Include applicable ownership, access, quality, compatibility and recovery checks for persistent data, schemas, ETL or migrations in the existing TODO; preserve real consent and privacy decisions without a separate governance contract.
Installs into .claude/skills of the current project.
Are you the author of Data Governance?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/papemati98-data-governance)
---
name: data-governance
description: Include applicable ownership, access, quality, compatibility and recovery checks for persistent data, schemas, ETL or migrations in the existing TODO; preserve real consent and privacy decisions without a separate governance contract.
---
# Data governance
Use when a change materially affects durable data, schemas, ETL, published metrics,
access, migrations, retention or recovery. Synthetic transient local examples do
not require a governance workflow. Do not assign new D0-D3 gates to ordinary work.
1. In the existing TODO identify system of record, owner, purpose, classification,
sources/consumers and compatibility. Reuse established policy; an unknown fact
stays unknown. No separate sealed contract or schema registry is required by
this skill for a local fix.
2. Choose only applicable checks: actual role allow/deny, formula/units/time range,
quality and source-to-target reconciliation, schema compatibility, replay
idempotency, or backup/restore where recovery is affected. A hash or row count
alone does not prove semantic correctness.
3. For a migration define the checkpoint, representative validation, backup/recovery
and narrow rollback in this same TODO. Destructive or production cutover retains
its exact explicit owner-consent gate.
4. Reuse existing lawful basis, retention, DPIA and transfer decisions. Do not invent
them or request all of them for every task. Ask for a real unresolved business
or legal decision only when the proposed operation depends on it. Independently
verify variable high-stakes facts from primary sources before using them.
5. Minimize data sent to models: schema, aggregate metrics and redacted evidence.
Existing privacy and no-training policies remain binding. This skill does not
authorize client-data access, contact, secrets, IAM or public integrations.
6. Independent product review checks current results and relevant data regressions.
Report evidence and missing decisions proportionately. Artifact/transport success
is separate from product acceptance and canonical Hub/session closeout.
Do not add an independent plan, contract, approval or review graph solely for
these checks. Shared changes still require private backup PUSHED/CURRENT.
Historical DRAFT/seal/compile-graph-fragment commands are preserved only in
references/legacy-recovery.md. Do not execute them for a new TODO.