMail) that reference specific, timely signals rather than generic opener lines.

Step 1 — Gather contact context

Ask for or confirm from conversation context:

• Contact name and title
• Company name and domain
• LinkedIn URL (if available — needed for contact-level signals)
• Your name and company (the sender)

Step 2 — Pull signal data from Saber

Signal results are the personalisation fuel. Pull as much as is available.

If the Saber CLI is available (saber --help works):

Company-level signals — check for existing subscription results for this domain:

saber subscription list
saber subscription get <subscriptionId>   # for any subscription covering this company

If no results exist for this company, offer to run a quick spot-check:

saber signal --domain <domain> --question "<question>" --answer-type boolean

Suggested spot-check questions if no signals are pre-defined:

• "Is this company actively hiring in sales or revenue roles?"
• "Has this company raised funding in the last 12 months?"
• "Is this company expanding into new markets or geographies?"

Contact-level signals — if a LinkedIn URL is available:

saber signal --profile <linkedin-url> --question "Is this person posting about [relevant topic]?"
saber signal --profile <linkedin-url> --question "Has this person recently changed roles or been promoted?"

If the Saber CLI is not available: ask the user to share any signal results they have, or describe what they know about the contact and company. Proceed with whatever context is available.

Step 3 — Identify the best personalisation angle

From the signal results and context, pick the strongest hook — the one most likely to resonate with this specific contact. Prioritise:

1. Recent, specific events (funding, launch, hiring surge, leadership change) — shows you did your homework
2. Role-specific pain (their title implies a problem; signals confirm it) — shows relevance
3. Contact's own words (if contact-level signals show recent posts or activity) — most personal

Avoid generic openers like "I noticed you're growing fast" unless signals specifically confirm growth.

Step 4 — Confirm value prop and ask

Before writing, confirm:

• What outcome does the sender want from this message? (book a call, get a reply, share a resource)
• What's the core value prop in one sentence?

If this is already clear from conversation context (e.g. previous signal-discovery run), use it directly.

Step 5 — Write the outreach

Write all three variants using the chosen personalisation angle:

Cold email

• Subject line: specific, not clever — reference the signal hook
• Opening line: the personalisation hook (1 sentence, no fluff)
• Value prop: what you do and why it's relevant to them right now (2–3 sentences)
• CTA: one clear, low-friction ask (reply, 15-min call, a specific question)
• Length: under 120 words in the body

LinkedIn connection request (300 chars max)

• Lead with the personalisation hook
• One-line value prop
• No hard CTA — the connection itself is the ask

LinkedIn InMail

• Similar structure to the cold email but slightly warmer tone
• Can be slightly longer (150 words) given the LinkedIn context

Step 6 — Present variants and offer to adjust

Show all three variants clearly labelled. Ask:

• Which angle feels most accurate to the contact's situation?
• Does the value prop feel right?
• Offer to write 2–3 subject line alternatives for the email

Notes on quality

• Never fabricate signal data — only reference what was actually returned
• If a signal came back negative or ambiguous, don't use it as a hook
• The best outreach sounds like it was written by someone who actually read the signals, not generated by AI