| name | crm-hygiene-check |
| description | Audit CRM hygiene; never make unreviewed bulk changes. |
| metadata | {"short-description":"Audit CRM quality and propose safe fixes"} |
CRM Hygiene Check
Outcome
A prioritized, explainable data-quality report with exact proposed corrections and confidence levels.
CRM-neutral operating contract
- Discover available CRM, communication, calendar, knowledge, and file capabilities before choosing a retrieval plan. If no live CRM is available, use user-supplied exports when that still satisfies the request; never imply a live write is possible.
- Map records by meaning, not vendor labels: organization, person, lead, opportunity, activity, product, service record, user or team, and knowledge item. Inspect schemas and custom fields, preserve native IDs, stages, currencies, timezones, and fiscal periods.
- Keep observed facts, derived conclusions, and missing information distinct. Anchor consequential claims to record IDs, fields, timestamps, or named sources, and state access or freshness limits.
- Start read-only. Before creating, updating, assigning, sending, or scheduling, show the exact proposed action and obtain confirmation unless the current request already authorized that exact mutation.
- Obey connected-system permissions, resolve ambiguous records or people instead of guessing, check for duplicates before creation, and report partial failures without repeated retries.
Workflow
- Resolve the user, target records, requested scope, reporting window, timezone, and business objective. Ask only for missing details that materially change the result.
- Retrieve the minimum authorized evidence needed from the following inputs: Requested object and record scope, schema, validation and uniqueness rules, ownership, required fields, stage rules, associations, timestamps, and relevant source-of-truth data.
- Normalize relevant entities and fields while retaining native record identifiers and source timestamps.
- Apply the analysis rules below. When evidence conflicts, show the conflict rather than averaging it away.
- Deliver the required output in a compact form suited to the scope. Put urgent or decision-relevant items first.
Analysis rules
- Confirm scope and rules before scanning; profile fields and missingness without retrieving unnecessary sensitive data.
- Test for required-field gaps, invalid values, stale dates or next steps, contradictory stage data, duplicates, orphan associations, ownership conflicts, and suspicious outliers.
- Distinguish definite errors, policy violations, likely issues, and review candidates; explain the rule and evidence for every finding.
- For duplicates, preserve record history and relationships and never select a merge survivor from completeness alone.
Required output
- Scope and rule set
- Issue counts by severity and type
- Record-level findings
- Proposed field-level fixes
- Merge or ownership review queue
- Coverage caveats
Actions and boundaries
Default to report-only. Require explicit approval for a bounded change set; never delete or bulk merge through this skill.