Draft outreach from a signal

Turn one signal into a short, personalized first message that references what actually happened.

A signal already tells you who moved and why. This playbook turns that into a first message a rep can send, built around the real event instead of a generic template.

What you need

At least one signal in your workspace, so a keyword or watchlist agent that has run. The skills or the MCP get you there in minutes.

The result

Hi Jane, saw your post on buying intent at Acme. We just helped two RevOps teams
cut their qualification time in half on exactly that problem. Worth a quick look?

Two ways to run it

With the MCP connected or the sillage-api skill installed, point your assistant at a signal and let it write.

Take my most recent signal. Look up who it's about and what they posted, then
write a short first LinkedIn message that references it specifically. Under four
sentences. Sound like a person, not a template, and skip "I hope this finds you
well".

Ask for three variations, or a colder or warmer tone, until it fits your voice.

Read a signal, pull the person and topic off it, and hand both to your model.

Get a signal

Grab the newest detection. Filter by agent_id or company_id to target a specific account instead.

const key = { Authorization: `Bearer ${process.env.SILLAGE_API_KEY}` }
const base = 'https://api.getsillage.com/api/v2'

const { data: [signal] } = await fetch(`${base}/workspace/signals?limit=1`, {
  headers: key,
}).then((r) => r.json())

Draft with your model

Keyword detections carry the author and the keywords they matched on. Feed those to whatever model you already use.

const person = signal.data?.author // { full_name, headline, linkedin_url }
const topic = signal.data?.keywords_found?.join(', ')

// your own LLM call (Claude here)
const message = await claude(
  `${person.full_name} (${person.headline}) posted about ${topic}.
   Write a short, specific LinkedIn opener that references it. Under 4 sentences, no fluff.`,
)
console.log(message)

Want the model to read the actual post, not just the keywords? Each keyword detection carries a data.content_id. Fetch it from the Contents API and pass the text into the prompt.

Make it yours

Loop over a whole morning's signals to draft a batch, one message per account.
Pull the post text with data.content_id so the model quotes a real line.
Add your value proposition to the prompt so every draft ties the event back to what you sell.
Route the drafts into a review step instead of sending straight away.

Next steps

Who to contact today decides which signal to act on first.
Signal reference explains what the signal means before you draft around it.
Contents API returns the post behind a detection.