Change Review Skill | Agent Skills

Change Review

Review of CRM and PM data changes before PR -- a data equivalent of code-review

Public repository: https://github.com/your-org/claude-change-review-skill

When to use

Before a PR with CRM changes
Before a PR with PM changes (tasks, projects)
"check my CRM changes"
"change review"
"data review"

Paths

| What | Path | |------|------| | CRM | $CRM_PATH/ | | CRM Schema | $CRM_PATH/schema.yaml | | PM | $PM_PATH/ |

How to execute

Step 1: Read schema.yaml

# ALWAYS read the schema first -- it contains all the rules
cat $CRM_PATH/schema.yaml

Schema contains:

primary_key -- unique identifier
required -- required fields
unique -- unique fields
foreign_keys -- relationships between tables
composite_unique -- composite unique keys
enums -- allowed values
id_format -- regex for ID format
rules -- business rules

Step 2: Get the diff

git diff HEAD -- sales/crm/

Step 3: Load data

import pandas as pd
import yaml

# Read schema
with open('$CRM_PATH/schema.yaml') as f:
    schema = yaml.safe_load(f)

# Load all tables
base_path = '$CRM_PATH/'
tables = {}
for table_name, table_def in schema['tables'].items():
    tables[table_name] = pd.read_csv(base_path + table_def['file'])

Step 4: Validate against schema

import re

def validate_table(name, df, table_schema, all_tables):
    issues = []

    # 1. Primary key uniqueness
    pk = table_schema['primary_key']
    if df[pk].duplicated().any():
        dups = df[df[pk].duplicated()][pk].tolist()
        issues.append(('critical', f'Duplicate {pk}: {dups}'))

    # 2. Required fields
    for field in table_schema.get('required', []):
        missing = df[df[field].isna()]
        if len(missing) > 0:
            issues.append(('critical', f'Missing required {field}: {len(missing)} rows'))

    # 3. Unique fields
    for field in table_schema.get('unique', []):
        if field == pk:
            continue
        dups = df[df[field].notna() & df[field].duplicated()]
        if len(dups) > 0:
            issues.append(('high', f'Duplicate {field}: {dups[field].tolist()}'))

    # 4. Foreign keys
    for fk_field, ref in table_schema.get('foreign_keys', {}).items():
        ref_table, ref_field = ref.split('.')
        valid_values = all_tables[ref_table][ref_field]
        invalid = df[df[fk_field].notna() & ~df[fk_field].isin(valid_values)]
        if len(invalid) > 0:
            issues.append(('critical', f'Invalid {fk_field} references: {invalid[fk_field].tolist()}'))

    # 5. Composite unique
    for fields in table_schema.get('composite_unique', []):
        dups = df[df.duplicated(subset=fields, keep=False)]
        if len(dups) > 0:
            issues.append(('critical', f'Duplicate {"+".join(fields)} combination'))

    # 6. Enum values
    for field, valid_values in table_schema.get('enums', {}).items():
        invalid = df[df[field].notna() & ~df[field].isin(valid_values)]
        if len(invalid) > 0:
            issues.append(('high', f'Invalid {field} values: {invalid[field].unique().tolist()}'))

    # 7. ID format
    if 'id_format' in table_schema:
        pattern = table_schema['id_format']
        invalid = df[~df[pk].str.match(pattern, na=False)]
        if len(invalid) > 0:
            issues.append(('low', f'Invalid ID format: {invalid[pk].tolist()}'))

    return issues

# Validate all tables
for name, table_schema in schema['tables'].items():
    if name in tables:
        issues = validate_table(name, tables[name], table_schema, tables)
        for severity, msg in issues:
            print(f'[{severity.upper()}] {name}: {msg}')

Step 5: Business rules (from schema.rules)

# Example: won_lead_has_client
if 'leads' in tables and 'clients' in tables:
    won_leads = tables['leads'][tables['leads']['stage'] == 'won']
    for _, lead in won_leads.iterrows():
        client_exists = (
            (tables['clients']['company_id'] == lead['company_id']) &
            (tables['clients']['product_id'] == lead['product_id'])
        ).any()
        if not client_exists:
            print(f"[HIGH] Won lead without client: {lead['lead_id']}")

Step 6: Output

## Change Review Summary

**Schema version:** 1.0
**Files reviewed:** [list]
**Risk Level:** Critical / High / Medium / Low

### Critical Issues (must fix)
- [table:field] Description

### High Priority
- [table:field] Description

### Validation by Schema
- [x] Primary keys unique
- [x] Required fields present
- [x] Foreign keys valid
- [x] Enum values valid
- [x] Business rules passed

Severity Levels

| Level | Schema Rule | Examples | |-------|-------------|----------| | Critical | primary_key, required, foreign_keys, composite_unique | Duplicate ID, broken FK | | High | unique, enums, rules | Duplicate email, invalid status | | Medium | (manual check) | Missing optional fields | | Low | id_format | Wrong ID pattern |

Troubleshooting

| Problem | Solution | |---------|----------| | Broken FK | Add parent record or fix ID | | Duplicate PK | Change ID or remove duplicate | | Invalid enum | Use values from schema.yaml | | Wrong ID format | See id_format in schema.yaml |

Related skills

add-lead -- adding records (also reads schema.yaml)
update-lead -- updating records
code-review -- for code (not data)

Agent Skills: Change Review

Install this agent skill to your local

Skill Files