Imported from Prism-Shadow/GDPevo (experiments/claude_code_opus_4_8_xhigh/reports/skills/task_group_001/self/self_attempt_02/SKILL.md). Install upstream with npx skills add Prism-Shadow/GDPevo --skill self_attempt_02. Copyright stays with the author.

HarborCRM Front-of-Funnel Handoff Skill

Solves four recurring task families in the HarborCRM domain:

1. Post-event CRM handoff / reconciliation (sponsor status + qualified non-sponsor leads).
2. Trade-show prospecting (qualify exhibitors by platform fit, rank, size opportunities).
3. Raw-contact import hygiene (normalize, dedupe, suppress, CRM-match).

Each test gives a prompt + an answer_template.json. The template is authoritative: emit exactly its keys, enums, ordering, and types — nothing more, nothing less. Per-task enum sets and schema shapes vary; always read the template first and copy its allowed values verbatim. Output JSON only, no prose.

0. The remote API (read-only, HTTP-only)

Base URL: supplied by runner (e.g. <remote-env-url>). Use curl or python urllib. Never WebFetch (forces HTTPS; host is HTTP-only).

Endpoints and what they are for:

• GET /api/policies — controlled value enums + notes. Sparse; most rules are inferred from data + template. Always fetch it but do not expect it to give thresholds.
• GET /api/events/{id} — start_date, end_date, status, campaign_code, lead_opportunity_amount, followup_days_after_end, sponsor_followup_days_after_end.
• GET /api/events/{id}/orders and /sponsor_packages — sponsor orders (account_id, account_name, amount, package_level, order_status ∈ {confirmed, proposal_sent, canceled}, ticket_contacts). Orders and sponsor_packages are usually the same rows.
• GET /api/events/{id}/badges — scanned attendees (badge_id, badge_type ∈ {sponsor, attendee, student, press, …}, company_name, contact_name, email, phone, job_title).
• GET /api/finance/invoices?event_id={id} — invoice_id, amount, paid_amount, deferred_amount, status ∈ {paid_deferred, open, …}, payment_date.
• GET /api/crm/accounts — account_id, name, domain, status ∈ {prospect, customer, disqualified}, disqualified_reason.
• GET /api/crm/contacts — contact_id, account_id, name, email, phone, opted_out.
• GET /api/crm/opportunities — account_id, amount, event_id, stage.
• GET /api/crm/campaign_members?event_id={id} — existing members (account_id, contact_id, status ∈ {attended_sponsor, registered_sponsor, attended}).
• GET /api/tradeshows, /api/tradeshows/{id}/exhibitors, /api/tradeshows/{id}/meeting_interest.
• GET /api/import_batches, /api/import_batches/{id}/raw_contacts, /api/import_batches/{id}/suppression.

Fetch only what the task family needs (see each SOP). Match accounts to companies by email/website domain, not by display name (names vary: "HelioWare Manufacturing" vs "HelioWare Mfg.").

1. Normalization rules (used everywhere)

• Email: trim surrounding whitespace, lowercase. If the result is empty/blank → empty string "". Domain = substring after @ (for account matching).
• Phone: strip everything except digits ([^0-9] removed). Keep leading country-code digits if present in the source (e.g. +1 415 555 0188 → 14155550188; (415) 555-0188 → 4155550188). Do not invent or strip a country code that wasn't there. Empty/blank → "".
• A contact is contactable iff it has a non-empty normalized email or non-empty normalized phone. Email-only and phone-only are both fine.
• Output amounts as integer USD (no decimals, no currency symbol).

2. Sponsor status classification (event reconciliation)

For every account that has a sponsor order/package for the event, reconcile order + invoice:

• open_invoice open balance = invoice.amount − invoice.paid_amount (use amount, not deferred_amount).
• Canceled orders: how to report depends on the template's sponsor_status enum:
  • If the enum includes not_sponsor → list the canceled/non-sponsor account with sponsor_status = not_sponsor (amount = its package/order amount, or 0 if none).
  • If the enum omits not_sponsor (only paid_deferred/open_invoice/proposal_only) → exclude the canceled account from sponsor_statuses and instead surface it in the exclusion list with reason inactive_sponsor_record.
• Revenue-by-status totals: sum package/order amount per status bucket. open_invoice total = full package amount of open-invoice sponsors; open_invoice_balance = sum of their open balances (amount − paid_amount). proposal_only total = sum of proposal amounts.
• Sort sponsor_statuses by account_name ascending.

Unpaid / sponsor-finance follow-up set = sponsors that are NOT paid_deferred, i.e. open_invoice + proposal_only. Their total = sum of their package amounts. (If a template clearly scopes it to "open invoices only," restrict to open_invoice. Default: include both.) paid_deferred sponsors are settled and need no finance follow-up.

3. Badge / lead classification (event reconciliation)

For each badge, classify into sponsor_attendee / qualified_non_sponsor_lead / excluded:

1. sponsor_attendee — the badge's company has ANY sponsor order for the event (confirmed OR proposal_sent — proposal-stage accounts are on the sponsor track, not the lead track) or the badge_type is sponsor. Exclusion reason sponsor_attendee. Not a sales lead.
2. excluded — non_business_badge — badge_type is non-business (student, press, guest, media, etc.). Reason non_business_badge.
3. excluded — existing_disqualified — the badge's company matches a CRM account whose status == disqualified (has a disqualified_reason). Reason existing_disqualified.
4. excluded — missing_contact — badge has neither email nor phone after normalization (not contactable). Reason missing_contact.
5. Otherwise → qualified_non_sponsor_lead (business attendee, contactable, account not disqualified, company is not a sponsor).

Precedence when several apply: sponsor_attendee → non_business_badge → existing_disqualified → missing_contact. (A disqualified-account attendee that is also a canceled sponsor's contact is genuinely ambiguous; prefer the account-level existing_disqualified if the CRM account is flagged disqualified, else inactive_sponsor_record.)

CRM action for each badge (use the template's crm_action enum — names differ per task):

• Qualified lead, company not in CRM → create account + contact + campaign member (create_account_contact_campaign_member, or in the leaner schema: crm_account_action=create_account, crm_contact_action=create_contact, campaign_member_action=add_campaign_member).
• Qualified lead, company in CRM but contact missing → create_contact_campaign_member (account update_existing, contact create_contact).
• Qualified lead already a campaign member → update_campaign_member / add_campaign_member per template; sponsor attendee already a correct member → no_action.
• Excluded badge → no_import (or no_action for sponsor attendees already handled).

badge_decisions sort by badge_id ascending.

4. Campaign-member actions & target status (event reconciliation)

Compare each subject (sponsor ticket contacts + qualified leads) against existing campaign_members:

• Already a member with the correct status → no_action.
• New subject who should be a member → create.
• Member whose status should change → update.
• Excluded subject → no_import.

target_status mapping:

• Sponsor account, contact attended (has a badge scan) → attended_sponsor.
• Sponsor account, registered only (no badge scan) → registered_sponsor.
• Qualified non-sponsor lead who attended → attended.
• Excluded → excluded.

Sort by subject_key ascending (or as the template specifies).

crm_action_counts (when present) tally the implied work across qualified leads (and any member updates): accounts_create/update, contacts_create/update, campaign_members_create/update. New leads each contribute one create per dimension as applicable; an existing account contributes accounts_update.

5. Opportunity & pipeline math (event reconciliation)

• Each qualified non-sponsor lead's opportunity = the event's lead_opportunity_amount.
• lead_pipeline_total / open_opportunity_total_usd = (count of qualified non-sponsor leads) × lead_opportunity_amount. open_opportunity_count = that count.
• qualified_non_sponsor_account_names = distinct company names of qualified leads, sorted ascending.

6. Follow-up due dates (event reconciliation)

• Lead follow-up due date = event end_date + followup_days_after_end (calendar days).
• Sponsor finance follow-up due date = event end_date + sponsor_followup_days_after_end.
• Format YYYY-MM-DD. Use the end_date (not start_date) as the base; single-day events have start_date==end_date.
• lead_task_count = number of qualified-lead handoff tasks (= qualified lead count). sponsor_finance_task_count = number of unpaid sponsor accounts (Section 2).

7. Trade-show prospecting SOP

Goal: from exhibitors, pick those that manufacture / OEM-build target underwater platforms, rank them, size opportunities, and list near-miss exclusions.

Endpoints: /api/tradeshows/{show}/exhibitors, /.../meeting_interest, /api/crm/accounts, /api/policies.

Platform enums (fixed): AUV, ROV, Underwater Camera. Always list platforms in that enum order.

Read the exhibitor description to assign platforms:

• "AUV", "autonomous underwater vehicle", "AUV scout" → AUV.
• "ROV", "remotely operated", "inspection-class ROV", "pen-cleaning ROV" → ROV.
• "underwater camera", "camera modules/arrays", "optics manufacturer", "camera maker" → Underwater Camera.
• A maker can have multiple platforms (e.g. "builds AUVs and ROVs" → [AUV, ROV]).

Qualified vs excluded (the key judgment — exclude near-misses):

Qualified = the company builds/OEM-manufactures at least one target platform. Exclude companies that only sit adjacent to the platform, with a controlled reason. Use the exact exclusion enum from the template (it differs per task):

• Distributor / reseller / dealer / sales agent ("does not manufacture", "imported brands") → distributor_only (relationship distributor).
• Service / consulting / operator ("operates rented ROVs", "analytics dashboard using partner ROV feeds", "no hardware manufacturing") → service_only (relationship service_provider).
• Sensor-only vendor ("sensor-only DO/salinity probes for integration by platform partners") → sensor_vendor_only or sensor_only (copy the template's spelling) (relationship sensor_vendor).
• Research / academic lab → research_only (relationship research).
• Wrong market entirely → not_target_market (only if the template offers it).
• All excluded exhibitors get crm_action = no_import and stay visible in the exclusion list.

Critical: a company that makes the sensor but not the platform is NOT qualified (it's the sensor vendor we are selling FOR, not the OEM target). A company that merely uses or resells platforms is NOT qualified.

CRM overlap & action:

Match exhibitor to CRM account by crm_account_id (if exhibitor row provides it) or by website/email domain.

• Qualified + already in CRM → crm_action = update_existing, set crm_account_id.
• Qualified + not in CRM → crm_action = create_account, crm_account_id = null.
• existing_crm_overlap_count / existing_crm_overlap_account_ids count only qualified leads that have a CRM account; ids sorted ascending.

Priority tier & opportunity sizing:

Join exhibitor to meeting_interest by company_name to get interest_score and requested_demo. Default tiering (used when the prompt doesn't override): based on demo request + score:

• A = requested_demo == true AND interest_score >= 90.
• B = requested_demo == true AND interest_score >= 80 (and not A).
• C = everything else qualified. If the prompt gives explicit tier thresholds or dollar values, use those verbatim. Typical opportunity sizing: A = 120000, B = 90000, C = 50000 (USD) — but always take the prompt's stated values if given. total_estimated_opportunity_usd = sum of qualified leads' opportunity estimates.

Ranking (when the template has rank):

Order qualified leads by: (1) requested_demo true first, (2) interest_score descending, (3) broader platform coverage (more platforms first), (4) company_name ascending. Assign 1-based contiguous rank.

Counts:

• platform_coverage_counts / platform_counts: count of qualified leads covering each platform (a multi-platform lead increments each of its platforms).
• priority_counts: count of qualified leads per tier A/B/C.
• qualified_total / qualified_lead_count, excluded_*_total/excluded_count.

Sort qualified_exhibitors/ranked_leads per template (company_name asc, or rank asc); excluded_* by company_name ascending.

8. Import-batch hygiene SOP

Endpoints: /api/import_batches/{batch}/raw_contacts, /.../suppression, /api/crm/accounts, /api/crm/contacts, /api/policies. Campaign code comes from the import_batch record (campaign_code).

Process raw rows in this order:

1. Normalize every row (Section 1): trimmed-lowercased email, digits-only phone.
2. Drop unusable rows = not contactable (no email AND no phone) → removal reason missing_contact, action no_import.
3. Dedupe remaining rows. Key = normalized email (fallback to normalized phone when email is empty). Within a duplicate group pick the winner:
   • Primary: latest captured_at (freshest record wins).
   • Tie-break: lowest row_id (ascending) when timestamps are equal. Losers → removal reason duplicate. Record {key, winner_row_id, removed_row_ids} per group; sort duplicate_keys by key ascending.
4. Suppression: a winner whose normalized email or phone matches any suppression-list entry → removal reason suppressed, action suppress. (Suppression matches on email or phone; reasons like global_opt_out / privacy_request / role_account all suppress.)
5. CRM match the survivors by email domain → account.domain:
   • Domain matches an existing account → crm_action = update_existing; set existing_account_id; existing_contact_id = that account's matching contact_id if the exact person already exists, else null.
   • No account match → crm_action = create_account; existing_account_id = null, existing_contact_id = null.

Outputs:

• clean_contacts = the surviving rows only (the create_account + update_existing set). Each item uses the winner row's values: clean_contact_id = source_row_id = winning row_id; company_name, contact_name, normalized email, normalized phone, source_name, captured_at (winner's), crm_action, existing_account_id, existing_contact_id. Sort by clean_contact_id ascending.
• duplicate_summary: duplicate_removed_count = total duplicate losers; duplicate_keys as above.
• removal_summary: unusable_removed_count = missing_contact count; suppressed_removed_count; removed_rows = every removed row {row_id, reason} with reason ∈ {duplicate, missing_contact, suppressed}, sorted by row_id ascending.
• import_action_totals: integer counts over the deduped rows (winners + suppressed + unusable, i.e. excluding duplicate losers) for create_account, update_existing, no_import (missing_contact rows), suppress.
• campaign_member_import_count = number of surviving cleaned contacts (the clean_contacts list length).

9. Common misjudgments to avoid

• Don't add or rename keys/enums. Copy the template's exact enum spellings (e.g. sensor_vendor_only vs sensor_only; presence/absence of not_sponsor).
• Canceled sponsors are not active sponsors. Either not_sponsor or excluded (inactive_sponsor_record) depending on the enum — never proposal_only.
• Open-invoice total uses full package amount; the balance is a separate field (amount − paid_amount). Don't put the balance in the revenue total.
• Proposal-stage accounts are on the sponsor track, so their attendees are sponsor_attendee, not qualified leads.
• Match companies by domain, not display name (manufacturing/Mfg., punctuation, suffixes vary).
• Sensor-only and platform-using companies are NOT qualified prospects — they are near-miss exclusions.
• Use end_date (not start_date) for all follow-up date arithmetic; add the event's own day-offset fields.
• Lead opportunity amount comes from the event (lead_opportunity_amount), not from CRM opportunity rows; CRM opportunities are usually for sponsor/closed deals, not the new leads.
• Dedup winner = latest captured_at, tie-break lowest row_id. A duplicate with an identical timestamp is the signal to use the row_id tie-breaker.
• Suppression and missing-contact rows are removed, not in clean_contacts, but still counted in import_action_totals and listed in removal_summary.
• Contactable = email OR phone. An empty email alone does not disqualify a row that has a phone.
• Respect every ordering rule; ties on the primary sort key fall back to the secondary key named in the template.
• Output JSON only, integers for all amounts/counts, dates as YYYY-MM-DD.