Imported from kidclone3/deal-flow-matchmaker (.feynman/agent/skills/entity-enrichment/SKILL.md). Install upstream with npx skills add kidclone3/deal-flow-matchmaker --skill entity-enrichment. Copyright stays with the author.

Entity Enrichment

Collect structured profiles for VC funds, angel investors, and startups by searching and extracting from multiple public sources in parallel. Outputs a provenance-tracked JSON profile with confidence flags per field.

When to use

• Populating a deal-flow database with real company/investor data
• Enriching startup profiles before running a matching engine
• Building a seed dataset for a matchmaking platform
• Any time you need to collect structured entity data from the open web

Entity types and field coverage

This skill targets three entity types with different collection strategies:

Collection workflow

Phase 1: Source discovery (parallel search)

For each entity, run 4-6 searches with different angles simultaneously using web_search queries parameter:

For a VC fund named "Example Ventures":

1. "Example Ventures fund size AUM partners portfolio"
2. "Example Ventures investment thesis stage check size sectors"
3. "Example Ventures Crunchbase PitchBook profile"
4. "Example Ventures portfolio companies exits"
5. "Example Ventures Vietnam investments Asia Southeast Asia"
6. site:exampleventures.com "team" OR "portfolio" OR "strategy"

For an angel investor named "Jane Nguyen":

1. "Jane Nguyen angel investor portfolio investments"
2. "Jane Nguyen LinkedIn Crunchbase"
3. "Jane Nguyen startup investments exits"
4. "Jane Nguyen" site:linkedin.com

NEVER search for: DOB, passport, ID, tax ID, KYC, personal address, personal phone. Only search for: name + "investor" / "angel" / "portfolio" / "startup" / "LinkedIn" / "investments".

For a startup named "Example AI":

1. "Example AI startup Crunchbase funding valuation"
2. "Example AI founder team LinkedIn"
3. "Example AI product customers revenue growth"
4. "Example AI competitors market"
5. "Example AI funding round investors"
6. site:example.ai

Phase 2: Deep extraction (parallel fetch)

From Phase 1 results, identify the 3-6 highest-signal URLs and fetch them in parallel with fetch_content urls parameter:

Priority URL types (ordered by signal strength):

1. Crunchbase / PitchBook / Dealroom profile pages
2. The entity's own website (About, Team, Portfolio pages)
3. LinkedIn company/person page
4. Press releases announcing funding rounds
5. SEC Form D / ADV filings (US funds)
6. TechCrunch / VnExpress / DealStreetAsia coverage
7. startup.gov.vn / Viet.io listings (Vietnam-specific)

Phase 3: Document extraction (if applicable)

If the startup has a public pitch deck or the VC has a public fund factsheet PDF:

• Use document_parse to extract text from the PDF
• Use LLM extraction prompt (see Section 5 of dealflow research doc) to pull structured fields

Phase 4: Merge and annotate

Consolidate all extracted data into the output schema. For every field, record:

{
  "field": "ticket_size",
  "value": "$500K-$2M",
  "source_url": "https://crunchbase.com/organization/example-ventures",
  "source_type": "crunchbase",
  "confidence": "high"
}

Confidence rules:

• high — from entity's own website, official filing, or Crunchbase verified profile
• medium — from press article, LinkedIn, or third-party database
• low — inferred from context, single mention, or outdated source
• unavailable — searched but not found in any source

Phase 5: Output (agent→spine contract, R3)

Do NOT write files. Emit the enriched profile as your terminal message: a single fenced ```json block conforming to schemas/enrich.json. No prose before or after it. The write tool is for scratch notes only — the spine reads the JSON block, not the disk.

Top-level keys:

• entity_type — one of vc | angel | startup | corporation | university | research_institution
• the provenance-tracked profile sections below (each field {value, source_url, confidence})
• relationships — an array of graph edges surfaced during research, each {"kind": one of ["raised_from","invested_in","pilot_with","founded_by_alumni_of","partner_with"], "dst_name": "...", "source_url": "..."}, written from THIS entity's perspective (a startup raised_from a fund, is founded_by_alumni_of a university, etc.)
• collection_summary

Output schema

VC Fund

{
  "entity_type": "vc",
  "enriched_at": "2026-07-18T...",
  "legal_information": {
    "fund_name": {"value": "...", "source_url": "...", "confidence": "high"},
    "legal_entity": {"value": "...", "source_url": "...", "confidence": "medium"},
    "country": {"value": "...", "source_url": "...", "confidence": "high"},
    "founded": {"value": 2020, "source_url": "...", "confidence": "high"},
    "aum": {"value": "...", "source_url": "...", "confidence": "medium"},
    "gp_partners": {"value": ["..."], "source_url": "...", "confidence": "high"},
    "website": {"value": "...", "source_url": "...", "confidence": "high"},
    "linkedin": {"value": "...", "source_url": "...", "confidence": "high"}
  },
  "investment_thesis": {
    "stage": {"value": ["seed", "series_a"], "source_url": "...", "confidence": "high"},
    "check_size": {"value": "$500K-$2M", "source_url": "...", "confidence": "high"},
    "sector_focus": {"value": ["ai", "fintech"], "source_url": "...", "confidence": "high"},
    "geography": {"value": ["Vietnam", "Southeast Asia"], "source_url": "...", "confidence": "high"}
  },
  "portfolio": {
    "companies": {"value": [{"name": "...", "round": "seed"}], "source_url": "...", "confidence": "medium"},
    "vietnam_exposure": {"value": true, "source_url": "...", "confidence": "medium"},
    "notable_exits": {"value": ["..."], "source_url": "...", "confidence": "medium"},
    "lead_co_invest": {"value": "both", "source_url": "...", "confidence": "low"},
    "follow_on_rate": {"value": null, "source_url": null, "confidence": "unavailable"}
  },
  "deal_sourcing": {
    "key_contacts": {"value": [{"name": "...", "role": "Partner", "linkedin": "..."}], "source_url": "...", "confidence": "medium"},
    "emails": {"value": ["..."], "source_url": "...", "confidence": "low"},
    "warm_intro_status": {"value": null, "source_url": null, "confidence": "unavailable"}
  },
  "collection_summary": {
    "total_fields": 20,
    "fields_collected": 17,
    "fields_unavailable": 3,
    "sources_used": ["crunchbase.com", "exampleventures.com", "techcrunch.com"],
    "gaps": ["follow_on_rate", "warm_intro_status", "lead_co_invest"]
  }
}

Angel Investor

{
  "entity_type": "angel",
  "enriched_at": "2026-07-18T...",
  "identity": {
    "full_name": {"value": "...", "source_url": "...", "confidence": "high"},
    "nationality": {"value": "...", "source_url": "...", "confidence": "medium"}
  },
  "investment_profile": {
    "ticket_size": {"value": "$25K-$100K", "source_url": "...", "confidence": "medium"},
    "stage": {"value": ["pre_seed", "seed"], "source_url": "...", "confidence": "medium"},
    "sector": {"value": ["ai", "healthtech"], "source_url": "...", "confidence": "medium"},
    "geography": {"value": ["Vietnam"], "source_url": "...", "confidence": "medium"},
    "frequency": {"value": null, "source_url": null, "confidence": "unavailable"},
    "lead_co_invest": {"value": null, "source_url": null, "confidence": "unavailable"}
  },
  "portfolio": {
    "investments": {"value": [{"name": "Startup X", "year": 2024}], "source_url": "...", "confidence": "medium"},
    "exits": {"value": [], "source_url": "...", "confidence": "medium"},
    "board_seats": {"value": [], "source_url": "...", "confidence": "low"},
    "notable_startups": {"value": ["Startup X"], "source_url": "...", "confidence": "medium"}
  },
  "professional": {
    "current_company": {"value": "...", "source_url": "...", "confidence": "high"},
    "role": {"value": "...", "source_url": "...", "confidence": "high"},
    "experience": {"value": ["..."], "source_url": "...", "confidence": "high"},
    "education": {"value": ["..."], "source_url": "...", "confidence": "high"},
    "linkedin": {"value": "...", "source_url": "...", "confidence": "high"}
  },
  "pii_blocked": ["dob", "id_passport", "tax_id", "kyc_status", "accredited_status", "personal_phone", "personal_address"],
  "collection_summary": {
    "total_fields": 18,
    "fields_collected": 13,
    "pii_skipped": 7,
    "fields_unavailable": 2,
    "sources_used": ["linkedin.com", "crunchbase.com", "angellist.com"],
    "gaps": ["frequency", "lead_co_invest"]
  }
}

Startup

{
  "entity_type": "startup",
  "enriched_at": "2026-07-18T...",
  "company": {
    "legal_entity": {"value": "...", "source_url": "...", "confidence": "high"},
    "founded": {"value": 2023, "source_url": "...", "confidence": "high"},
    "country": {"value": "Vietnam", "source_url": "...", "confidence": "high"},
    "website": {"value": "...", "source_url": "...", "confidence": "high"},
    "industry": {"value": ["agritech", "ai"], "source_url": "...", "confidence": "high"}
  },
  "founder": {
    "name": {"value": "...", "source_url": "...", "confidence": "high"},
    "background": {"value": "...", "source_url": "...", "confidence": "medium"},
    "education": {"value": ["..."], "source_url": "...", "confidence": "medium"},
    "linkedin": {"value": "...", "source_url": "...", "confidence": "high"},
    "previous_exits": {"value": [], "source_url": "...", "confidence": "low"}
  },
  "product": {
    "description": {"value": "...", "source_url": "...", "confidence": "high"},
    "business_model": {"value": "B2B SaaS", "source_url": "...", "confidence": "medium"},
    "target_customer": {"value": "...", "source_url": "...", "confidence": "medium"},
    "usp": {"value": "...", "source_url": "...", "confidence": "medium"},
    "competitors": {"value": ["..."], "source_url": "...", "confidence": "low"}
  },
  "traction": {
    "users": {"value": null, "source_url": null, "confidence": "unavailable"},
    "customers": {"value": null, "source_url": null, "confidence": "unavailable"},
    "mau_dau": {"value": null, "source_url": null, "confidence": "unavailable"},
    "revenue": {"value": null, "source_url": null, "confidence": "unavailable"},
    "arr_mrr": {"value": null, "source_url": null, "confidence": "unavailable"},
    "growth": {"value": null, "source_url": null, "confidence": "unavailable"},
    "retention": {"value": null, "source_url": null, "confidence": "unavailable"}
  },
  "financial": {
    "funding_raised": {"value": "$1.2M", "source_url": "...", "confidence": "high"},
    "valuation": {"value": null, "source_url": null, "confidence": "unavailable"},
    "burn_rate": {"value": null, "source_url": null, "confidence": "unavailable"},
    "runway": {"value": null, "source_url": null, "confidence": "unavailable"}
  },
  "fundraising": {
    "current_round": {"value": "Seed", "source_url": "...", "confidence": "high"},
    "amount_seeking": {"value": "$2M", "source_url": "...", "confidence": "medium"},
    "existing_investors": {"value": ["..."], "source_url": "...", "confidence": "high"},
    "cap_table": {"value": null, "source_url": null, "confidence": "unavailable"}
  },
  "collection_summary": {
    "total_fields": 28,
    "fields_collected": 16,
    "fields_unavailable": 12,
    "sources_used": ["crunchbase.com", "example.ai", "techcrunch.com", "linkedin.com"],
    "gaps": ["users", "customers", "mau_dau", "revenue", "arr_mrr", "growth", "retention", "valuation", "burn_rate", "runway", "cap_table"]
  }
}

Source priority by entity type

VC Fund — sources ranked by reliability

Angel Investor — sources ranked by reliability

PII wall: Stop at professional data. Do not search for or attempt to extract:

• DOB, age
• ID numbers, passport numbers
• Tax IDs, social security numbers
• Home addresses
• Personal phone numbers
• KYC status, accredited investor verification

If a source accidentally surfaces this data, do not include it in the output. Mark it in pii_blocked.

Startup — sources ranked by reliability

Vietnam-specific sources

Always include these when enriching Vietnam-based entities:

Batch enrichment mode

When enriching multiple entities (e.g., all 25 NIC-supported startups):

1. Parallel search phase: Run web_search queries for 3-5 entities simultaneously (interleave queries so each entity gets 1-2 queries per batch)
2. Parallel fetch phase: Gather all high-signal URLs across entities, fetch 4-6 at a time with fetch_content urls

Note: the spine dispatches one entity per agent job, so each run emits exactly one entity's JSON block (per Phase 5). Batch parallelism applies to searches/fetches within that single-entity job, not to emitting many profiles at once.

Important rules

0. Emit JSON, not files (R3). Your terminal message is a single fenced ```json block per schemas/enrich.json. write is scratch-only.
1. Always record provenance. Every field must have a source_url. If you can't find a source, mark it unavailable — don't invent data.
2. Never collect PII for individuals. The pii_blocked list in the angel schema is hard enforcement, not a guideline.
3. Prefer entity-owned sources. The startup's own website beats Crunchbase beats a press article.
4. Flag low confidence. If you're inferring from context (e.g., guessing sector from company description), mark it low.
5. Date everything. Include enriched_at timestamp so downstream consumers know data freshness.
6. Do not paywall-shim. If data is behind a Crunchbase/PitchBook paywall, mark it unavailable and note the paywall. Do not try to bypass.
7. Vietnam first. For Vietnamese entities, always check startup.gov.vn, Viet.io, and VnExpress before global sources.
8. Always emit collection_summary.status — one of:
   • complete — most fields collected with sources.
   • partial — some fields collected, others unavailable.
   • not_found — the entity itself could not be confirmed to exist: its website does not resolve (NXDOMAIN / unreachable), and no independent source (Crunchbase, LinkedIn, press, Wikipedia, local registries) corroborates it. Do NOT fabricate a profile to fill the schema — return every field null/unavailable, put your evidence in collection_summary.finding, and set status: "not_found". The spine gates not_found (and all-null) entities to status='unverified' so they never enter matching. A wrong seed name/domain is a real outcome, not a failure to hide.