event-prospecting-pipeline

name	event-prospecting-pipeline
description	End-to-end event-driven prospecting — find attendees/speakers at conferences and events, research and qualify them against ICP, find decision-maker contacts, deduplicate against the contact cache, and hand off a clean qualified list for outreach. Orchestrates the event, qualification, contact-finding, and cache sub-skills.
metadata	{"version":"1.0.0","category":"lead-generation","type":"playbook"}

Event Prospecting Pipeline

A thin orchestration over existing sub-skills. The pipeline owns no scrapers itself — it routes to each sub-skill and lets the agent synthesize the final qualified list. (For Luma-only with Sheets + Slack alerting, use get-qualified-leads-from-luma.)

When to use

• "Find leads from [event/conference]." / "Who's speaking at [conference]? Get contacts."
• "Find AI events in SF and get decision-maker contacts."
• Full pipeline including outreach prep.

Sub-skills orchestrated

1. Source people — luma-event-attendees (Luma event/topic → attendees) or conference-speaker-scraper (conference site → speakers).
2. Research & enrich — keyless web research (run inbound-lead-enrichment's web_research.py, or the serp Search/Extract nodes) for funding stage, size, product, role, recent news. Skip this step if the user only wants a raw attendee list.
3. Qualify — lead-qualification (score each lead against the ICP).
4. Find decision-makers — company-contact-finder (Apollo + keyless fallback) for qualified companies; verify emails with an email-verify skill if available.
5. Dedup — contact-cache (check op) against known/contacted people by LinkedIn URL/email. Mandatory before outreach.
6. Export — the deduped qualified leads to a sheet/CSV + Agent Teams channel attachment.

How to run

Invoke each sub-skill's scripts by path, threading the output of one into the next, e.g.:

# 1. source (conference example)
python3 ${SKILL_DIR}/../conference-speaker-scraper/scripts/scrape_speakers.py \
  --url https://conf.example.com/speakers --output ${WORKSPACE}/people.json

# 3. qualify (after enrich)
python3 ${SKILL_DIR}/../lead-qualification/scripts/qual_prompt.py scaffold \
  --leads ${WORKSPACE}/people.csv --output ${WORKSPACE}/scorecard.json
# ... agent fills verdicts ...

# 4. contacts for qualified companies
python3 ${SKILL_DIR}/../company-contact-finder/scripts/apollo_people.py \
  --company "Acme Corp" --titles "VP Eng,CTO" --output ${WORKSPACE}/contacts.json

# 5. dedup (mandatory)
python3 ${SKILL_DIR}/../contact-cache/scripts/contact_cache.py \
  --store ${WORKSPACE}/contacts.csv check --linkedin-urls "<urls>"

Each sub-skill owns its own proxy/rate-limit/anti-block and degrade behavior — follow its SKILL.md when an optional key is missing.

Outputs

A qualified, deduplicated lead list (Name, Title, Company, LinkedIn URL, Email, Signal, Score) exported to a sheet/CSV + channel attachment, ready for outreach.

Credentials / env

• Required: none at the playbook level — gates live in the sub-skills it invokes.
• Optional (inherited per sub-skill): APIFY_API_TOKEN (Luma search mode / conference JS fallback), APOLLO_API_KEY (contact-finding), SUPABASE_URL + SUPABASE_SERVICE_ROLE_KEY (cache — required by contact-cache for durable cross-run dedup), email-verify keys, ANTHROPIC_API_KEY if not platform-provided.

Notes & edge cases

• Skip step 2 if the user only wants a raw attendee list.
• Each sub-skill owns its proxy/rate-limit/anti-block handling.
• Dedup (step 5) is mandatory before outreach to prevent re-contacting known leads.
• Degrade gracefully per each sub-skill's notes when an optional key is missing, and flag reduced depth.