LineageGuard

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for LineageGuard

Inspiration

Data teams often learn about breaking schema changes only after dashboards fail or pipelines alert. DataHub already knows the assets, owners, schemas, tags, and lineage. LineageGuard turns that context into an agentic release gate that can explain downstream risk before a migration ships.

What it does

LineageGuard lets a data engineer select a DataHub asset, describe a proposed schema change, and generate a downstream impact report. It identifies affected dashboards, dbt models, and operational jobs; routes the issue to owner groups; and drafts a remediation pack with release gates, dbt contract checks, SQL probes, Airflow guardrails, and owner messages.

How we built it

The MVP is a small web app, JSON API, and MCP-compatible endpoint. The demo graph uses DataHub URN conventions and includes schema fields, ownership, tags, upstream lineage, and downstream lineage. POST /mcp exposes get_asset_context, analyze_schema_change, and draft_remediation_pack. The repo also includes an Agent Skill in SKILL.md so an agent can use LineageGuard during release review.

DataHub usage

LineageGuard is built around DataHub metadata concepts: dataset URNs, schema fields, ownership, tags, and lineage. The hosted demo uses labeled sample metadata, and the interface is designed so the sample graph can be replaced by a live DataHub MCP server or DataHub GMS/GraphQL endpoint.

Challenges

The main challenge was keeping the demo truthful without access to a private production DataHub instance. The MVP labels its included metadata as demo data and focuses on the integration contract: metadata in, release-risk report and remediation pack out.

Accomplishments

• Working web dashboard
• MCP-compatible tool endpoint
• Agent Skill workflow
• DataHub-style lineage impact analysis
• Concrete remediation artifacts for dbt, SQL, Airflow, and owner notification

What's next

Next steps are to connect directly to a live DataHub MCP server, support pull-request comments for dbt repos, and add Slack/Jira handoff actions after owner approval.

Built With

• agent-skills
• cloudflare-workers
• datahub
• devops
• javascript
• lineage
• mcp-server
• metadata