GTM Email Sequencer — personalized cold email + Instantly push

You are helping me build a GTM Email Sequencer agent that runs on a 5-minute cron. It picks up one verified lead at a time from my workspace database, writes a personalized cold email using one of several contextual variants (chosen by signals on the lead — e.g. their role and the size of their company), stores the draft alongside the lead, and pushes the lead into my Instantly campaign for sending.

This blueprint is general — it works for any outbound motion where you have a stream of verified leads and want a personalized email per lead. Hiring outreach, sales outbound, partner outreach, fundraising — same shape.

REFERENCE DOCS (read these before writing code)
- Full agntdata API documentation: https://agnt.mintlify.app/apis/overview
- agntdata Instantly integration: bulkAddLeads — the only outbound this agent does
- Workspace DB access via agnt.db — used for the lead claim, draft write, and status updates

ABOUT MY PRODUCT
- Product name: <YOUR PRODUCT>
- One-line description: <WHAT IT DOES>
- ICP: <WHO YOU SELL TO — role + company stage>
- Sender voice: <FOUNDER NAME, role, signing-off line for emails>
- Instantly campaign id (UUID): <CAMPAIGN_ID>

YOUR DATA MODEL
- Leads table name: <LEADS_TABLE> (default: `leads`)
- "Ready to sequence" filter: `<READY_STATUS_COLUMN> = '<READY_VALUE>'` (e.g. `enrichment_status='verified'`, `qualification_status='qualified'`, or just `status='ready'`)
- Sequencing state column: `<SEQUENCE_STATUS_COLUMN>` (typical default: `sequencing_status`, with values `pending → in_progress → sequenced | sequencing_failed`)
- Variant selection signals — the 1–3 columns the agent reads to pick a variant. Common combos:
  - role + company size (default)
  - role + industry
  - title + tenure
  - trigger event type (job change / funding / signup) + company size
- Personalization tokens used in your variants: at minimum `[Name]`, `[company]`, and `[context]` — where `[context]` is whatever lead-specific value your hooks reference (job title, recent trigger, etc.). Tell me which column `[context]` maps to on your lead rows.

EMAIL VARIANTS
Give me 3–5 contextual variants of your cold email. For each, specify:
- WHEN TO USE — the trigger (e.g. mid-market + sales role, early-stage startup + technical founder)
- SUBJECT LINE — with placeholder tokens
- BODY — 3–5 short paragraphs, ending with a soft question close
- FOLLOW-UP — one universal follow-up the agent sends as a reply to every email

WHAT TO BUILD
- A scheduled agent on agntdata that runs every 5 minutes (claude-sonnet-4-6 — the variant-selection logic benefits from the smarter model).
- Each run: claim ONE row from <LEADS_TABLE> where `<READY_STATUS_COLUMN>='<READY_VALUE>'` AND `<SEQUENCE_STATUS_COLUMN>='pending'`. Order by `created_at ASC`. Immediately mark it `in_progress` so concurrent runs don't double-process.
- Pick the right variant using your variant-selection signals. Substitute placeholders from the lead row. Generate clean output: no double line breaks, no stray punctuation.
- Write the finalized `email_subject`, `email_body`, and `email_variant` back onto the lead row.
- Call Instantly `bulkAddLeads` with the campaign id; pass the subject, body, and follow-up as `custom_variables` (`custom_subject`, `custom_body`, `custom_followup`).
- Mark `<SEQUENCE_STATUS_COLUMN>='sequenced'` and `sequenced_at=now()` on success. On failure, mark `sequencing_failed` and surface the error in the run summary.

DATABASE (must already exist — upstream agents write to it)
- Your leads table is owned by upstream scraper / enricher / qualifier agents. This sequencer adds five columns to it on first run: `<SEQUENCE_STATUS_COLUMN>`, `sequenced_at`, `email_variant`, `email_subject`, `email_body`. The rest of the schema is whatever your pipeline already produces.
- If you don't have a leads table yet, look at the `linkedin-gtm-engager-scraper` blueprint (creates one) and `linkedin-lead-qualifier-enricher` blueprint (enriches + verifies). This sequencer is the closing piece of that pipeline.
- (Optional) `agent_config` — single-row config table for environment-specific values like the Instantly campaign id, when you don't want them in the system prompt.

DELIVERY
- After each run, the agent prints a one-line JSON summary: lead id, name, company, variant used, subject, instantly_status. That's how you'll spot failures from the agent_sessions list at a glance.

GUARDRAILS
- Process exactly ONE lead per run — keeps runs deterministic and easy to monitor.
- Never re-process a lead where `<SEQUENCE_STATUS_COLUMN> != 'pending'`. The `in_progress` lock plus the status filter together prevent double-sends.
- Multiple contacts at the same company is correct — sequence every verified lead regardless of whether a colleague already got their email.
- Never send a raw template — every email must have all placeholders filled.
- The agent does NOT scrape, enrich, or verify emails — that's upstream agents' jobs. This agent assumes "ready" means trustworthy.

When you're ready, start by asking me the ABOUT MY PRODUCT, YOUR DATA MODEL, and EMAIL VARIANTS questions.