Verify Leads Skill

You help the user qualify candidate companies by verifying their existence and applying business exclusion rules.

This skill moves candidates from pending status to either:

•Verified → moved to companies.csv (qualified prospects)
•Dropped → marked in candidates.csv with reason (excluded)

IMPORTANT: Always Start by Reading User Parameters

FIRST STEP - Read user parameters:

code

Read /data/user-parameters.md
Extract:
- Exclusions (SGX, MNC, holdco, size)
- Target sectors
- Current banker identification strategy

Step 1: Check for Uncommitted Changes

Before starting verification:

code

Check git status for uncommitted changes in:
- candidates.csv
- companies.csv

IF uncommitted changes exist:
  "You have uncommitted changes in [files].
   Commit before verifying? (yes/no)"

IF user says yes:
  Wait for user to commit ("commit this")
  Then proceed

IF user says no:
  Proceed with verification

Step 2: Read Candidates to Verify

Read /data/candidates.csv

Filter:

•WHERE status='pending'
•Optional: If user specified sector (e.g., "Verify logistics candidates"), filter by source or company name pattern

Count pending candidates

IF 0 pending candidates:

code

"No pending candidates to verify.

Current candidates.csv status:
- [X] already verified
- [Y] already dropped

Add new candidates with: 'Scan [company]' or 'Find [criteria]'"

STOP

IF > 0 pending candidates:

code

"Found [N] pending candidates to verify.
Processing..."

Continue to Step 3

Step 3: Verify Each Candidate

For EACH candidate with status='pending':

3a) Existence Verification

Check website:

code

Search: "[company_name] official website Singapore"
Look for official domain

Check LinkedIn:

code

Search: "[company_name] LinkedIn company"
Look for linkedin.com/company/[name]

IF NEITHER website NOR LinkedIn found:

•Mark as dropped
•Reason: "No website/LinkedIn found"
•SKIP to next candidate

IF website OR LinkedIn found:

•Continue to 3b

3b) MNC Subsidiary Check

Look for patterns in website/search results:

code

1. "[Company] subsidiary of [Parent MNC]"
2. "[Company] part of [Global Group]"
3. Website footer: "A [MNC] company"
4. About page mentions international parent
5. Global brand names (FedEx, DHL, UPS, Siemens, ABB, Schneider, etc.)

Common MNC patterns to watch:

•Courier: FedEx, DHL, UPS, TNT
•Logistics: DB Schenker, Kuehne+Nagel, DSV
•Industrial: Siemens, ABB, Schneider Electric
•Consulting: Accenture, IBM, Deloitte, PwC
•Tech: Microsoft, Oracle, SAP, Salesforce

IF MNC subsidiary:

•Mark as dropped
•Reason: "MNC subsidiary of [Parent]"
•SKIP to next candidate

IF not MNC:

•Continue to 3c

3c) Current Banker Identification

Search for banking relationships:

code

"[company_name] facility [bank]"
"[company_name] financing [bank]"
"[company_name] loan [bank]"
"[company_name] banking partner"

Look for mentions in:

•Press releases
•News articles (Business Times, Straits Times)
•Company announcements

Key phrases to identify current banker:

code

- "secured S$XXM facility from [Bank Name]"
- "banking partner [Bank]"
- "facility from [Bank]"
- "financed by [Bank]"
- "refinanced with [Bank]"
- "syndicated loan led by [Bank]"
- "[Bank] provided working capital"

Common Singapore banks:

•OCBC, UOB, DBS, Standard Chartered, HSBC, Maybank, CIMB, ANZ, Citibank

Record findings:

•IF banker mentioned → Note the bank name (will be added to companies.csv)
•IF no banker mentioned → Leave blank (not "unknown")
•IF multiple banks mentioned → Record primary/lead banker only

IMPORTANT:

•This is intelligence gathering, NOT an exclusion check
•Record banker information for strategic value
•Continue to 3d regardless of findings

3d) Holdco / Size Plausibility Check

Look for signals:

Red flags for >S$100M:

•"Holdings" or "Group" in legal name
•Listed parent company
•Wide regional presence (>5 countries, >20 locations)
•Employee count >1,000
•Multiple subsidiaries across industries
•Revenue publicly disclosed >S$100M

Search:

code

"[company_name] revenue"
"[company_name] group structure"
"[company_name] parent company"

Decision logic:

code

IF obvious signals (listed parent, disclosed revenue >100M, massive scale):
  Mark as dropped
  Reason: "Holdco revenue >S$100M"
  SKIP to next candidate

IF some signals but unclear (e.g., "Holdings" in name but no other info):
  Flag for review but DON'T auto-drop
  Set holdco_flag='yes' in companies.csv
  Continue to 3e (let user decide later)

IF no signals:
  Set holdco_flag='no'
  Continue to 3e

3e) Sector Validation

Check if company's sector is in target sectors (from user-parameters.md):

•Logistics, Supply Chain & Trade
•Manufacturing & Industrial
•Healthcare, Wellness & Life Sciences
•F&B, Retail & Consumer
•Technology & Professional Services

IF sector not in target list:

•Mark as dropped
•Reason: "Out of target sectors ([sector])"
•SKIP to next candidate

IF sector in target list OR unknown:

•Continue to 3f

3f) Extract Company Details (Passed All Checks)

If candidate passed all exclusion checks:

Step 3f-1: ACRA Lookup (UEN + Paid-up Capital)

Search for ACRA data using WebSearch:

code

Search query: "[Legal Name] UEN Singapore ACRA paid-up capital"

Look for results from:
- companies.sg
- sgpbusiness.com
- recordowl.com
- opengovsg.com

Extract:
- UEN (Unique Entity Number) - 9-10 digit code
- Paid-up Capital - SGD amount
- Company Status - Live/Struck Off
- Incorporation Date (optional)

Format paid-up capital as: "S$XXM" or "S$XX.XM"
Examples: "S$10.7M", "S$1M", "S$500K"

IF no ACRA data found:
  Leave UEN and paid_up_capital empty
  Continue with other details

Step 3f-2: Extract from website/LinkedIn:

code

- Legal name (if found, e.g., "[Name] Pte Ltd")
- Website URL
- LinkedIn URL
- Annual revenue (if publicly disclosed; otherwise estimate with "~" prefix)
  Format: "S$XXM" (declared) or "~S$XXM" (estimated)
  Examples: "S$45M", "~S$30M", "S$400M+"
  Use paid-up capital as reference if available (typical 5-20x multiplier)
- Sector (from user-parameters.md target list)
- Industry (sub-sector, e.g., "3PL/4PL", "Electronics", "Medical Devices")

Set defaults:

code

- priority = medium (user will adjust later if needed)
- confidence = MEDIUM (can upgrade to HIGH if award-winning, etc.)
- date_verified = today (YYYY-MM-DD)
- last_enriched = never
- current_stage = verified

Generate company_id:

code

- Kebab-case from company name
- Example: "YCH Group" → "ych-group"
- Example: "Supreme Components International" → "supreme-components-international"

Add to companies.csv:

CRITICAL CSV FORMAT REQUIREMENT:

•ALL 18 columns must be present for every row
•Empty fields must use empty string (consecutive commas: ,,)
•NEVER skip columns - preserve column order exactly
•UEN, paid_up_capital, and annual_revenue fields may be empty - must still include comma placeholders
•VALIDATION: Every row must have exactly 17 commas (18 fields total)

New column order:

code

company_id,legal_name,common_name,website,linkedin,uen,paid_up_capital,annual_revenue,sector,industry,priority,confidence,holdco_flag,date_verified,last_enriched,current_stage,assigned_to,notes

Example with all fields:

csv

ych-group,YCH Group Pte Ltd,YCH Group,https://www.ych.com,https://linkedin.com/company/ych-group,198003684Z,S$10.7M,~S$80M,Logistics,3PL/4PL,medium,HIGH,no,2025-12-30,never,verified,,Company notes here

^ 17 commas = 18 fields ✓

Example with missing website/LinkedIn/UEN but with paid-up capital:

csv

example-company,Example Company Pte Ltd,Example Company,,,200104742C,S$1M,~S$20M,Manufacturing,Electronics,medium,HIGH,no,2025-12-30,never,verified,,Award winner

^ Count after common_name: 3 empty fields (website, linkedin, uen) = 3 commas, then paid_up_capital = 17 commas total ✓

Example with all financial fields empty:

csv

new-company,New Company Pte Ltd,New Company,https://example.com,,,,~S$15M,Technology,SaaS,medium,MEDIUM,no,2025-12-30,never,verified,,

^ Count after common_name: website (filled), then 3 empty fields (linkedin, uen, paid_up_capital) = 3 commas before annual_revenue = 17 commas total ✓

CSV Row Construction Checklist: Before writing each row, verify the structure:

•company_id (field 1)
•legal_name (field 2)
•common_name (field 3)
•website (field 4) - may be empty
•linkedin (field 5) - may be empty
•uen (field 6) - may be empty
•paid_up_capital (field 7) - may be empty
•annual_revenue (field 8) - may be empty
•sector (field 9)
•industry (field 10)
•priority (field 11)
•confidence (field 12)
•holdco_flag (field 13)
•date_verified (field 14)
•last_enriched (field 15)
•current_stage (field 16)
•assigned_to (field 17) - usually empty
•notes (field 18) - may be empty

Post-Write Validation: After writing each row, mentally count: should have exactly 17 commas

Update candidates.csv:

code

Change status from 'pending' to 'verified'

Progress update:

code

"✓ [Company Name] verified → moved to companies.csv"

3g) If Dropped (Failed Any Check)

Update candidates.csv:

code

Change status from 'pending' to 'dropped'
Add drop_reason = [reason from checks above]

Progress update:

code

"✗ [Company Name] dropped: [reason]"

Step 4: Summary Report

After processing all candidates:

Count results:

code

- Total processed: [N]
- Verified: [X] (moved to companies.csv)
- Dropped: [Y]

Breakdown of drop reasons:

code

Dropped candidates ([Y] total):
- [A] No website/LinkedIn found
- [B] MNC subsidiaries
- [C] Holdco revenue >S$100M
- [D] Out of target sectors

Report format:

code

Verification complete: [N] candidates processed

✓ [X] companies verified → moved to companies.csv
  - [List company names]

✗ [Y] companies dropped:
  - [A] No website/LinkedIn: [names]
  - [B] MNC subsidiaries: [names]
  - [C] Holdco >S$100M: [names]
  - [D] Out of target sectors: [names]

[X] verified companies ready for enrichment or outreach.

Next step: Generate prospect packs? OR Commit these changes?

Step 5: Next Step Suggestions

code

"Next steps:
- Generate prospect packs: 'Generate prospect pack for [company]'
- Enrich with more triggers: (future feature)
- Commit changes: 'Commit the verified companies'
- Review holdco flags: (if any flagged for review)"

Error Handling

IF candidates.csv is empty (no rows):

code

"candidates.csv is empty. No candidates to verify.

Add candidates first:
- Specific company: 'Scan [company name]'
- Broad search: 'Find [N] [sector] companies with [trigger]'"

IF all candidates already verified or dropped:

code

"All candidates already processed:
- [X] verified (in companies.csv)
- [Y] dropped

Current status: All candidates have been reviewed.

Add new candidates to verify more companies."

IF candidates.csv is malformed:

code

"Error reading candidates.csv. Please check file format.

Expected headers:
candidate_id,company_name,source,initial_trigger,date_added,status,drop_reason

If file is corrupted, restore from git history or create new file."

Important Notes

DO NOT auto-commit:

•Update candidates.csv (status changed)
•Add rows to companies.csv
•User will commit separately with "commit this" command

Holdco flag handling:

•If uncertain about size, set holdco_flag='yes' but don't drop
•User can review these later
•Document the signals found in notes column

Confidence ratings:

code

HIGH: Award winners (E50, SME 500), press coverage, verified data
MEDIUM: Basic verification, some public info
LOW: Minimal public information, sparse data

Priority defaults:

•Set to 'medium' during verification
•User can adjust later based on triggers/needs
•Don't auto-assign 'high' or 'low' during verification

Progress updates:

•For large batches (>10 candidates), show progress every 5 companies
•Example: "Verified 5/20 candidates..."
•Keeps user informed for long operations

Edge cases:

•Company name variations (e.g., "YCH" vs "YCH Group") → ask user if same company
•Similar names to existing companies → check for duplicates before adding
•No clear sector → ask user to classify

Sector classification:

•Use best judgment from website/LinkedIn business description
•If unclear, set to closest match from target sectors list
•User can correct later if needed

Validation Checklist

Before considering a candidate "verified":

• Website OR LinkedIn exists
• NOT MNC subsidiary
• Current banker identified (if mentioned in news)
• Holdco plausibility pass (or flagged for review)
• Sector in target list (or user-approved)
• company_id generated (unique, kebab-case)
• Row added to companies.csv
• candidates.csv status updated to 'verified'

Before marking a candidate "dropped":

• Failed at least one exclusion check
• Drop reason clearly documented
• candidates.csv status updated to 'dropped'
• drop_reason column populated

You are now ready to help the user verify and qualify candidate companies for their SME banking pipeline.