LinkedIn-fed outbound system — scrape → qualify → sequence

You are helping me build a two-agent LinkedIn-fed outbound system on agntdata. Agent 1 scrapes LinkedIn engagers daily. Agent 2 enriches them, qualifies them, finds + verifies their email, and sequences cold email outreach (via Instantly) + LinkedIn outreach (via HeyReach).

REFERENCE DOCS
- agntdata API documentation: https://agnt.mintlify.app/apis/overview
- Endpoints behind one agntdata key: LinkedIn search/reactions/comments/profile/company, agntdata email finder + verifier, Exa search + getContents
- Integrations: HeyReach (campaign-add-leads-v2), Instantly (bulkAddLeads)

ABOUT MY PRODUCT
- Product name: <YOUR PRODUCT>
- One-line description: <WHAT IT DOES>
- ICP: <WHO BUYS IT — role + company stage>
- Topics on LinkedIn my buyers engage with (standard run pool): <8-12 KEYWORDS>
- Hashtags for the standard run pool: <3-6 HASHTAGS>
- High-intent keywords (used in the 2 high-intent runs of the day): <8-12 KEYWORDS>
- Geo gate: who counts as a target geography? (Default: US-only.)
- Closed-won pattern analysis: paste the deepest analysis you have of your last 5-10 won deals so the qualifier can pattern-match.

CAMPAIGNS
- One HeyReach campaign id: every qualified LinkedIn-outreach-ready lead is added here.
- One Instantly campaign id: every qualified deliverable-email lead is added here.

PERSONAS (optional)
- Names of 3-7 buyer personas if you want the sequencer to classify leads for opener context. Skip if you don't want persona tagging.
- (Advanced) If you want per-persona campaign routing instead of the single-campaign default — one HeyReach + one Instantly campaign id per persona — say so and provide the persona → campaign id tables. Don't do this unless you already run distinct campaigns per persona.

WHAT TO BUILD
- Two scheduled/queue-driven agents on agntdata. The scraper runs on cron (5 standard + 2 high-intent times per weekday). The sequencer drains a queue.
- Single `leads` table is the join point. Scraper writes; sequencer enriches.
- DNC table for suppression.

DATABASE (create fresh — agnt_db)
- `leads` — id (uuid PK), linkedin_profile_url (text, partial unique index where not null), first_name, last_name, title, company, company_domain, location, email, email_verification_status, score, score_rationale, persona, linkedin_source_post_url, created_at, updated_at.
- `linkedin_scraped_posts` — dedup/audit. PK post_url + per-post counts.
- `linkedin_search_history` — search_term, search_type, searched_at. Used to avoid repeating recent searches.
- `do_not_contact` — linkedin_profile_url, email, first_name, company, company_domain, reason, created_at.

DELIVERY
- Each scraper run posts a one-line summary (posts processed, new leads, leads queued).
- Each sequencer run outputs one line per lead with score + persona + sequencing routing.

GUARDRAILS
- Scraper caps at 5 posts per run.
- Sequencer skips email-finder credits for sub-50 / null-persona leads.
- DNC check before both HeyReach and Instantly.
- Cold-email send (Instantly) requires `email_verification_status === 'deliverable'`. LinkedIn send (HeyReach) doesn't.
- US-confirmation: positive geo signal required for both outbound legs.

When you're ready, start by asking me the ABOUT MY PRODUCT + CAMPAIGNS questions. Only bring up per-persona campaign routing if I mention it or if I already have multiple campaigns per channel.