Fewshot Attempt 02

HarborCRM Trade-Show Pipeline Skill

Reusable workflow guidance for processing marine-robotics trade-show exhibitor data through HarborCRM.

1. Task Taxonomy

The pipeline generally follows one of four patterns. Read the prompt carefully to identify which schema to produce:

2. API Query Habits

• Always start with GET /api/v1/shows/{show_id}/exhibitors to get the raw list.
• Iterate per exhibitor with GET /api/v1/exhibitors/{exhibitor_id} for booth, country, website, product lines, etc.
• Fetch relationships with GET /api/v1/relationships?entity=exhibitor&entity_id={exhibitor_id} to determine relationship_type (manufacturer, distributor, service_provider, sensor_vendor, research).
• Enrichment via GET /api/v1/enrichment/vendors/{vendor_id} when a vendor_id is present to discover platforms and sensor capabilities.
• Campaign eligibility via GET /api/v1/campaigns/{campaign_id}/eligibility to confirm qualification rules for the specific campaign.
• CRM deduplication via GET /api/v1/crm/accounts (search by website domain) when producing import or prospecting outputs.

Only query endpoints and entity IDs explicitly named in the prompt. Do not call global list/index endpoints to discover extra shows, campaigns, or batches.

3. Business Rules for Qualification

3.1 Priority Tiers

3.2 Platform Normalization

• Normalize whatever strings appear in product_lines / platform_tags to the canonical set: AUV, ROV, Underwater Camera.
• Platform arrays in JSON outputs must be ordered: ["AUV", "ROV", "Underwater Camera"] (omit any the exhibitor does not have).
• In CSV outputs, platforms are pipe-separated in the same order: AUV|ROV, ROV|Underwater Camera, etc.

3.3 Exclusion Reasons

Use these exact strings:

• distributor_only
• service_only
• sensor_vendor_only (qualification tasks) or sensor_only (prospecting tasks — follow the answer template enum)
• research_only

4. Output Conventions & Schemas

4.1 Qualification Output (Pattern A)

{
  "show_id": "<exact_show_id>",
  "campaign": "<campaign_id_from_prompt>",
  "qualified_exhibitors": [ ... ],
  "excluded_near_misses": [ ... ],
  "aggregate_counts": { ... }
}

• qualified_exhibitors objects contain: company_id, company_name, platforms, priority_tier, booth, country, website.
• excluded_near_misses objects contain: company_id, company_name, exclusion_reason.
• aggregate_counts must include all expected keys even when zero:
  • qualified_total
  • platform_counts: { "AUV": N, "ROV": N, "Underwater Camera": N }
  • priority_counts: { "A": N, "B": N, "C": N }
  • excluded_near_misses_total

4.2 CRM Import Output (Pattern B)

{
  "show_id": "<exact_show_id>",
  "crm_import_report": [ ... ]
}

• Each entry: company_id, company_name, crm_account_id (string or null), crm_action, status, match_basis (string or null).
• match_basis for website matches should contain the matched domain.
• Report covers all exhibitors (qualified + excluded), sorted by company_id ascending.
• Excluded exhibitors receive "crm_action": "no_import".

4.3 CSV Report (Pattern C)

• Columns (exact order): Company Name, Country, Priority Tier, Booth, Platforms, Interest Score
• Header row is required.
• show_name in the filename/prompt context should be the human-readable show name (not the show_id).
• Only qualified exhibitors appear.

4.4 Lead Ranking / Prospecting Output (Pattern D)

{
  "show_id": "<exact_show_id>",
  "summary": { ... },
  "ranked_leads": [ ... ],
  "excluded_exhibitors": [ ... ]
}

• summary contains:
  • qualified_lead_count, excluded_count
  • existing_crm_overlap_count, existing_crm_overlap_account_ids (sorted ascending)
  • total_estimated_opportunity_usd
  • platform_coverage_counts: count every platform occurrence across all qualified leads (e.g., a company with ROV + Underwater Camera contributes 1 to each).
• ranked_leads entries include all qualified exhibitor fields plus rank, crm_account_id, crm_action, requested_demo, interest_score, priority_tier, opportunity_estimate_usd.
• excluded_exhibitors entries include company_id, company_name, relationship_type, exclusion_reason, crm_action.

5. Ordering Rules

Apply these sort orders consistently:

6. CRM Action Rules

7. Common Pitfalls

1. Wrong show name in CSV — use the human-readable show name for filenames/context, but the machine show_id in JSON show_id fields.
2. Missing zero counts — platform_counts and priority_counts must explicitly include every expected key with 0 when empty.
3. Platform ordering — JSON arrays and CSV pipe-strings must follow AUV → ROV → Underwater Camera.
4. Exclusion reason vocabulary — sensor_vendor_only vs sensor_only differ by task type; match the template enum exactly.
5. Rank contiguity — rank in ranked_leads must be 1-based and contiguous (1, 2, 3, …) with no gaps.
6. Not deduplicating CRM — always check /crm/accounts by website domain before deciding create_account vs update_existing.
7. Over-querying — never call global list endpoints to discover IDs; only use IDs explicitly provided in the prompt.
8. Answer template mismatch — if the on-disk answer_template.json does not match the prompt, trust the prompt description and the solved output/answer.json as the ground-truth schema.