onfire-prospecting

The orchestrator for the OnFire MCP. Reads user intent first, normalises the inputs, then routes the work across the five atomic skills. Read this one first; it picks the path. Each atomic skill it points to covers the input shapes, batch caps, output fields, polling rules, and pitfalls in detail.

When the agent picks this skill

The user is asking for any of:

• Prospecting a company or list of companies (find buyers, champions, decision-makers).
• Scoring a specific person at a company.
• Resolving people or companies to verified LinkedIn profiles.
• Enriching contacts with emails or phones.
• Editing the prospecting schema (regions, keywords, prompts, ranking).

Trigger phrases the agent listens for: “prospect this company”, “find buyers at X”, “who should I talk to at Y”, “score Lana”, “get me emails for this list”, “what’s Lana Patel’s LinkedIn”, “tweak my hot keywords”.

Atomic skills it routes to

Skill	When this skill defers to it
match-company	Need a verified company LinkedIn URL before prospecting.
match-person	Need a verified person LinkedIn URL before scoring or enriching.
ai-prospecting	Run a ranked list (run) or score one person (get_prospect).
contact-data-enrichment	Append emails / phones — strictly enforced consent above 10 contacts.
manage-ai-prospecting	Edit the schema via the shadow lifecycle.

Cross-cutting rules it enforces

• Pick from intent, not input shape. A bare website doesn’t imply prospecting; “find me champions at acme.com” does.
• Always offer enrichment after a prospecting run — it is the standing follow-up, not optional.
• Dataset handles flow through the pipeline. An ai_prospecting run returns a dataset_id; hand it straight to contact_data_enrichment (or to query_datasets for slicing) instead of rebuilding contact dicts.
• Surface download URLs verbatim. When the user wants the CSV, the agent calls download_dataset and pastes the returned download_url directly — that link is the deliverable.

Underlying MCP tools

match_company · match_person · ai_prospecting · contact_data_enrichment · manage_ai_prospecting · describe_dataset · query_datasets · download_dataset · list_datasets