Crm Import

CRM Import

Import new data from staging to master CRM database with validation and approval.

When to use

• "import CRM data from X"
• "add new people/companies/leads from file Y"
• "validate staging data"
• A file appeared in $CRM_PATH/_staging/incoming/

Input

• CSV file with new records
• Table type (auto-detect or explicitly specified)

How to execute

Step 1: Run the agent

cd $AGENTS_PATH/crm-import

# Auto-detect table type
python3 crm-import_agent.py sales/crm/_staging/incoming/new_people.csv

# Or specify explicitly
python3 crm-import_agent.py data.csv --table companies

Modes:

• --dry-run -- validation without git commit
• --validate-only -- schema validation only
• --show-duplicates -- show potential duplicates

Step 2: The agent will perform

1. Schema validation (required fields, ID format, enums)
2. Deduplication check (exact + fuzzy matching)
3. Foreign key validation
4. Business rules check
5. ID generation (if missing)
6. Git branch creation + commit
7. Telegram notification

Step 3: You will receive a Telegram message

CRM Import Ready for Review

Table: people
New records: 5
Branch: crm-import-people-new_people-2026-02-12
Commit: a1b2c3d4

Next steps:
1. Review changes: git diff main..crm-import-people-new_people-2026-02-12
2. Approve: gh pr create (then merge)
3. Reject: git branch -D crm-import-people-new_people-2026-02-12

Step 4: Review changes

cd $PROJECT_ROOT
git diff main..crm-import-people-new_people-2026-02-12

Step 5: Approve or Reject

Approve:

gh pr create --base main --head crm-import-people-new_people-2026-02-12 \
  --title "Import new people records" \
  --body "Importing 5 new people from staging"

# Merge PR
gh pr merge --merge

Reject:

git branch -D crm-import-people-new_people-2026-02-12

Output

• Git branch with new records
• Validation report (if there are issues)
• Telegram notification with summary
• Console log with details

Validation Rules

Critical (stops the import)

• Missing required fields
• Invalid ID format
• Invalid enum values
• Foreign key violations

High (flagged for review)

• Potential duplicates (fuzzy match)
• Business rule violations (e.g., person without email AND phone)

Info (log only)

• Auto-generated IDs
• Auto-set timestamps
• Skipped exact duplicates

ID Generation

If ID is missing, the agent generates it automatically:

Deduplication

Exact match (skip):

• Primary key match
• Email match (people)
• Website match (companies)
• LinkedIn URL match

Fuzzy match (review):

• People: similar name + same company
• Companies: similar name (ignoring Inc/Ltd/etc)

Examples

Example 1: Import new people

# Create staging file
cat > sales/crm/_staging/incoming/new_people.csv << EOF
person_id,first_name,last_name,email,company_id,role,created_date,last_updated
,John,Smith,john@example.com,comp-acme,Engineer,,
,Jane,Doe,jane@example.com,comp-acme,Manager,,
EOF

# Import
python3 crm-import_agent.py sales/crm/_staging/incoming/new_people.csv

# Result:
# - ID: p-acme-001, p-acme-002
# - Branch: crm-import-people-new_people-2026-02-12
# - Telegram notification

Example 2: Dry-run before import

# Verify without commit
python3 crm-import_agent.py data.csv --dry-run

# If OK -> run
python3 crm-import_agent.py data.csv

Example 3: Show duplicates

python3 crm-import_agent.py data.csv --show-duplicates

Rollback

If you need to revert:

# Find merge commit
git log --oneline --grep="Import.*records"

# Revert
git revert <commit-sha>

Staging Directory

sales/crm/_staging/
├── incoming/     # New files for import
├── review/       # Duplicate review reports
└── archive/      # Processed files

Errors

"Cannot auto-detect table type"

-> Add --table people

"Foreign key violation"

-> Import the parent record first (e.g., company before people)

"Critical schema issues"

-> Fix required fields, ID format, enum values in the CSV

"Git commit failed"

-> Check git status, the working directory may be dirty

Related skills

• change-review -- called automatically by the agent
• git-workflow -- for PR creation and merge

Dependencies

• Python 3.8+
• pandas, pyyaml
• git, gh CLI
• tg-tools (Telegram)

Limitations

• CSV only (no Google Sheets support yet)
• Insert only (no updating existing records)
• One table at a time
• Synchronous processing (for <1000 records)

Future improvements

• Google Sheets integration
• Bulk update mode
• Multi-table import (denormalized -> normalized)
• Auto-scheduling (watch directory)
• ML fuzzy matching