get-qualified-leads-from-luma（从 luma 获得合格的销售线索）

Get Qualified Leads from Luma Events

Search Luma for events by topic and location, extract all attendees and hosts, qualify them against your ICP, export to a Google Sheet, and send a Slack alert with the top leads.

This is a 5-step pipeline that chains together luma-event-attendees, lead-qualification, Google Sheets output, and Slack alerting.

Step 0: Clarify Search Parameters

Before doing anything, make sure you have clear answers to these questions. If the user's prompt already covers them, skip ahead. Otherwise, ask:

1. Location — Where should events be? (e.g., "San Francisco", "New York", "London")
2. Topics/Keywords — What event topics? Suggest 3-5 keyword variations to maximize coverage. For example, if the user says "growth marketing", also suggest: "GTM", "demand gen", "startup growth", "growth hacking", "marketing leadership"
3. Timeframe — How recent should the events be? (e.g., "past 2 weeks", "past month", "this quarter"). Default to past 30 days if the user doesn't specify. Luma search can return events from months or years ago, so always confirm a timeframe to avoid stale results.
4. Qualification prompt — Does the user have an existing qualification prompt in skills/lead-qualification/qualification-prompts/? If not, what's their ICP at a high level? (Can use lead-qualification intake mode to build one)
5. Slack channel/webhook — Where should the alert go? A webhook URL or Slack channel name?
6. How many top leads in the Slack alert? (default: 5)

Present these as a numbered list. The user can answer in one shot.

Step 1: Search Luma and Extract Attendees

Use the luma-event-attendees skill with multiple keyword variations to maximize coverage.

Run parallel searches

Generate 3-5 keyword variations combining the user's topic with their location. Run them all in parallel:

# Run each search variation in parallel
python3 skills/luma-event-attendees/scripts/scrape_event.py --search "AI San Francisco" --output /tmp/luma_search_1.csv
python3 skills/luma-event-attendees/scripts/scrape_event.py --search "Growth Marketing San Francisco" --output /tmp/luma_search_2.csv
python3 skills/luma-event-attendees/scripts/scrape_event.py --search "GTM San Francisco" --output /tmp/luma_search_3.csv

Filter by timeframe

After collecting results, filter out events outside the user's specified timeframe using the event_date column. Luma search returns events from all time periods, so this step is essential to avoid stale leads. If no timeframe was specified, default to the past 30 days.

Deduplicate

Merge and deduplicate by name (case-insensitive). Handle None names gracefully — skip entries with no name.

Save the deduplicated result as a CSV:

/tmp/luma_all_attendees.csv

Report to the user:

• How many total results before dedup
• How many unique people after dedup
• How many have LinkedIn profiles
• How many events were covered

Step 2: Save Attendee Data to CSV

Work with CSVs throughout the pipeline — Google Sheets creation happens only at the end (Step 4) because writing large datasets to Sheets mid-process is slow and error-prone.

The CSV from Step 1 (/tmp/luma_all_attendees.csv) is your working file. Columns should include:

Step 3: Qualify Leads

Use the lead-qualification skill (Mode 2: reuse prompt) to qualify all attendees.

Prepare batches

1. Read the qualification prompt from the file the user specified (e.g., skills/lead-qualification/qualification-prompts/ai-event-attendees-gtm.md)
2. Split attendees into batches of ~15-20 leads each
3. For each lead, include: id (row number), name, event_role, bio, title, company, linkedin_url, event_name

Run parallel qualification

Launch all batches simultaneously using the Task tool with sonnet model subagents:

Task: "Qualify leads batch 1/N"
  - Include the full qualification prompt text
  - Include the batch of leads as JSON
  - Ask for output as JSON array: [{id, name, qualified, confidence, reasoning}]

Task: "Qualify leads batch 2/N"
  ... (launch ALL at once)

Merge results

1. Collect all batch results
2. Merge into a single JSON array, preserving original IDs
3. Sort qualified leads by confidence (High first, then Medium, then Low)
4. Save results:

- /tmp/all_qual_results.json — all 195 results - /tmp/qualified_leads.json — only qualified leads, sorted by confidence

Report to the user:

• Total leads processed
• Qualified count and percentage
• Breakdown by confidence level (High / Medium / Low)
• Top disqualification reasons

Step 4: Create Google Sheet with Results

Now create the Google Sheet with all data — both raw attendees and qualification results.

Use Rube MCP for Google Sheets

1. Use RUBE_SEARCH_TOOLS to find Google Sheets tools (search for "google sheet create")
2. Create a new sheet named: Luma Leads - [Topic] - [Date]
3. Sheet 1 ("All Attendees"): Write all attendee rows with original columns PLUS:

- Qualified — Yes / No - Confidence — High / Medium / Low - Reasoning — 2-3 sentence explanation

1. Sheet 2 ("Qualified Leads"): Only qualified leads, sorted by confidence

Writing strategy for large datasets

The Google Sheets API can be slow for large datasets. Use this approach:

• Write the header row first
• Write data in chunks of 50 rows using batch update operations
• If a chunk fails, retry once before moving on

Fallback

If Rube/Sheets is unavailable, save as CSV:

/tmp/luma_qualified_leads_[date].csv

Present the Google Sheet link (or CSV path) to the user.

Step 5: Send Slack Alert

Send a formatted Slack message with the top N qualified leads (default: 5, or whatever the user specified).

If the user provided a webhook URL

Use Python with urllib.request to POST to the webhook:

import json
import urllib.request

message = {
    "blocks": [
        {"type": "header", "text": {"type": "plain_text", "text": "Top N Qualified Leads from [Topic] Events"}},
        {"type": "section", "text": {"type": "mrkdwn", "text": "_From X attendees across Y events, Z qualified (P%). Here are the top N:_"}},
        # For each lead:
        {"type": "section", "text": {"type": "mrkdwn", "text": "*1. Name* [Confidence]\n   LinkedIn: url\n   Bio: ...\n   Why: reasoning"}},
        {"type": "divider"},
        # Link to spreadsheet at the bottom
        {"type": "section", "text": {"type": "mrkdwn", "text": "<sheet_url|View full spreadsheet> (X attendees, Y qualified)"}}
    ]
}

req = urllib.request.Request(webhook_url, data=json.dumps(message).encode(), headers={"Content-Type": "application/json"})
urllib.request.urlopen(req)

If the user wants Slack via Rube MCP

Use RUBE_SEARCH_TOOLS to find Slack tools, then send via SLACK_SEND_MESSAGE or similar.

Message format

The Slack alert should include for each top lead:

• Name and confidence level
• LinkedIn URL (clickable)
• Bio — one-line summary
• Why — the qualification reasoning (truncated to ~150 chars if needed)

End with a link to the full Google Sheet.

Cost Estimate

Typical run: ~200 attendees across 3-5 search variations costs essentially just the Apify subscription + a few cents in LLM tokens.

Example Prompts

Quick run with existing prompt:

"Find qualified leads from AI and growth events in SF. Use the ai-event-attendees-gtm qualification prompt. Send top 5 to Slack webhook: https://hooks.slack.com/..."

Full specification:

"Search Luma for startup, SaaS, and AI events in New York. Extract all attendees. Qualify them against our Series A founders ICP. Put everything in a Google Sheet and Slack me the top 10."

Minimal (triggers clarifying questions):

"Find me leads from SF tech events"

Troubleshooting

Apify token not set

export APIFY_API_TOKEN="your_token"
# Or check skills/luma-event-attendees/.env

No guests found

Some Luma events have show_guest_list disabled. The Apify scraper can still get featured guests, but full attendee lists may not be available for all events.

Google Sheets writing is slow

This is normal for large datasets. The skill writes in 50-row chunks. If it's too slow or fails, results are always available as CSV in /tmp/.

Slack webhook returns error

Verify the webhook URL is correct and the Slack app is still installed in the workspace. Test with a simple curl:

curl -X POST -H 'Content-Type: application/json' -d '{"text":"test"}' YOUR_WEBHOOK_URL