Fewshot Attempt 01

HarborCRM Skill

Environment

Base URL: http://34.46.77.124:8001. All API calls use this prefix. Never start local env/setup.sh or use localhost — environment_access.md overrides any task text that mentions local URLs.

API Reference

Events (post-event reconciliation tasks)

Tradeshows (prospecting tasks)

Finance

CRM

Import Batches

Policies

GET /api/policies — returns contact_hygiene, prospecting.platform_enums (always ["AUV","ROV","Underwater Camera"]), sponsor_handoff.status_enums.

Normalization Rules

• Email: lowercase, trim leading/trailing whitespace. If blank/whitespace-only after trim → treat as empty string.
• Phone: strip all non-digit characters (+, -, ., (, ), spaces). If result is empty → empty string.

Sponsor Status Derivation

For event reconciliation tasks, classify each sponsor order:

1. Order is canceled (order_status == "canceled") → exclude entirely (inactive sponsor record). Do NOT include in sponsor_statuses. Include in excluded_records with reason inactive_sponsor_record.

2. Has invoice with status == "paid_deferred" → paid_deferred. paid_amount = invoice.paid_amount, open_balance = 0 (deferred revenue recognized, fully paid).

   • Revenue total: sum package_amount (the order/invoice amount).

3. Has invoice with status == "open" → open_invoice. paid_amount = invoice.paid_amount, open_balance = invoice.amount - invoice.paid_amount.

   • Revenue total: sum invoice amount. open_invoice_balance = sum of all open_balance values.

4. No invoice exists but order is active (order_status == "confirmed" or "proposal_sent") → proposal_only. paid_amount = 0, open_balance = 0, invoice_id = null.

   • Revenue total: sum order amount.

5. Not a sponsor (not_sponsor) — use only when a task template explicitly requires it.

Sponsor Revenue Totals

{
  "paid_deferred": <sum of amounts for paid_deferred sponsors>,
  "open_invoice": <sum of amounts for open_invoice sponsors>,
  "proposal_only": <sum of amounts for proposal_only sponsors>,
  "open_invoice_balance": <sum of (amount - paid_amount) for open_invoice sponsors>
}

All values are integers (USD).

Sponsor Finance Follow-up

• Accounts needing follow-up: sponsors with status open_invoice or proposal_only (not paid_deferred).
• Due date: event.end_date + event.sponsor_followup_days_after_end days (format YYYY-MM-DD).
• Task count: number of accounts needing follow-up.

Lead Qualification (Event Tasks)

From badge scans, identify qualified non-sponsor leads:

1. Exclude sponsor attendees: any badge whose company_name matches a sponsor order's account_name (case-sensitive exact match). Include in excluded_records with reason sponsor_attendee.
2. Exclude non-business badges: badge_type is "student", "press", or similar non-business types → reason non_business_badge.
3. Exclude disqualified CRM accounts: look up badge company_name in CRM accounts. If the account has status == "disqualified" or non-null disqualified_reason → reason existing_disqualified.
4. Remaining badges → qualified non-sponsor leads.

CRM Actions for Qualified Leads

• Account: if the company has a matching CRM account (matched by name or by email domain — extract domain from the badge contact's normalized email, compare to CRM account domain field) → update_existing. Otherwise → create_account.
• Contact: if a CRM contact exists with the same normalized email → update_existing. Otherwise → create_contact. (Also check same account_id.)
• Campaign member: all qualified leads get add_campaign_member.

Opportunity Amount

Use event.lead_opportunity_amount for each qualified lead. lead_pipeline_total = sum of all lead opportunity amounts.

Lead Follow-up

• Due date: event.end_date + event.followup_days_after_end days (format YYYY-MM-DD).
• Task count: number of qualified leads.

Trade Show Prospecting

Qualification

Read each exhibitor's description field. An exhibitor is qualified if it manufactures/OEM-builds at least one of the target platforms: AUV, ROV, or Underwater Camera. Use the description text to determine platform coverage:

• "builds AUVs", "manufactures ROVs", "OEM underwater cameras" → qualified
• "distributor", "reseller", "does not manufacture" → not qualified
• "sensor-only", "analytics dashboard using partner hardware", "consulting/operates rented" → not qualified

Exclusion Reasons (controlled vocabulary)

Use the exclusion_reason enum from the task's answer template. Common values:

• distributor_only — reseller/distributor, doesn't manufacture platforms
• service_only — consulting/services, operates but doesn't build
• sensor_vendor_only / sensor_only — sensor-only vendor, no platform manufacturing
• research_only — research institution
• not_target_market — doesn't fit campaign criteria

Match relationship_type to exclusion_reason when both are required: distributor→distributor_only, service_provider→service_only, sensor_vendor→sensor_only, research→research_only.

Platform Enums

Always ["AUV", "ROV", "Underwater Camera"] in that order. Sort platforms lists in this enum order, not alphabetically.

Priority Tiers & Opportunity Sizing

Derived from meeting interest data (joined by company_name):

Ranking (when task requires ranked output)

1. Demo requested first (true before false)
2. Interest score descending
3. Platform coverage count descending (more platforms ranks higher)
4. Company name ascending (alphabetical tiebreaker)

CRM Actions (Trade Show)

• Exhibitor has crm_account_id not null → update_existing
• Exhibitor has crm_account_id null → create_account
• Excluded exhibitors → no_import

CRM Account Matching

• Match exhibitor crm_account_id to CRM account account_id directly when the field is present.
• For event tasks without explicit exhibitor CRM links: extract domain from normalized email → match against CRM domain field. Fall back to company_name match if domain lookup fails.

Aggregate/Summary Counts

• qualified_total: number of qualified exhibitors
• platform_counts: count of qualified exhibitors covering each platform (a company may cover multiple platforms)
• priority_counts: count of qualified exhibitors per tier
• existing_crm_overlap_count: number of qualified exhibitors with non-null crm_account_id
• total_estimated_opportunity_usd: sum of opportunity_estimate_usd across all qualified leads

Batch Import Cleaning

Deduplication

Group raw contacts by normalized email (lowercase, trimmed). When multiple rows share the same email:

Winner selection (source priority, highest first):

1. partner_upload
2. webinar_form
3. badge_scan
4. sponsor_form
5. exhibitor_form
6. manual_upload

If same source, pick the row with the later captured_at timestamp. The winner becomes the clean_contact. Its row_id is used as both clean_contact_id and source_row_id.

Suppression

Check each surviving clean contact (after dedup) against:

1. Suppression list (GET .../suppression): match by normalized email OR digits-only phone.
2. CRM contacts with opted_out == true: match by normalized email.

A contact is suppressed if it matches EITHER source. Suppressed contacts get crm_action: "suppress", removal reason "suppressed".

Missing Contact

A row is unusable if contact_name is blank/null OR normalized email is empty/whitespace-only. Removal reason: "missing_contact".

CRM Account/Contact Matching (Batch Import)

• Match company to CRM account: extract domain from normalized email → find CRM account with matching domain. If email is empty, try company_name match.
• Match contact to CRM contact: normalized email match against CRM contacts. Only set existing_contact_id if the matched contact belongs to the matched account.
• If account found → update_existing. If not → create_account.

Duplicate Key Format

"email:<normalized_email>" — one entry per duplicate group.

Removal Summary

• unusable_removed_count: rows removed for missing_contact
• suppressed_removed_count: rows removed for suppressed
• duplicate_removed_count: rows removed as losers in dedup (in duplicate_summary)
• removed_rows: ALL removed rows (duplicates + suppressed + missing_contact), sorted by row_id ascending

Import Action Totals

Count actions across ALL raw contact rows (not just clean ones):

• create_account: clean contacts with no matching CRM account
• update_existing: clean contacts with matching CRM account
• no_import: removed duplicates + missing_contact rows
• suppress: suppressed rows

Campaign Member Count

Number of clean contacts that survived dedup, suppression, and missing-contact removal — i.e., clean_contacts.length.

Badge-Level Handling (Event Reconciliation with Badges)

When a task requires per-badge decisions:

Classification

• sponsor_attendee: badge company_name matches a sponsor order's account_name
• qualified_non_sponsor_lead: not a sponsor attendee, not disqualified, business badge type
• excluded: non-business badge OR existing disqualified CRM account

CRM Action per Badge

• New account + new contact needed → create_account_contact_campaign_member
• Existing account, new contact → create_contact_campaign_member
• Existing account and contact, just add to campaign → add_campaign_member
• Sponsor attendee with existing campaign member → update_campaign_member or no_action
• Excluded badges → no_import
• Sponsor attendee (contact needed but lead-excluded) → create_contact_campaign_member (adds the contact record but they remain classified as sponsor_attendee for lead purposes)

Campaign Member Actions

For each unique (account, contact) pair relevant to the event:

Subject key format: "acct_{id}:cont_{id}" for CRM contacts, "badge:{badge_id}" for badge-only leads.

Badge-Only Contacts

Contacts from badges that have no existing CRM contact record. Include normalized email (lowercase, trimmed; empty string if none) and normalized phone (digits only; empty string if none). Include sponsor attendee badge-only contacts too. Sort by company_name ascending.

Exclusion Counts

Count excluded badges by reason: sponsor_attendee, non_business_badge, existing_disqualified, missing_contact. Integer values.

Sorting Rules (by task type)

General Rules

1. Return JSON only — no explanatory prose outside the JSON object.
2. Match the answer template exactly — do not add or omit top-level keys. Use the field names and enum values from the template.
3. Integer amounts — all monetary values are integers (USD). No decimals or floats.
4. Date format — always YYYY-MM-DD. Calculate follow-up dates as end_date + offset_days.
5. Empty strings, not null — for missing emails/phones in normalized output, use "" not null, unless the template explicitly requires null.
6. account_id / invoice_id nulls — use JSON null for absent IDs per the template (e.g., no invoice → invoice_id: null; no CRM account → account_id: null or crm_account_id: null).
7. Campaign member actions span both CRM contacts and badge-only contacts — include all relevant (account, contact) pairs for the event, not just qualified leads.
8. Canceled sponsor orders are inactive — exclude from sponsor_statuses; include in excluded_records with reason inactive_sponsor_record.
9. Don't double-count — when a sponsor contact appears in both campaign_members and badges, don't create duplicate campaign member actions.
10. Domain extraction — from user@example.com extract example.com. Use this to match against CRM account domain field.