local-client-discovery

Files (7)

• README.md
• SKILL.md
• openai.yaml · agents/
• manifest.json
• query-patterns.md · references/
• schema-and-scoring.md · references/
• normalize_leads.py · scripts/

SKILL.md (first lines)

# Local Client Discovery

Use this skill to build clean outreach-ready lead lists for local businesses in a specific geography.

## Use this skill when
- The user wants leads in a precise place, for example a city, island, district, or country.
- The target is local business outreach, for example restaurants, bars, beach clubs, cafes, agencies, hotels, or venues.
- The task involves search APIs, scraping, Instagram profile matching, or ranking prospects.
- The user wants CSV or JSON outputs suitable for outreach.

## Core workflow
1. Define the target clearly:
   - geography
   - business type
   - outreach angle
2. Choose one or more source paths:
   - search API results
   - search-engine queries
   - Google Maps style business discovery
   - Instagram profile discovery
   - website/contact-page discovery
3. Normalize records into a common schema.
4. Add simple scoring signals.
5. Export a clean lead file and a short note about how the leads were sourced.

## Recommended record shape
Use these fields when possible:
- business_name
- area
- full_location_query
- address
- business_type
- category_for_outreach
- instagram_url
- instagram_match_type
- google_rating
- website
- foreign_owner_signal
- recent_business_signal
- outreach_priority
- search_title
- search_snippet

## Scoring heuristics
Prefer lightweight explainable scoring over complex ranking.

Useful signals:
- category fit
- location fit
- visible Instagram presence
- website present
- contact route present
- foreign-owner or international-brand signal when relevant
- signs of recent activity
- premium visual positioning when relevant to the offer

## Output rules
- Prefer CSV for human review.
- Also write JSON when the list will be consumed by code.
- Keep the raw source query or route when possible.
- Include one short sourcing note with assumptions and caveats.

## Quality rules
- Do not pretend a lead is verified if it is only inferred from search snippets.
- Separate profile matches from post-only matches.
- Keep the pipeline easy to audit.
- Deduplicate aggressively.

## Bundled resources
- Read `references/schema-and-scoring.md` before designing or normalizing a dataset.
- Read `references/query-patterns.md` when you need search ideas for local business discovery.
- Use `scripts/normalize_leads.py` to convert raw CSV or JSON into a normalized scored export.

From ruota/open-sinapseed, MIT.

More from ruota/open-sinapseed