personalized-outreach

<skill name="personalized-outreach"> <overview> Personalized outbound doesn't scale by hand, so a periodic sweep does the research and the writing per contact instead. It reads HubSpot for contacts in {{hubspot_list}} that haven't been handled yet, enriches each one, drafts a first-touch and follow-up sequence grounded in the account's real context using the angles and proof points in memory, logs every draft back to HubSpot, and holds the batch in {{approval_channel}} for a person to approve and send — capped at {{daily_cap}} contacts per run.

Fresh session per sweep — state lives on the HubSpot contact record itself (a kortix_outreach_drafted property + timestamp marks a contact handled), not in a local ledger file. A single sweep can find several new contacts at once; handle each one as an independent unit — a research or drafting failure on one contact is logged and skipped, and never blocks or corrupts the draft for any other contact in the same sweep. The outreach approach itself (angles, proof points, what to avoid, which signals are worth writing to) lives as memory that travels with the agent and improves as message patterns prove out. </overview>

<when-to-load> - The periodic outreach sweep fires. - A human asks the agent to run outreach against a specific list or contact. - The outreach playbook (angles, proof points, what to avoid) changes and needs to be reflected in the next batch of drafts. </when-to-load> <workflow>

Step 0 — Orient: find what's new

{{hubspot_list}} is a named HubSpot list, not a contact property — resolve it to a list ID first, then page through its memberships:

GET /crm/v3/lists/object-type-id/0-1/name/{{hubspot_list}}

Take the list.listId from the response, then page through membership records (100 at a time, following after until the response has no more pages):

GET /crm/v3/lists/{listId}/memberships?limit=100&after={after}

Collect the recordIds this returns, then batch-read those contacts and drop anything already carrying kortix_outreach_drafted:

POST /crm/v3/objects/contacts/batch/read
  properties: [... the properties Steps 1-4 need ...]
  inputs: [{ id: recordId }, ...]   // the memberships page(s) above

Anything already carrying kortix_outreach_drafted was handled by a prior run — skip it, even if it hasn't been sent yet. Sending is a human decision, not a signal to re-draft. Take at most {{daily_cap}} contacts this run, oldest or highest-priority first, so the batch stays a size a person can actually review.

Step 1 — Enrich each contact

Pull the account's real context: role and seniority, company size and industry, and recent signals (funding, hiring, product launches, news) that make a message specific instead of generic.

Step 2 — Pick the angle

From memory: the angles that resonate, the proof points to use, what to avoid, and which signals in an account are worth writing to. Choose the angle that actually matches this account's context — don't default to the same opener for every contact.

Step 3 — Draft the sequence

Write a first-touch email and its follow-ups grounded in what Steps 1–2 found: one specific reference to the account's real context, the proof point that fits, and a single clear ask. This is not a template with a name and company swapped in — if the draft would read the same with the enrichment data removed, it isn't personalized enough yet.

Step 4 — Log the draft to HubSpot

Write the drafted sequence, the research notes, and the angle used onto the contact's record/timeline so the history is complete even before the email sends or if a human edits the draft first.

Step 5 — Hold for approval, mark handled

Place the batch of drafted sequences in {{approval_channel}} for a person to review, edit, approve, and send. Do not send anything yourself. Set kortix_outreach_drafted = true (with a timestamp) on each contact you drafted this run so the next sweep skips them.

</workflow> <guardrails> - **Never send.** Every sequence is a draft held in {{approval_channel}} for a human to review, edit, and send. - **Daily cap.** Draft at most {{daily_cap}} new contacts per run, regardless of how large {{hubspot_list}} is — volume never outruns what a person can review. - **HubSpot writes are scoped** to research notes, drafted sequence content, and the `kortix_outreach_drafted` marker. Never change deal stage, pipeline, or lifecycle stage. - **No duplicate drafts.** Always check the `kortix_outreach_drafted` marker before researching or writing — a contact drafted once is never re-drafted, even across many sweeps. - **Scoped secrets.** HubSpot and enrichment credentials are injected by the connector at runtime, scoped to this agent's grant. </guardrails> </skill>